Pogbagram

With memory costs soaring by 300%, the Chinese mobile phone market has entered the "Year of Major Price Hikes," and budget phones are rapidly disappearing.

This year, China's smartphone market is experiencing an unprecedented wave of price increases across the board. The core reason is the sharp rise in the cost of upstream memory and storage chips, coupled with the capacity squeeze caused by the surge in demand for AI servers. The industry generally believes that 2026 will become the "first year of major price increases" for the mobile phone industry.

Multiple supply chain and channel sources have confirmed that the procurement cost of smartphone memory chips has increased by more than 80% compared to the same period last year, and the upward trend shows no signs of slowing down.

As a result, major brands such as OPPO, OnePlus, vivo, Xiaomi, iQOO, and Honor have planned to launch a new round of price adjustments in early March, which will be the largest and most significant collective price adjustment in the past five years.

Sina Tech, citing multiple sources, reports that starting in March, mobile phone price increases will accelerate, with new models seeing price hikes of at least 1,000 yuan, and mid-to-high-end flagship models potentially seeing increases of 2,000-3,000 yuan.

Meanwhile, some brands will also raise the prices of older models. In 2026, the Chinese mobile phone market may see multiple price adjustments within a year for the first time. The market for phones priced around 1,000 yuan may also face the dilemma of "selling at a loss and losing money."

It is worth noting that, according to South Korean media DealSite, Samsung Semiconductor originally planned to raise prices by only about 60% in negotiations to supply Apple with LPDDR5X memory needed for the iPhone 17 series, but Apple quickly agreed to accept the initial price of 100% .

This result not only established Samsung's absolute pricing power in the memory market, but also triggered a chain reaction, forcing Samsung's own mobile business unit to bear the same price increase pressure. Ultimately, it introduced chips from its competitor Micron to ensure supply, and adopted a "50/50" strategy in the initial batches of the new S26 series.

Yesterday, Xiaomi REDMI product manager Hu Xinxin (@馨心_Mia) also shared her views on this matter, saying that the current storage supercycle is truly a "ghost story" for mobile phone manufacturers .

Multiple institutions predict that the memory shortage will continue until 2027, and with TSMC's more advanced processes entering mass production in 2028, costs are unlikely to fall back to previous levels. This means that mobile phone prices will remain high for the next three years, and the industry will enter a long-term structural price increase cycle.

Related reading: "Losing money on releases, losing money on sales": The era of reduced specifications and increased prices for mobile phones has arrived.

Nano Banana 2 Launched: High Image Quality and High-Speed ​​Generation Achieved for the First Time

Today, Google officially launched its next-generation image generation model, Nano Banana 2 (Gemini 3.1 Flash Image), which focuses on further improving image quality, understanding, and subject consistency on the basis of high-speed generation. It is positioned as a lightweight alternative to Nano Banana Pro and is open to a wider range of users.

Nano Banana 2 continues the Gemini series' real-world knowledge base and combines it with real-time information from web searches, making the model perform more naturally in terms of spatial understanding, scale relationships, lighting and shadow processing, and Chinese text rendering.

• It can keep the faces of up to 5 characters unchanged and the appearance of 14 objects consistent in a single generation, making it suitable for complex scenarios such as comic serialization and storyboard production.
• It covers output from 512px to 4K, with sharper image quality and more natural lighting compared to its predecessor, and can be used directly for marketing materials or visual design;
• It adopts the same high-efficiency architecture as Gemini Flash, emphasizing a "near real-time" high-speed generation experience, suitable for rapid iteration and multiple rounds of modification;
• In terms of cost, regular users can generate 100 images per day for free in the Gemini app and Google search, while Pro users can generate up to 1,000 images per day; the API price is halved compared to the previous generation Pro model, with the cost of generating a 4K image being approximately $0.15.
• In terms of content credibility, Google has upgraded SynthID digital watermarking and C2PA content credentials to enhance AI-powered content traceability capabilities.

Related reading: Nano Banana 2 just released! Affordable, generous portions, and even understands Chinese better – I discovered these details after trying it out.

Apple's first new product of 2026 was unveiled on Monday.

Last night, Apple CEO Tim Cook officially announced on social media:

An exciting week is about to begin. It all kicks off next Monday evening Beijing time!

Apple had previously announced that it would release new products at 10 PM on March 4th. However, according to the latest news, Apple is likely to adopt a release schedule of one new product per day to maintain buzz throughout the week.

According to multiple sources, this release is expected to bring updates to several entry-level products.

• iPhone 17e: It will align with the standard iPhone 17, maintaining the same core specifications such as processor and MagSafe, but will retain the single-camera design of its predecessor. This new iPhone is expected to target emerging markets and enterprise procurement channels with a highly competitive price.
• Entry-level MacBook: Multiple sources indicate that Apple is testing a low-cost MacBook solution equipped with A-series chips, with a target price potentially dropping to the $600 range;

In addition, there is a possibility of updates to the MacBook Air and MacBook Pro product lines equipped with M5 series chips, and Apple's own Studio Display is also expected to be updated.

Google's official "Gemini Phone" revealed: Allows Gemini to directly control apps.

According to 9to5Google, Google will launch a new API in the Android system to enable AI agents to control apps, similar to the "Bean Bun Phone".

According to the official Android documentation, AppFunctions is a platform capability of Android 16, along with the Jetpack library, which allows developers to expose specific functions within their applications to AI Agents in a structured manner.

Developers can generate the necessary code using annotations, metadata, and the KSP compiler, enabling AI Agent applications such as Gemini to execute tasks directly in the background of the device without requiring users to navigate to the application interface.

At yesterday's Samsung Galaxy S26 series launch event, Google demonstrated the practical application of AppFunctions: users only need to tell Gemini, "Show me photos of cats in Samsung Gallery," and the system can automatically call the gallery's functions and return the results directly, without having to manually open the app to search.

This feature will first be available on the Galaxy S26 and Pixel 10 series in the United States and South Korea, initially supporting only some food delivery, fresh food, and ride-hailing apps. It will also be rolled out to more Samsung devices with One UI 8.5.

Nubia President Ni Fei stated on Weibo yesterday that while the combination of Samsung and Google demonstrates the automation capabilities of mobile phones, it still falls under the category of "partial capabilities" and has not achieved the "full-scenario system-level" autonomous driving AI experience of the Nubia M153 "Doubao Phone".

He emphasized that Nubia had already released a preview version of the Doubao mobile phone technology last year, promoting the deep application of AI intelligent agents on mobile phones at the system level, and looked forward to more manufacturers joining in to improve scenario coverage and implementation depth.

Domestically developed large-scale models topped the OpenRouter charts in February, with MiniMax and Kimi leading the global token usage volume.

According to the Science and Technology Innovation Board Daily, in February this year, Chinese large-scale models dominated the global token call volume rankings on the OpenRouter platform, with domestic models occupying four of the top five spots, showing a trend of simultaneous leaps in application demand and technical capabilities.

The latest data from February shows that Chinese models such as MiniMax, Kimi, and DeepSeek have seen significant growth globally.

MiniMax M2.5, released just two weeks ago, ranked first in monthly usage with 4.55 trillion tokens; Kimi K2.5, from the darker side of the moon, ranked second with 4.02 trillion tokens. Google Gemini 3 Flash Preview, DeepSeek V3.2, and Anthropic Claude Sonnet 4.5 followed.

Weekly data also shows the strong performance of domestic models. From February 9th to 15th, the number of calls to Chinese models exceeded that of American models for the first time; from February 16th to 22nd, the weekly call volume of Chinese models further increased to 5.16 trillion tokens, with a cumulative increase of 127% over three weeks.

According to the Dark Side of the Moon, Kimi K2.5 has already surpassed its total revenue for the entire previous year in less than a month since its release, with particularly significant growth in overseas paying users.

Apple releases official speed test tool

Recently, Apple officially launched a network speed test tool in the "Apple Support" app, which supports iOS 18 and later versions.

After updating to version 5.12, the new speed test tool in the application allows users to test the download, upload speed and latency of the current network environment with one click.

This speed test tool maintains the minimalist style of system-level applications, with smooth interface animations and no ad interference. Compared to third-party speed test apps on the market, it is more lightweight and direct.

However, the entry point for this function is hidden quite deep, located under the "Internet Speed ​​Test" submenu of "Device Performance".

Former Apple and Meta executive Pang Ruoming joins OpenAI

According to The Information, Ruoming Pang, former head of Apple's foundational modeling team and former head of AI infrastructure at Meta Super Intelligence Lab, has officially joined OpenAI.

Related reading: Just now, the most expensive Chinese in Silicon Valley turned down a record-breaking $1.4 billion offer and joined OpenAI with a bucket.

Last year, Pang Ruoming left Apple to join Meta, and less than a year later he jumped ship again. According to multiple reports, his compensation package at Meta was worth more than $200 million, including multi-year incentives.

Before joining Meta, Pang Ruoming led Apple’s Basic Model Team (AFM), leading about 100 people to develop large-scale language models, providing underlying capabilities for features such as Apple Intelligence.

Companies like Meta, OpenAI, and Apple are attracting top researchers with high salaries. In the past year, several core members of the Meta AI team have left, including executives such as Yann LeCun.

IBM responds to the AI ​​revolution: Modernization is far more than just changing code.

Recently, IBM Senior Vice President Rob Thomas published an article responding to discussions surrounding whether AI can quickly replace COBOL and mainframes.

Previously, Anthropic announced that Claude Code could automatically sort out COBOL dependencies, generate documentation, and identify risks, raising concerns in the market about the impact on IBM's mainframe business. IBM's stock price recorded its biggest single-day drop in nearly 26 years on Monday, with a market value loss of approximately $31 billion.

The article points out that the core value of the IBM Z mainframe platform lies not in COBOL, but in its vertically integrated architecture from chip to operating system, which can provide extremely high transaction processing capabilities, security, and availability.

IBM stated that simply converting COBOL to other languages ​​cannot replicate these capabilities, nor can it solve real modernization challenges, including data architecture restructuring, runtime replacement, transaction consistency, disaster recovery capabilities, encryption systems, and integration with peripheral systems.

The article also points out that IBM Z's advantages are even more pronounced in the AI ​​era. AI tools are helping companies automatically analyze legacy system architectures, accelerate DevOps processes, reduce manual operations, and alleviate the skills gap caused by COBOL talent retirement.

IBM believes these trends mean that businesses should continue to increase their investment in IBM Z.

IBM emphasizes that while AI can indeed bring value to the code space, it should not be misunderstood as a "one-click replacement" of mainframe platforms. True modernization is far more complex than language translation.

Genshin Impact quietly removed the clause allowing voice data to be used for AI training.

According to Eurogamer, Genshin Impact's privacy policy previously included a clause allowing the processing of voice communication data when players use the voice chat feature. This clause has recently been quietly removed.

Like many AI-driven experiences, we might use user input you generate (such as chat data) to train and optimize the models used to provide services.

User "Cevian" first noticed this change on the social platform Bluesky. However, Chapter 2 of the new policy still retains the statement that "users can exit model training at any time in the service-related settings," and emphasizes that exiting will not affect the existing game experience.

The report further points out that this deletion occurred after the privacy policy was updated on January 14th, and the official reason for the deletion has not yet been publicly explained. However, it has been verified that the Chinese version of the Genshin Impact privacy policy does not contain such a clause.

The report also mentioned that miHoYo has publicly expressed its investment in generative AI on multiple occasions in recent years, including features such as the "artificial intelligence ecosystem simulation system," "all-terrain combat AI," and "highly intelligent NPCs" shown in the trailer for the Unreal Engine 5 large-scale MMO project released last year.

XPeng's second-generation VLA has entered routine L4 testing, and mass production is "just around the corner."

Yesterday, XP-Candice Tingting, Senior Director of Autonomous Driving Products at XPeng Motors, shared a video on Weibo of the Robotaxi test team filmed at the factory.

She stated that the model equipped with the second-generation VLA has passed third-party track testing and obtained the Guangzhou Intelligent Connected Vehicle Road Test Permit. It is currently undergoing routine L4 public road testing, and mass production is "just around the corner."

According to Zheng Rongqing, Product Marketing Director of XPeng Motors, the XPeng GX has been able to complete a full range of operations, including starting from a standstill, autonomous driving, temporary parking, picking up passengers, and starting again, under conditions where the driver is unmanned and there is no map in the park, demonstrating a high degree of stability in autonomous driving.

The second-generation VLA was released last November, highlighting its "brain-based reasoning" model capabilities, which can generate response strategies for different scenarios. Previous reports indicate that the system has brought several capability upgrades, including small-path NGP, with an average takeover mileage of approximately 13 times higher for complex small paths.

XPeng Chairman He Xiaopeng previously revealed that Volkswagen will be the first customer for the second-generation VLA model.

Lei Jun's first live stream of the Year of the Horse is scheduled for tonight, where he will explain Xiaomi's automotive safety system in detail.

Xiaomi Auto announced that Lei Jun will host his first live stream of the Year of the Horse at 7:30 PM tonight, focusing on the theme of "safety," at the Xiaomi Auto factory.

Lei Jun stated that this live stream will systematically introduce Xiaomi Auto's entire safety system, and several experts will participate. He emphasized that the content is "very professional and may be a bit dry."

Yesterday, Caixin revealed key details of the drunk driving accident involving a Xiaomi SU7 Ultra in Chengdu on October 13th last year:

The report cited an opinion from the Sichuan Xihua Traffic Judicial Appraisal Center, which showed that the vehicle was traveling at a speed of 167 km/h at the moment of the first collision. The power battery was squeezed, causing a short circuit in the cells. High-voltage current rushed into the low-voltage circuit, causing the low-voltage system to lose power and rendering the electronic release function of the door handles ineffective.

Because the vehicle was not equipped with an external mechanical emergency handle, the doors could not be opened from the outside, and the driver ultimately died from burns caused by the vehicle catching fire. Chengdu police had previously determined that the driver bore full responsibility for the accident due to drunk driving and excessive speeding.

This incident has also raised public awareness of electronic door handles, battery safety, and emergency escape mechanisms.

Paramount raises offer to $111 billion; Netflix abandons acquisition of Warner.

According to Variety, Netflix has officially abandoned its bid to acquire Warner Bros., marking a key turning point in the months-long Hollywood merger battle.

Netflix stated that continuing the bidding was "no longer financially attractive" after Paramount made a higher offer. Warner's board of directors determined in late February that Paramount's latest proposal was a "superior offer."

Ultimately, Netflix withdrew, making Paramount the most likely buyer to acquire Warner.

Previously, Warner Bros. and Netflix had reached a preliminary merger agreement, but Paramount submitted a company-wide acquisition proposal of approximately $111 billion on February 24, far exceeding the nearly $83 billion deal in which Netflix acquired Warner Bros. Pictures and its streaming business alone.

Paramount's offer included $31 per share in cash, a $7 billion regulatory breakup fee, and the assumption of the $2.8 billion termination fee that Warner was required to pay to Netflix.

The acquisition covers all of Warner's businesses, including CNN, Discovery, HBO Max, DC Studios, and content assets such as Harry Potter, White Lotus Resort, and Succession. If the deal goes through, it will further reduce the number of Hollywood's five remaining traditional film studios to four.

In its statement, Netflix emphasized that its business is "healthy and continues to grow," and that it will invest approximately $20 billion in content production this year and resume its stock buyback program. Co-CEOs Ted Sarandos and Greg Peters stated that the acquisition of Warner was "icing on the cake, not a necessity."

With Netflix's exit, Warner's ultimate fate will depend on regulatory scrutiny and shareholder votes. If Paramount succeeds, it will be one of the largest and most far-reaching media mergers in recent years.

iQiyi's revenue reached 27.29 billion yuan last year, marking its fourth consecutive year of profitability.

According to Jiemian News, iQiyi released its unaudited financial results for the fourth quarter and full year ending December 31, 2025 yesterday, showing that the platform continues to maintain profitability under the adjustment of its content strategy and business structure, and has achieved operating profitability for four consecutive years.

The financial report shows that membership business, advertising business and overseas business all maintained growth momentum, becoming key factors driving the overall revenue improvement.

The company's total revenue for the year was RMB 27.29 billion, with non-GAAP operating profit reaching RMB 640 million; of which, total revenue for the fourth quarter of 2025 was RMB 6.79 billion, achieving both year-on-year and quarter-on-quarter growth, with quarterly non-GAAP operating profit of RMB 140 million.

Baidu's AI business revenue reached 40 billion yuan last year, and Luobo Express has accumulated over 20 million trips.

Yesterday, Baidu released its financial report for the fourth quarter and full year of 2025, with AI cloud, AI applications, and autonomous driving constituting the three core growth drivers.

• Total revenue in 2025 is 129.1 billion yuan, a year-on-year decrease of 3%; AI business revenue is 40 billion yuan.
• Fourth-quarter revenue reached 32.7 billion yuan, a year-on-year increase of 5%, with AI business revenue accounting for 43% of general business revenue, exceeding market expectations;
• In 2025, Baidu's net profit attributable to shareholders was 5.6 billion yuan; excluding the impact of long-term asset impairment, the net profit was 19.4 billion yuan; non-GAAP net profit was 18.9 billion yuan, with a net profit margin of 15%.
• AI cloud revenue for the year was approximately RMB 20 billion, representing a year-on-year increase of 34%; Q4 AI high-performance computing facility subscription revenue increased by 143% year-on-year.
• AI-native marketing services revenue increased by 301% year-on-year.
• AI application revenue exceeded 10 billion yuan for the year; Baidu's intelligent agent "Famou" has attracted more than 2,000 companies to apply for trial use;
• Baidu App's monthly active users (MAU) reached 679 million; Wenxin Assistant's MAU reached 202 million, with the Spring Festival red envelope campaign driving a fourfold year-on-year increase in MAU.
• In the fourth quarter, Carrot Express provided 3.4 million autonomous driving rides globally, a year-on-year increase of over 200%; the cumulative number of rides exceeded 20 million, covering 26 cities worldwide, with a total mileage of over 190 million kilometers of fully autonomous driving.

Jensen Huang: AI assistants will not replace the software industry

In a recent interview with CNBC, Nvidia CEO Jensen Huang reiterated concerns about the impact of AI on the software industry.

He emphasized that the market has "misjudged" this issue, and that AI assistants will not only not replace software tools, but will further improve the efficiency of the software industry.

He pointed out that AI assistants will become "users of tools," not replacements. Whether it's Cadence, Synopsys, ServiceNow, or SAP, the fundamental reason for the existence of these tools will not change; AI will further enhance productivity on top of these tools.

We need tools to accomplish specific tasks and to provide feedback to us in a way that we can understand.

This statement came after Nvidia released strong financial results.

In the fourth fiscal quarter of 2026, Nvidia's revenue reached $68.127 billion, a year-over-year increase of 73%, exceeding market expectations. During the earnings call, Jensen Huang emphasized that the inflection point for intelligent agents (AI) has arrived, and global companies are increasing their investment in computing power.

In this new world of AI, computing power equals income.

OPPO Find N6 real device exposed: creases are almost invisible to the naked eye

Recently, a reporter from Guangdong News Network visited the OPPO mobile phone production line and released a real video of the upcoming foldable screen product Find N6, showing that its crease is "so shallow that it is almost invisible".

According to an official article released by OPPO, following the previous generation's entry into the 8mm thin and light era, the new phone has undergone a comprehensive upgrade in folding structure, materials, and manufacturing precision. Staff explained that to achieve seamless folding, the technical team spent three years trying dozens to hundreds of solutions, constantly overturning designs and re-verifying them.

It is worth noting that Zhou Yibao, the product manager of OPPO Find series, also revealed on Weibo yesterday that the Find N6 will be equipped with "the only Hasselblad 200MP ultra-clear quad camera on a foldable screen" and will be the first foldable screen to feature a lens for reproducing the colors of Danxia landforms.

Vivo X300 Ultra Reportedly to be the World's First to Feature Sony's 200MP LYT-901 Sensor

According to the blogger "Digital Chat Station", vivo's upcoming flagship imaging phone, the X300 Ultra, will be the world's first to feature Sony's 200-megapixel LYT-901 main camera.

Released last November, this sensor features a large 1/1.12-inch sensor and integrates AI-based image processing circuitry, supporting up to 4x lossless zoom for a single camera.

In addition to the main camera, the new phone is expected to also feature a 3.5X periscope telephoto lens with the same 200MP Samsung HPB sensor, and a 50MP ultra-wide-angle lens that industry insiders call the "strongest in the mobile phone industry".

The railway authorities responded to the case of a 1700 yuan train ticket that was successfully booked through the midnight waitlist and subsequently invalidated.

Yesterday, the railway authorities responded to the online rumor that "a 1,700 yuan train ticket that was successfully booked in the middle of the night was invalidated," stating that the relevant reports were untrue.

An article published on the official WeChat account of "China Railway" shows that the 12306 Railway Technology Center has verified the relevant orders and confirmed that the waiting list process was normal and the notification was timely. The so-called "cancellation" is a misinterpretation.

The railway department stated that Chen submitted a waitlist order for the train from Xiamen (Xiamen North) to Yixing on February 19, and purchased another ticket for train G1656 on February 22, but did not cancel the waitlist order.

After fulfilling the waitlist order in the early morning of February 23, the system sent notifications to the three passengers according to their selected method. Since the travel times of the waitlist train (G1654) and the purchased train (G1656) do not overlap, the system will not determine that there is a travel conflict, and therefore there is no "automatic cancellation".

The railway authorities also pointed out that the ticket purchase interface shown by some media outlets was not the official 12306 website, and reminded passengers to purchase tickets through official channels. If they have already purchased other tickets, they should cancel their waitlist orders in a timely manner to avoid misunderstandings.

Ctrip's Director and President Resign

According to reports from The Paper and Beijing News, Trip.com Group released its unaudited financial results for the fourth quarter and full year of 2025 yesterday, and disclosed major personnel changes in the board of directors and management. Two co-founders, Fan Min and Ji Qi, resigned from their positions as director and president and director of the company, respectively .

At the same time, the company appointed Wu Yihong and Xiao Yang as new independent directors, and added Li Jipei as a member of the board's remuneration committee.

• Trip.com's net revenue for the fourth quarter of 2025 was approximately RMB 15.398 billion, representing a year-on-year increase of approximately 20.83%.
• Net profit attributable to the parent company was approximately RMB 4.281 billion, representing a year-on-year increase of approximately 98.47%. Full-year net revenue was approximately RMB 62.409 billion, representing a year-on-year increase of approximately 17%.
• Net profit attributable to the parent company was approximately RMB 33.294 billion, representing a year-on-year increase of approximately 95%.

The significant increase in profits was mainly due to a substantial rise in "other income".

• In the fourth quarter of 2025, Trip.com recorded other revenue of RMB 2.038 billion, compared to expenses of RMB 137 million in the same period last year.
• Other revenue for the year reached RMB 21.321 billion, nearly ten times higher than the previous year, mainly due to proceeds from the sale of shares in the Indian online travel platform "MakeMyTrip". Excluding this impact, Ctrip's operating profit for the full year of 2025 would increase by approximately 11% year-on-year.

Trip.com CEO Jane Sun revealed in a conference call that the core OTA business will generate approximately RMB 1.1 trillion in transaction volume in 2025, with accommodation and air tickets contributing approximately RMB 280 billion and RMB 550 billion respectively.

It is worth noting that Ctrip mentioned in its financial report that it received an investigation notice from the State Administration for Market Regulation in January this year. The investigation is still ongoing, and the company is fully cooperating with the regulatory authorities.

Discontinued CCD cameras are making a comeback, with prices rising several times over.

According to CCTV Finance, CCD digital cameras, which had been out of production for many years, have recently become popular again, becoming a "hot-selling" product among young consumers. Several merchants in Huaqiangbei stated that models that originally sold for only a few hundred yuan are now generally priced at over 2,000 yuan, with some models even seeing price increases of several times.

Merchants in Huaqiangbei revealed that in earlier years, CCD cameras were even sold by weight, at about 100 yuan per kilogram. The current price surge is mainly due to the fact that production has been suspended for many years, the limited stock, and the influence of aesthetic trends.

However, merchants also warn that most CCD cameras currently on the market are old, second-hand devices, which generally have risks such as aging batteries, sensor wear and tear, and shutter life approaching its limit, significantly affecting the failure rate.

Avatar 3 grosses over 1.2 billion yuan in 70 days.

According to data from Maoyan Professional Edition, the cumulative box office of the movie "Avatar 3: Ashes" has exceeded 1.2 billion RMB 70 days after its release.

The film was released simultaneously in the United States and mainland China last December. Directed by James Cameron, the story continues to revolve around Jake Sully and Neytiri's family, focusing on the family's growth and choices amidst the war. It also introduces new groups such as the "Wind Merchants" and the "Ash Tribe," expanding the cultural and conflict structure of Pandora.

On Douban, Avatar 3 received a score of 7.6.

Mortal Kombat 2 releases Chinese trailer

A new Chinese trailer for the latest installment in the "Mortal Kombat" series, "Mortal Kombat 2," has been released. The film will be released in North America on May 15th, and its release in China has been confirmed.

This film continues the series' high-intensity, bloody style, with director Simon McQuaid returning to direct and screenwriter Jeremy Slater writing the script. The story is still adapted from the fighting game IP of the same name created by Ed Bourne and John Tobias.

The film will revolve around the ultimate showdown between the heroes and the villain Shao Kang, with the story scaled up further and emphasizing the themes of "cruelty and glory".

Clips from "Panda Project: Tribal Adventure" released.

The family-friendly comedy film "Panda Project: Tribal Adventure" released a clip titled "When Darkness Falls," in which Yu Yang, playing the tribal prince Turu, gathers everyone and decides who will carry out a special mission by drawing lots. The seemingly fair process conceals a hidden "scheme."

Directed by Xu Hongyu and starring Hu Hu, Jackie Chan, Ma Li, Qiao Shan, and Yu Yang, the film is currently showing nationwide and is positioned as a "family-friendly movie".

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

ifanr | Original Link · View Comments · Sina Weibo

Editor's Note:

When AI begins to search for its own shape, some of its choices are unexpected.

AI has given rise to a dedicated button on smartphones, seemingly rekindling their long-lost evolutionary drive. Glasses, with their natural access to sight and hearing, are beginning to resemble the next generation of personal terminals. Some small, focused devices seem more reliable than all-in-one devices at certain moments. Meanwhile, radical attempts to replace smartphones with disposable devices have met with a cold reception.

The implementation of technology is never just about stacking up functions; it's also about people's habits, the fit of scenarios, and the redefinition of "easy to use".

iFanr launches the "AI Gadgets Chronicle" column, aiming to explore with you how AI is changing hardware design, reshaping human-computer interaction, and, more importantly, how AI will enter our daily lives.

I originally thought the Samsung Galaxy S26 series had already been leaked, and the launch event would just be a formality. It turns out Samsung and Google had something else in store.

The two companies jointly showcased the new Gemini AI capabilities featured in the S26: with a simple verbal command, Gemini can help you hail a ride with Uber or order takeout on DoorDash.

This feature is currently in early preview and is only available in the United States and South Korea.

You can think of it as Google and Samsung teaming up to create a global version of "Doubao Phone" (more accurately, Doubao Phone Assistant). The Galaxy S26 series is just the beginning; these capabilities will be rolled out to Google Pixel 10 phones and more Android 17 devices later.

Having seen and used many mobile phone/computer system-level AI agents, and having used the "Doubao Phone" extensively, I feel that the discussion about the Gemini agent should not be limited to just a "new feature".

Admittedly, this is not the first time that the underlying framework of the Android operating system has been deeply customized to accommodate intelligent agents—many manufacturers, including OPPO, Honor, and Huawei, have already made considerable early attempts.

But this is Google, the absolute owner of the Android operating system.

If ByteDance, as an "outsider," was "disrespectful" to a national-level app, then Google's involvement in this matter is entirely different.

But hold on, let's first take a look at what exactly this "beanbag phone" from Google and Samsung is all about.

How is the Samsung "Bean Bun Phone"?

The "Gemini Auto Task" capability showcased by Samsung and Google can mimic human operation of a phone to automate tasks. The underlying approach combines AI screen reading understanding with system-level/application-level APIs.

It's worth noting that the "Doubao Phone," jointly developed by ByteDance and Nubia, heavily utilizes system-level permissions and screen reading capabilities, rather than APIs. You could say that the Doubao Phone primarily adopts a "forced" approach, implementing its functionality without consulting app developers beforehand, which provides ammunition for national-level apps to block and boycott it.

The Gemini smart agent that Samsung and Google have implemented in the Galaxy S26 series combines elements of both. According to information revealed by Samsung, it supports the top 200 apps in its app store (but only specific apps are guaranteed to work well, as will be detailed later) – indicating that Samsung and Google have at least largely reached out to these app developers.

Let's take a look at the experience of a Wired magazine editor: She simply called up Gemini, told it she wanted to go to the airport, and the Gemini app itself opened a "virtual window" to open Uber and started executing the action in the background. Users can click to enter at any time to view the execution process of Gemini.

Since there are several different airports in the area, Gemini quickly reminded the user to select the appropriate destination; when placing an order, Gemini also pushed the interface in front of the user to make it easier for the user to select the appropriate vehicle and make payment.

Gemini's "virtual window" can be understood as a sandboxed "virtual machine," a consideration by Google for protecting user privacy. Previous Gemini systems ran on Android, but this new Gemini agent operates within this sandbox and does not access other parts of the device when interacting with applications.

One more thing: If you've used smart agent products with cloud computer/cloud phone capabilities such as Manus, Kimi computer, or AutoGLM, you should easily understand the logic of this Gemini virtual machine.

This is a fairly simple task; many domestically produced AI mobile assistants had already mastered this scenario a year ago.

Gemini's even more killer feature is its combination with the screen reading and information capture capabilities that it has already been developing for a long time.

For example, when a user and their friends are talking about ordering pizza for a party, the user can simply call out Gemini and say "get the order sorted out." Gemini can then directly retrieve the pizza shops mentioned in the chat, and even specific types of pizza, and organize everyone's needs.

Then, users can have Gemini order takeout on the food delivery platform Grubhub. The AI ​​will automatically add all the food items to the shopping cart in the background according to the order requirements that have just been sorted out, and then deliver it to the user for confirmation and order placement.

Sometimes, ordering food doesn't go smoothly, and Gemini will try to handle unexpected situations herself and provide solutions for users. Once, when a pizzeria limited the number of large pizzas that could be ordered during peak hours, Gemini asked if she could order two medium pizzas instead.

Here's another example: A Google Keep note was used to list the attendees of a barbecue party, with vegetarians noted. Gemini could calculate the total number of hot dogs and buns needed for the entire party, then send it to purchase the ingredients, and within minutes all the items were in the DoorDash shopping cart.

Sammer Samat, president of Google's Android ecosystem, revealed that Gemini did not "memorize" the steps and routes of these platform operations in advance, but rather it truly used reasoning abilities to mimic how humans view the screen and perform the next operation. This means that Gemini can unleash its potential in more scenarios in the future.

Here you can see that Gemini's initial focus is on food ordering and ride-hailing scenarios, which is more similar to what Qianwen did before the Spring Festival.

Another "bean bun phone" from Android official.

Compared to the truly "all-in-one" Doubao mobile assistant, which can even help find WeChat favorites (at least before the boycott), Gemini's current capabilities are still quite limited, focusing on everyday scenarios such as ride-hailing, food delivery, and grocery shopping. Although its underlying technology is stronger, the actual user experience is not much different from that of domestic mobile AI assistants such as HarmonyOS's Xiaoyi and Honor's YOYO.

However, as mentioned at the beginning of the article, Google controls the entire Android ecosystem and has absolute influence and control.

With the release of Gemini's automation capabilities, Google also revealed in detail the underlying layout and future plans of the Android system—there are two directions, which, simply put, are both "Apple" and "Doubao".

First, Google released a framework called "AppFunctions" last year, which allows developers to expose entry points for specific functions and features of their applications so that AI assistants can call them.

Google likens AppFunctions to Android's "Model Context Protocol" (MCP), which can be simply understood as a dialogue standard that helps third-party apps and AI models connect.

This framework is similar to Apple's App Intents. In Apple's vision, users can use Siri to operate various apps to achieve functions, and the underlying implementation is through App Intents—given the delayed arrival of the next-generation Siri, App Intents are sufficient to provide a decent experience.

The same applies to Google's AppFunctions.

For example, a user might give an instruction to find a recipe in a friend's email and add the relevant ingredients to a shopping list. Upon receiving the command, the AI ​​first uses the "search" function in the email app to retrieve and extract the relevant content, and then uses the "shopping list" function in the notes app to fill in and organize the data.

Some AppFunction features have already been implemented in the Samsung Galaxy S26 and One UI 8.5 system. For example, users can give commands to Gemini to find specific photos in their album and send them to friends via SMS.

It's important to note that throughout the entire process, Gemini doesn't need to open the Photos and Messages apps, or even leave the Gemini app. Instead, it uses AppFunctions to retrieve the corresponding entry points and execute the operations within Gemini, making it more efficient.

Essentially, the implementation based on AppFunctions follows the same logic as the old API path. This is a pre-arranged solution.

However, not all apps are properly adapted. No worries, Google has another solution.

In an article published yesterday on the Android developer blog, Google explicitly stated that the company is also developing a UI automation framework that will allow AI assistants and third-party applications to mimic humans and directly open the app and perform step-by-step operations.

—This is the knock-off of the "Doubao Phone".

However, despite Google's claim that UI automation will take on the real "heavy lifting" in the future, in this Galaxy 26 series, UI automation is only an "early preview version".

If AppFunctions requires app developers to do additional adaptation work, then UI automation frameworks leave all the work to the AI ​​agent, requiring no additional adaptation. However, the effectiveness depends heavily on the capabilities of the AI ​​agent, and its advantage lies in its ability to cover a large number of applications as soon as it goes live.

As you can see now, in Google's Android Gemini intelligent agent project, AppFunctions and UI automation are two complementary paths: ensuring maximum compatibility through standardized and traceable interfaces, while laying the foundation for screen reader interaction modes that truly represent the future.

Google also stated that this is not just a feature exclusive to Gemini, but a feature of the Android system.

This also means that in the future, whether it's the AI ​​assistant built into the phone manufacturer or third-party applications like ChatGPT, they will be able to call AppFunctions to perform tasks or "understand" the phone's UI to perform automatic operations.

It's worth mentioning that even though Gemini isn't available on the Chinese version of the Samsung Galaxy S26, the Bixby assistant can still perform functions like ordering takeout, hail a ride, and comparing prices on e-commerce platforms. We can reasonably infer that Samsung has also found a model supplier in China to replace Gemini. As for which of these large and small companies is involved, it likely depends on who has achieved more outstanding results in mobile smart agents over the past year.

The path to AI smartphones will not be a solitary endeavor.

Last year, the "Doubao Phone" made a stunning debut, only to be unfortunately discontinued prematurely. While deeply regrettable, this also prompts us to consider: is the AI ​​automation model the ideal model for AI phones?

This question won't be answered for at least three to five years. At least, Doubao phones aren't acting alone; Google, which owns the Android system, has also chosen this path and has far more influence.

After the Beanbag phone became popular overseas, some netizens began to imagine that if Google promoted this technology on Pixel and Android phones, the prospects would be very broad.

Although I think Google doesn't really have a very clear answer to the question of "AI phone", it seems more like they have AI, system and hardware at hand, and they're trying each direction, hoping that one of them will work out.

But at least Google has set a good example of "system-level automation" for Android, and many new phones in the future have the potential to become "steamed bun phones".

This wave may not be limited to the Android camp. Don't forget, Apple has partnered with Google, and Gemini will become the technical support for Siri. And App Intents and App Functions are very similar…

Looking a little further ahead: Gemini intelligent agents aren't even limited to AI smartphones. Sammer Samat envisions that in the future, smart glasses, AI pendants, and even cars, as long as they have Gemini, can use it to accomplish complex tasks—of course, such scenarios are still far from being realized.

However, Google has only successfully implemented AI automation at the technical level. The establishment of a paradigm doesn't mean the problems have disappeared. The various contradictions encountered by Doubao phones will become challenges that later entrants will inevitably face.

First and foremost, there are privacy and security concerns. Google's vision is ambitious; in the future, the ability to access and manipulate mobile apps will extend beyond Gemini. Some third-party AI applications will be able to delve deeper into the core of user data, and if malicious applications in disguise exploit these interfaces, it could cause even greater losses.

The more intense conflict lies in the struggle among mobile phone hardware manufacturers, model/intelligent agent capability providers, and large platform applications for the new "entry point" in the AI ​​era. This was also the most difficult hurdle for the original Doubao phone to overcome.

After all, using Gemini to hail a ride may mean that users will no longer see Uber's membership promotions and advertising recommendations, and may even lose brand loyalty, directly harming the revenue of app service providers/advertising industries.

China has its internet/AI giants, and the same is true overseas. Long-time rivals like Meta and Amazon possess powerful platforms and ecosystems, and they may not be willing to open up to Google and allow Gemini to automate everything. Whether citing privacy, security, or platform rules, restrictions and higher barriers to entry will inevitably arise, and the competition will intensify.

Google, at least, is very confident about the future. Sammer Samat believes that AI technology is already in progress, and instead of racking their brains to fight against it, developers should think about a suitable way to embrace it.

The clash between the new and the old is inevitable, and the ultimate victors will be those players who bravely pursue change on the eve of it.

References:
https://android-developers.googleblog.com/2026/02/the-intelligent-os-making-ai-agents.html

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

ifanr | Original Link · View Comments · Sina Weibo

When using AI to generate images, one dilemma always arises: speed or quality?

However, speed and quality are not necessarily mutually exclusive. Just now, Google officially released their next-generation image generation model: Nano Banana 2 (Gemini 3.1 Flash Image).

Without many world-shaking slogans, it simply packed better image quality and a more human-like understanding into a completely new underlying architecture. This one thing makes AI-generated images feel less like a matter of luck and more reassuringly usable.

Having connected to the entire internet, this time the AI ​​truly understands what you're saying.

To explain this change, we need to recall how difficult it was to use AI-generated images three years ago.

If you ask it to draw "braised pork," it might obediently draw a piece of burning meat; but if you ask it to write a Chinese sentence on a poster, it will often produce a bunch of meaningless scribbles. This lack of common sense about the real world is the most frustrating aspect of first-generation AI.

The current Nano Banana 2 has changed a lot. Like its predecessor, the Nano Banana Pro, it accesses Gemini's vast real-world knowledge base and can combine it with real-time information from web searches, making it feel more like a worldly-wise person with some common sense.

The first change it noticed was that it began to better understand space and proportion.

In the case above, the AI ​​accurately recreated Shanghai's landmarks and handled the light and shadow and perspective relationships between the giant cat and the miniature city extremely naturally.

The most noticeable change is that it has finally learned to recognize and write characters. For example, when asked to paint a traditional Chinese ink painting of "Mooring at Maple Bridge at Night," it not only neatly wrote the entire poem "The moon sets, crows cry, frost fills the sky" in calligraphy at the top of the painting, but even the layout and the artistic conception of the ink painting were quite well captured.

Beyond its poetic qualities, it can also handle fairly complex UI scenarios—in the image below, the complex semi-transparent data panel, the floating shopping list, and the precise Chinese display are all arranged in an orderly manner by AI, and the hierarchical relationship between the information is truly clarified.

The layout of the double-page black-and-white Japanese comics is extremely meticulous, and it's a piece of cake.

Alternatively, this Chinese infographic on "Kung Fu Tea" with step-by-step instructions provides a visual solution that can be used directly, from layout to artistic conception.

An early beta user of Nano Banana 2 gave a fairly objective assessment: "It's not perfect, but it's the first model that can handle truly complex images and charts with a certain degree of consistency."

To test the limits of this new model's comprehension, he casually threw out an extremely tricky test question: "Draw me a picture of 'Where's Waldo,' set in ancient Venice, but the object you're looking for can't be a person; it has to be an otter wearing a blue striped flight suit."

Nano Banana 2 finally figured out the logic, not only avoiding any mistakes in the drawing, but also delivering a solid answer.

Make up quickly, so we don't have to choose one or the other anymore?

Besides its understanding of common sense, the powerful "subject consistency" is another major selling point of this Nano Banana 2 update.

During a single generation process, it can maintain the facial features of up to 5 characters or the appearance of 14 items. This means you can confidently use it for drawing comic strips or creating film storyboards.

Moreover, its image quality is high enough to be used directly in work.

It can handle everything from 512px images to 4K ultra-high-definition posters. Input a prompt about "Chongqing hot pot," and it will generate a cyberpunk-style rainy night street scene, with meticulous detail on the wet asphalt, the reflections of red and blue neon lights, and the "Open 24 Hours" sign.

It can also handle the Pop Art style with its strong color intensity.

Or perhaps this kind of fashion editorial, which possesses a touch of absurdity yet exudes a high-end feel:

In this case study, taken from a bird's-eye view, the AI ​​masterfully simulated the unique texture of an old-fashioned LOMO camera. The actress stands alone on the ground covered with black and white posters, creating a powerful cinematic tension and a strong sense of narrative.

However, it is not without its obvious shortcomings. When it forces anime characters, pencil sketches, and Nendoroids into the same real coffee shop scene, the integration of the sketched characters appears very awkward, and the transition at the edges is not natural enough.

Clearly, in terms of cross-dimensional fusion, it is far less natural than the previous generation of models, and there is still room for improvement.

In fact, despite the official blog's effusive praise of Nano Banana 2 (Gemini 3.1 Flash Image), the actual quality and speed of the generated images did not show any noticeable improvement, and in some scenarios, they were even inferior to the previous generation models.

What truly allowed the Nano Banana 2 to gain a foothold was its extremely affordable price-performance ratio.

Starting today, you can easily use it in the Gemini app and Google search box. Regular users without a subscription plan can get 100 free tickets within 24 hours; while Pro subscribers get up to 1000 tickets.

For developers, the API price has been halved, now only half the price of the previous generation Pro model. This translates to a cost of generating a single 4K high-resolution image of around $0.15.

Of course, the more AI can mass-produce high-resolution images at extremely low cost and speed, the more uneasy people become. With fake images flooding the internet, the saying "seeing is believing" is no longer reliable. If anyone can generate a nearly indistinguishable photo in a second, how can we tell the difference?

Google is well aware of this, so they have upgraded their anti-counterfeiting technology accordingly. Nano Banana 2 further enhances support for SynthID digital watermarking and C2PA content credentials, making it clearer whether an image is AI-generated and how it has been modified.

According to statistics, this verification function in Gemini has been called more than 20 million times since November last year.

The development of AI-powered graphics in the past two years has indeed been dazzling. We experienced the initial amazement of Nano Banana Pro, as well as the tedious setup and long wait. The emergence of Nano Banana 2 attempts to further combine quality and speed, and greatly lowers the barrier to entry for users.

An idea that pops into your head no longer needs repeated revisions and anxious waiting. The moment you press enter, it's there. Natural, simple, and instantly available—this sounds ordinary, but being able to do it is actually quite remarkable.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

ifanr | Original Link · View Comments · Sina Weibo

After four years, the red iPhone may be making a comeback.

According to Bloomberg reporter Mark Gurman, Apple may introduce a deep red color option for the "flagship model" of the iPhone 18.

Related reading: Is the red iPhone making a comeback? (PRODUCT)RED and Apple's 20-year rise and fall.

Apple is reportedly testing a new deep red color option for the iPhone 18 Pro series. This marks the first time red has appeared on an iPhone Pro model, and its return after four years since the iPhone 14 in 2022.

(In March 2017, the special edition red iPhone 7 and iPhone 7 Plus were released, and Apple has not released any since the iPhone 14.)

The return of red is thanks to the unibody aluminum design adopted by the iPhone 17 Pro. This new process significantly improves the color performance and consistency of the metal surface, making the Pro series more feasible in terms of color expansion.

According to leaks, the iPhone 18 Pro may adopt a "dual flagship color" strategy: a deep red as a new color, while retaining the popular "Starry Orange" from the previous generation. Previously rumored colors such as purple and brown were merely test variations of the deep red under different manufacturing parameters.

It's worth noting that Apple has adopted a completely different color strategy for the foldable iPhone Fold: offering only two basic colors, dark gray/black and white/light silver, returning to the classic color scheme logic of the original iPhone.

Regarding sales information, blogger "Shana Digital" recently posted that the iPhone Fold foldable screen will begin mass production around July, just like the iPhone 18 Pro, and will be released simultaneously.

This news aligns with analyst Ming-Chi Kuo's previous predictions, who indicated that these two products would be launched in the second half of 2026, while the standard iPhone 18 and iPhone 18e would be delayed until the first half of 2027.

In addition, on New Year's Eve, iFanr received an invitation from Apple, officially announcing that Apple will hold its first product launch event of 2026 at 10 pm on March 4.

Based on previous predictions, this spring event will be the start of Apple's packed product lineup for 2026, and we expect to see a large number of new products unveiled, including but not limited to:

• iPhone 17e: Featuring the same A19 processor as the iPhone 17, and adding MagSafe; the new phone will also serve as a testing platform for Apple's next-generation self-developed cellular network baseband and wireless chips (C1X and N1); and it will most likely switch to a dynamic island screen.
• The Muji MacBook with the A18 processor is expected to feature the A18 Pro processor and a 12.9-inch screen. The new machine will be available in a variety of vibrant colors, including light yellow, light green, blue, pink, as well as classic silver and space gray.
• MacBook Pro M5 Pro/Max
• The new generation iPad Air and Muji iPad
• New Studio Display, Apple TV, and HomePod mini

Gurman also recently reported that Apple is accelerating the development of three new wearable devices: smart glasses, wearable accessories, and AirPods with cameras.

Positioned at the highest end of the value chain, this smart glasses, codenamed N50, is the flagship product in Apple's AI hardware lineup, competing with the Meta AI glasses series. Similar to Meta's Ray-Ban glasses, it does not have a display screen, but interacts through speakers, microphones, and cameras.

Apple will differentiate itself through high-end materials in manufacturing processes and camera technology, and will not rely on third-party frames, instead developing its own designs entirely. Reportedly, Apple has already prototyped various colors and materials for different models.

In terms of functionality, the glasses will be equipped with dual cameras: one for taking high-resolution photos and videos, and the other for computer vision, used to understand the environmental context (similar to the technology in Vision Pro).

As for the new AI-powered accessories and AirPods, this represents a lighter, more accessible approach. This is a low-barrier solution Apple has prepared specifically for users who don't want to wear glasses.

• One is a pendant similar to the Humane AI Pin, which can be clipped to clothing or worn as a necklace, acting as an "external sensor" for your phone.
• There is also an AirPods with a camera, which mainly relies on a low-resolution camera to assist AI operation. It has a discreet design and is widely available.

Currently, the development progress of these three products is different: AirPods with cameras are progressing the fastest and are expected to be released as early as this year; AI accessory devices are currently in the early stages and, if not canceled, will be launched as early as next year; while the mass production plan for smart glasses is scheduled to start in December and is expected to be officially released in 2027.

It's worth noting that the computing power of these three AI hardware products still relies on the iPhone, and they are not independent computing platforms.

The Spring Festival AI battle has come to an end: new models have emerged, and there's even a robot Spring Festival Gala.

During the joyous Spring Festival holiday, AI companies went all out to offer users a series of new products and marketing campaigns. We've compiled this super roundup of the "Spring Festival AI Battle":

• Alibaba's 1000 Questions

Qianwen officially announced yesterday that during the Spring Festival, it helped users place nearly 200 million orders with a single sentence. It also launched its first wave of "3 billion RMB worth of free orders during the Spring Festival" on February 6th.

On New Year's Eve, Alibaba also released and open-sourced the Qwen3.5 series of models, and quickly deployed them on the chat.qwen.ai platform.

Benchmark results show that Qwen 3.5 leads other open-source models in several core capability dimensions, including multimodal understanding, complex reasoning, programming, and agent intelligence, and is comparable to top-tier closed-source models such as Gemini-3-pro in multiple benchmarks.

Related reading: Alibaba presents its Spring Festival "main course"—1000 questions 3.5! I had it make a New Year's greeting webpage, and the result was unexpected.

• bean bag

Before the Spring Festival, the Doubao team successively released the video generation model Seedance 2.0, the raw image model Seedream 5.0, and the Doubao large model version 2.0.

Seedance 2.0, in particular, has garnered significant attention and praise both domestically and internationally due to its exceptional visual consistency and unique multi-camera storytelling capabilities. Feng Ji, producer of *Black Myth: Wukong* and founder of Game Science, even praised it as "the strongest video generation model currently available."

Related reading: The most comprehensive beginner's guide to Seedance 2.0: 4 main entry points, popular gameplay, and detailed tips are all here.

The subsequently released Doubao Big Model 2.0 series achieved the highest scores in the industry on benchmarks such as VLMsAreBiased and OmniDocBench.

Related reading: I hand-crafted macOS using Doubao's large model 2.0; ByteDance is giving away another AI gift for Chinese New Year after Seedance 2.0.

• Tencent Yuanbao

Before the Spring Festival, Yuanbao updated the group chat AI gameplay of "Yuanbao Party" and subsequently launched the "Share 1 billion yuan cash red envelope" event.

On February 18, Tencent announced that Yuanbao's daily active users (DAU) exceeded 50 million and monthly active users (MAU) reached 114 million.

The day before, Tencent released a report on its 1 billion yuan cash red envelope giveaway event, showing that the main venue for the Yuanbao Spring Festival had over 3.6 billion draws and users completed over 1 billion AI tasks through the "Creation" section.

• Robot Spring Festival Gala

The 2026 Spring Festival Gala witnessed the "highest level of mechanical intelligence" in history: Songyan Power Robot, Cai Ming, and Doubao AI jointly performed a skit; Yushu led the martial arts program "Wu BOT" and cosplayed the Monkey King; Magic Atom's quadruped robot dog, wheeled robot dog and robot panda appeared on the stage.

Meanwhile, Seedance 2.0 was used in the visual production of several programs, including the opening dance "A Miracle Imminent", the dance "Ancient Charm of the Silk Road", the song "Galloping Cheese Horse", and so on.

Related reading: Unexpectedly, the first field that robots "replaced" was the Spring Festival Gala.

• The creator of OpenClaw announces joining OpenAI

On Lunar New Year's Eve (February 16), Peter Steinberger, the developer of the globally popular AI agent building platform OpenClaw, officially announced that he had joined OpenAI.

It is understood that OpenAI offered very attractive terms for the negotiations—to have them take charge of personal intelligent agent-related work at OpenAI, participate in the development of other products, and even discuss the establishment of a foundation specifically to operate the OpenClaw open source project.

Related reading: OpenClaw creator announces joining OpenAI, Zuckerberg's attempt to recruit him fails.

• Gemini 3.1 Pro Released

On February 20, Google officially released the preview version of Gemini 3.1 Pro.

In terms of performance, in the ARC-AGI-2 benchmark, which tests the ability of a model to solve new logical patterns, the Gemini 3.1 Pro achieved 77.1%, more than double that of the previous generation 3 Pro (31.1%), and also outperformed Anthropic's Opus 4.6 (68.8%) and OpenAI's GPT-5.2 (52.9%).

In terms of programming capabilities, the competitive programming benchmark LiveCodeBench Pro achieved an Elo score of 2887, surpassing the 3 Pro's 2439 and GPT-5.2's 2393. On SWE-Bench Verified, the 3.1 Pro scored 80.6%, essentially matching the Opus 4.6's 80.8%.

Related reading: Gemini 3.1 Pro released! Tsinghua University's Yao Shunyu promotes it; Karpathy: The era of app stores is over.

• The Dark Side of the Moon becomes the fastest domestic company to advance to the decacorn level.

According to The Paper, after raising more than $1.2 billion in funding, Kimi, a large model unicorn, has received the highest amount of funding in the large model industry in nearly a year, and has also set a record for the fastest progress of a domestic company from its establishment to becoming a decacorn company (valued at more than $10 billion).

Public data shows that it took ByteDance more than 4 years to break through the $10 billion mark, and Pinduoduo more than 3 years. Kimi, on the other hand, went from a $300 million angel round to a valuation of over $10 billion in just over two years, and his net worth has increased more than 30 times.

According to reports, less than a month after the release of Kimi's K2.5 flagship model, Kimi's cumulative revenue in the past 20 days has already exceeded its total revenue for the entire year of 2025. The growth is mainly driven by a surge in global paying users and API calls, especially the rapid growth of overseas paying users. Kimi's overseas revenue has now surpassed its domestic revenue.

• OpenAI will release a ChatGPT Pro Lite version.

According to blogger "Tibor Blaho" on February 21, OpenAI will be adding a Pro Lite subscription plan to ChatGPT, priced at $100/month (approximately RMB 690).

Based on the leaked information, the Pro Lite may offer 3 to 5 times the Deep Thinking Model usage quota of the Plus version, but it is not expected to include the Pro Mode exclusive to the Pro version.

• Nvidia to invest $30 billion in OpenAI

According to the Financial Times, Nvidia is in final-stage talks with OpenAI for a potential investment of $30 billion (approximately RMB 207.2 billion).

According to sources familiar with the matter, this $30 billion equity investment is part of a larger funding round that is expected to raise more than $100 billion and increase OpenAI’s valuation to $730 billion, excluding the new funds.

It's worth noting that sources also added that OpenAI will reinvest most of its new capital in Nvidia hardware, but the two companies will not proceed with their $100 billion multi-year investment partnership announced in September.

According to the terms of the previous $100 billion agreement, as OpenAI's demand for computing power grows, Nvidia will invest $10 billion in ten installments over ten years in exchange for a significant stake in the artificial intelligence startup.

The sequel to Yushu's "Wu BOT" is now available online.

Yesterday, Unitree Robotics released a video titled "Unitree BOT Temple of Heaven Blessing" and showcased its G1 robot swarm demonstration.

It is worth mentioning that during this year's Spring Festival Gala, Yushu used dozens of G1 robots to achieve the world's first fully automated performance of a humanoid robot swarm, setting several world records.

Related reading: Spring Festival Gala robot blows up international websites! Foreigners analyze frame by frame to see if it's real or fake: exclaiming "amazing!"

In an interview with CCTV's Spring Festival Gala, Wang Xingxing also exclusively revealed the "training secrets" of "Wu BOT". The report stated that for these two events, the Spring Festival Gala production team and the Unitree Robotics team worked together to refine several "world's first" robot maneuvers, such as "continuous fancy table-flipping parkour" and "Airflare's seven and a half giant spins".

Wang Xingxing also explained that the catapult launch that made people nervous was actually a humanoid robot jumping onto a customized catapult, which could then "jump" to a height of 2 to 3 meters, and perform a frontal somersault and a side somersault in the air before landing smoothly.

Ant Financial's "Afu" app and Alipay's "AI Pay" both surpass 100 million users.

On the seventh day of the Lunar New Year in the Year of the Horse, Ant Group released its actual performance data for the Spring Festival holiday, showing a full-scale surge in its core AI business:

Alipay's "AI Pay" has surpassed 100 million users, and the total number of users of the Ant Fortune APP has also exceeded 100 million.

In the first Spring Festival when AI technology entered large-scale application, Ant Financial achieved a differentiated breakthrough in the Spring Festival AI battle thanks to its performance in the payment and health fields.

In the realm of AI-native payments, Alipay's "AI Pay" saw explosive growth during the Spring Festival as AI technology penetrated deeper into everyday consumption scenarios.

Data shows that after surpassing 120 million payment transactions on February 12, the number of users also officially exceeded 100 million. With this, Alipay's "AI Pay" became the world's first AI-native payment product to surpass 100 million in both payment transactions and users, marking the product's official entry into large-scale commercial use.

In the realm of independent AI health applications, "Ant Afu" quickly gained national attention and garnered widespread interest through the "Health Blessing" campaign and its appearance on CCTV's Spring Festival Gala in the Year of the Horse.

Latest data shows that the total number of users of the Ant Financial AI App has rapidly exceeded 100 million, making it the world's number one health AI app. During the Spring Festival holiday, the enthusiasm of young people returning home to teach their families how to use Ant Financial's AI App drove a continuous surge in downloads, topping the Apple App Store's overall app download chart for several consecutive days.

It's worth noting that 52% of the new users of the Ant Financial App during the Spring Festival came from third-tier cities and below.

Honor will launch its first humanoid robot.

According to the Science and Technology Innovation Board Daily, Honor will launch its first humanoid robot at the 2026 Mobile World Congress in Barcelona (MWC 2026), focusing on the consumer market and becoming the world's first mobile phone company to enter the humanoid robot market.

According to Xinhua News Agency, at the Honor 400 series launch event last May, Honor CEO Li Jian announced the company's entry into the robotics industry, recounted the behind-the-scenes story of employees and robot research and development, and revealed that the robot they developed had reached a running speed of 4m/s, breaking the previous industry record.

It's worth mentioning that Honor also showcased its ROBOT Phone project last year. This product boasts "AI brain + robotic mobility + ultimate imaging," featuring a foldable and multi-angle rotating gimbal camera.

Tesla's wireless charging technology approved

The U.S. Federal Communications Commission (FCC) has officially approved Tesla to deploy ultra-wideband (UWB) technology on its CyberCab model for high-precision positioning in its wireless charging system.

According to documents disclosed by the FCC, the system is defined as a pulsed UWB radio device that establishes point-to-point communication links through UWB transceivers mounted on both the vehicle and the ground charging pad.

When a vehicle approaches the charging area, the system can track its spatial position in real time and guide the CyberCab to precisely park at the alignment point with the highest charging efficiency.

It is worth noting that Tesla's UWB signal is only briefly activated when the vehicle approaches the charging pad, and mainly propagates in the near-field space between the vehicle body and the ground.

This low-power, short-duration, and directional communication strategy not only ensures positioning accuracy but also effectively avoids potential interference with surrounding electronic devices.

Tesla specifically emphasizes that the signal is completely turned off when not in operation, and the vehicle structure itself also acts as a shield, further reducing the risk of electromagnetic leakage.

Despite the regulatory approval for wireless charging technology, the CyberCab prototype currently in the public eye is still connected to a traditional Tesla Supercharger.

It is speculated that the initial production models will likely be compatible with both wireless charging and the existing Supercharger network—the former representing a future experience, and the latter ensuring current practicality.

Amazon denies that AI caused a 13-hour AWS outage.

Recently, the Financial Times, citing sources familiar with the matter, reported that in mid-December last year, Amazon's AI programming assistant "Kiro," when allowed to make modifications, determined that the best course of action was to "delete and rebuild the environment," resulting in a 13-hour outage of the AWS system.

Several Amazon employees also told the media that this was the second time in recent months that AI tools had been involved in a service outage, and a senior employee said that these outages were "small but entirely predictable."

Amazon strongly denied these claims in an official blog post titled "Correcting Financial Times Reporting on AWS, Kiro, and AI".

Amazon acknowledged that a single service outage did occur in a specific region last December. However, this was entirely due to a "user error (role misconfiguration)" during access control configuration, and not a defect in the AI ​​tool itself.

Amazon emphasizes that such issues can occur in any developer tool (whether AI-based or manual) or in manual operation.

Amazon stated in a blog post that the Financial Times' claim about a second AWS outage was "completely false."

An AWS spokesperson later clarified to GeekWire that the second incident mentioned in the report did not occur within AWS operations, but rather in another Amazon division, therefore the media report was incorrect on this point.

Amazon stated that it has implemented several new protective measures following the incident, including mandating human peer review for access to the production environment.

Ultraman: Humans eating for 20 years is not as effective as training AI.

Recently, at a forum on Express Adda in India, Sam Altman discussed many AI topics, from AGI to the US-China AI competition, and data center water usage issues.

Related reading: Ultraman: Humans eating for 20 years is not as good as training AI. The whole internet is in an uproar. Netizens: Say that again?

During the event, Altman interacted with the audience in a lighthearted and humorous manner, discussing various aspects of AI development. He stated, "AI has evolved from being able to do high school math to pushing the boundaries of human knowledge."

Altman believes the impact of artificial intelligence on jobs is difficult to predict. "I enjoy reading the history of technology. During the Industrial Revolution, people panicked about jobs, and they were surprisingly wrong. Change won't happen as quickly as some people in society predict. But eventually, people will find new jobs," he said.

The most popular part of the discussion was Altman's response to criticisms about AI's energy consumption: "People always talk about how much energy it takes to train AI models… but training humans also takes a lot of energy, it would take 20 years and consume that much food to make them smart."

Simply put, Altman feels it's unfair that criticisms of AI often compare the total energy consumption of "training a model" to the instantaneous energy consumption of a human "answering a question." In his view, AI is not an energy killer, but a necessity for future civilization, just as some worried about unemployment in the candle industry when the light bulb was invented.

WeChat launches face-to-face photo sending/receiving of red envelopes

According to our observations, WeChat recently launched a "My QR Code" section on its scan interface. After entering, you can see that in addition to adding friends, WeChat now also supports receiving photos, documents, and red envelopes face-to-face.

WeChat's face-to-face file transfer this time uses the "WiFi + nearby devices" method, similar to the transfer of chat history.

Our tests showed that this feature can complete the transfer of photos/files without adding friends.

This also means that another way to transfer data between devices will be added in the future. Currently, "face-to-face photo and file transfer" only supports iOS and Android, but HarmonyOS has also launched a related "scan entry".

Nothing Phone (4a) Design Revealed

Yesterday, Nothing officially released the appearance information of its new phone, Phone (4a).

The new phone retains the Nothing's transparent back cover design. Compared to the 3a's design, the 4a's battery and small board area are more organized and simpler, while the camera module uses a metallic color process and adopts a triple-camera setup.

It is worth noting that the Phone(4a) uses a brand-new Glyph Bar light, which is displayed through 9 independently controlled mini-LED beads, and the brightness is increased by 40% compared to the previous Phone(a) series; at the same time, the rear red recording light is retained.

The new device will be released on March 5th.

Samsung S26 Ultra promotional materials leaked

Recently, promotional materials for the Samsung Galaxy S26 Ultra were leaked.

In terms of appearance, the new phone's overall design is similar to the standard/plus version, featuring a double-stepped racetrack-shaped camera module, along with two independently placed lens groups. The front adopts a flat screen design and provides privacy display technology.

In terms of imaging capabilities, the S26 Ultra features a 200MP main camera (supporting 2x optical quality cropping), a 50MP ultra-wide-angle lens, a 5x 50MP super telephoto lens, and a 10MP 3x mid-range lens; the front camera is 12MP (supporting AF autofocus).

The battery has a capacity of 5000mAh and can be charged to 75% in 30 minutes. According to the information revealed, the European version of the new phone is expected to use the Exynos 2600 processor, which is said to have a 39% improvement in NPU performance, a 24% improvement in GPU performance, and a 19% improvement in CPU performance (compared to the S23 Ultra).

According to blogger "Shana Digital", the price of the Galaxy Buds 4 Pro, which was released at the same time, will be lower than that of the Buds 3 Pro.

The new device is expected to be officially released at 2 AM on February 26th.

Xbox's new CEO embroiled in public relations crisis

Recently, former Microsoft Core AI president Asha Sharma took over as Xbox CEO. To dispel doubts about her "lack of experience in the gaming industry," she publicly shared her personal Xbox account (Gamertag: AMRAHSAHSA) the day after taking office and frequently responded to player questions on Xbox.

However, this public relations initiative, which aimed to bring players closer together, unexpectedly backfired and is now facing widespread data scrutiny and a crisis of trust within the community.

According to overseas media outlets such as Insider Gaming and Windows Central, as well as data miners, Sharma's Xbox account, which it attempted to use to "prove player identity," showed serious signs of unfair score manipulation.

Public records show that the account was created in mid-January 2026, only one month ago. However, in just four weeks, the account played nearly 29 games and earned tens of thousands of achievement points.

Not only did she exhibit the unnatural phenomenon of "unlocking achievements across multiple unrelated games on the same day," but the records also showed that while playing BALL x PIT for 43 hours in a week, she also achieved 100% completion of all achievements in multiple walking simulation games such as Firewatch at lightning speed.

Besides the abnormal game account data, Sharma's frequent interactions on social media have also sparked huge controversy.

Tech4Gamers and other tech media outlets and industry insiders pointed out that the responses were characterized by highly structured, concise, and formulaic use of emojis, leading to widespread suspicion that they were "public relations rhetoric" generated with the assistance of AI tools such as Copilot or even fully automatically.

In a highly dramatic turn of events, when confronted by netizens with the question, "Are you an AI robot?", the former AI business executive simply replied with the onomatopoeic sound of a system prompt: "Beep Boop Beep Boop."

Ironically, in her inaugural address to all staff, Sharma made a clear commitment to "resist soulless AI garbage content."

Neither Microsoft nor Asha Sharma herself has made any substantive response to the allegations of "account data manipulation" and "AI-managed social media."

Genshin Impact will be adding a "voice chat feature".

Yesterday, Genshin Impact announced that the "Starry Sky" will soon be getting a "voice function," supporting the ability to speak freely.

According to reports, the voice chat feature will be launched in the upcoming "Fifth of the Month" update. The developers hope that the voice chat feature will make gatherings in the lobby smoother, cooperation in levels more seamless, and provide players with a more immersive and engaging adventure experience in the game.

Guangzhou Railway: Several major high-speed rail stations will be open 24/7.

According to China Railway, in order to cope with the continuously increasing return passenger flow, China Railway Guangzhou Group Co., Ltd. has taken multiple measures to increase transport capacity:

Additional overnight high-speed trains will be added from February 21 to 28 and from March 5 to 6, covering popular high-speed rail lines such as the Beijing-Guangzhou, Guangzhou-Shenzhen-Hong Kong, Hangzhou-Shenzhen, Guiyang-Guangzhou, and Nanning-Guangzhou lines, to fully meet the travel needs of passengers.

Specifically, during the period of additional night-time high-speed trains, stations such as Guangzhou South, Shenzhen North, and Changsha South will operate 24 hours a day, providing passengers with manual ticketing, refund, rescheduling, and priority passenger services around the clock.

It is reported that Guangzhou South Railway Station is expected to handle 21.35 million passengers during this year's Spring Festival travel rush, making it the "busiest high-speed rail station in the country".

The official statement also indicated that a "non-disturbing" service mode will be implemented on the train: the crew will walk, speak, and move quietly in the carriages to create a comfortable "sleep-inducing" environment for passengers, and will provide "point-to-point" reminders and wake-ups to passengers 15 minutes before arrival at the station to ensure that passengers can disembark on time.

China's box office revenue is expected to exceed 8 billion yuan in 2026.

According to CCTV.com, as of February 23, 2026, China's box office (including pre-sales) has exceeded 8 billion yuan, with the Spring Festival box office exceeding 5.6 billion yuan. Currently, China's box office leads the global film market with nearly 1 billion yuan more than the North American market.

The report states that this year's Spring Festival film lineup features a wide variety of genres, including comedy, action, animation, and science fiction, offering audiences a diverse range of viewing options.

Among them, "Pegasus 3" has grossed over 2.8 billion yuan in just 7 days, becoming the box office champion. Following closely behind are "Silent Night" and "The Swordsman: The Wind Rises in the Desert," both achieving impressive box office results.

Related reading: The narrative logic is still clichéd, but "Pegasus 3" is still the best Chinese-language racing film.

The report points out that, driven by the "2026 Film Economy Promotion Year," various regions have also carried out distinctive film economy promotion activities, such as issuing movie-watching vouchers to stimulate audience enthusiasm for watching movies.

At the same time, the "Movie+" initiative has undergone a comprehensive upgrade, with activities such as "traveling with movies, visiting markets, tasting food, and appreciating intangible cultural heritage" promoting the further deep integration and development of "movie IP + cultural tourism resources," transforming movie "traffic" into "incremental consumption."

The 4K restored version of "Rainbow Cat and Blue Rabbit Seven Swordsmen" is scheduled for release tonight.

On February 22, the official Weibo account of "Rainbow Cat and Blue Rabbit" announced:

The 4K restored version of "Rainbow Cat and Blue Rabbit: The Seven Swordsmen" has been officially scheduled for release. Starting February 24 (the eighth day of the Lunar New Year), it will be broadcast every night at 18:00 on the "Revisiting Classics" channel, which is guided by the State Administration of Radio and Television and hosted by China Radio and Television, with two episodes airing consecutively each day.

It is reported that the original "Rainbow Cat and Blue Rabbit Seven Swordsmen" is a martial arts animated TV series produced by Hunan Hongmeng Cartoon Communication Co., Ltd., with He Mengfan as the chief screenwriter, Wang Hong as the chief director, and Fu Yilin, Yan Ting, Liu Rong, Liu Juan and other main voice actors.

The story mainly revolves around the young hero Rainbow Cat, who, following his father's dying wish, shoulders the responsibility of saving the forest environment. Filled with resentment, he descends the mountain to find the other six sword heirs, and eventually gathers all seven swords to defeat Black Heart Tiger.

The animation premiered on CCTV Children's Channel on September 5, 2006, and won the "First Prize for Domestic Animated Film" at the 2006 Children's Program and Animation Excellence Awards and the "Best TV Animation Award" at the China Original Animation Awards Ceremony.

The third season of Wednesday has been announced to begin filming.

Recently, Netflix's hit series "Wednesday" officially announced the start of filming for its third season, which is currently in production near Dublin, Ireland.

To maintain the IP's global appeal and build a strong competitive advantage, Netflix is ​​injecting highly marketable fresh blood into the third quarter.

In addition to Eva Green, who was previously announced to play the mysterious Aunt Ophelia, Hollywood veterans Winona Ryder, Chris Sarandon, and Noah Taylor have also been confirmed to join the cast.

The return of renowned director and executive producer Tim Burton, who expressed high expectations for the reunion with these old friends and collaborators, provides a strong guarantee for the show's quality and signature gothic aesthetic.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

ifanr | Original Link · View Comments · Sina Weibo

If you want to destroy a company, you don't need to beat it, you just need to make people believe that it can be defeated.

IBM shares plunged 13.1% to $223.39 per share by the close of trading on Monday, marking the largest single-day drop since the dot-com bubble burst in 2000. Its market capitalization fell from $240.8 billion to approximately $208.7 billion, a loss of about $31 billion.

Was it a financial report bomb? No. Was it a major customer who absconded? No.

The trigger for all of this was a blog post published that day by AI company Anthropic, announcing that its programming tool Claude Code could help transform the old COBOL system, directly targeting IBM's core and most profitable legacy systems consulting business.

No one has proven that IBM's business has collapsed, but in the face of a panicked market, no evidence is needed.

COBOL was IBM's competitive advantage, but now it's being targeted by AI.

To understand why IBM suffered such a severe setback, we must first understand what COBOL is and how IBM made money from it.

COBOL is a programming language that originated in the 1950s and still drives about 95% of ATM transactions in the United States today, as well as the daily operations of a large number of banking, government, and aviation systems.

IBM has a long history of selling mainframes optimized for COBOL and providing retrofitting and consulting services around them.

The reason this business is profitable lies in the extreme difficulty of transformation. Understanding decades-old, undocumented code often takes consultant teams several years. This high cost of comprehension is IBM's core competitive advantage.

Anthropic's blog states that Claude Code can automatically sort out dependencies between thousands of lines of code, complete workflow documents that no one remembers, and identify system risks that would take human analysts months to discover.

It further claims that modernization projects that were originally measured in years can be compressed into a few quarters using AI.

When the market heard this news, it reacted immediately: IBM's consulting model, which relies on large teams stationed on-site for extended periods, may be about to see its pricing power loosen.

IBM's situation is therefore particularly sensitive.

In 2025, its consulting revenue was approximately $21.055 billion, infrastructure revenue was approximately $15.718 billion, and the cumulative contract value for generative AI exceeded $12.5 billion. When the cost of understanding legacy code is significantly reduced by large models, the perceived value of these revenues is diminished in the market's eyes.

Interestingly, IBM and Anthropic are not simply competitors.

In October 2025, the two companies announced a strategic partnership to integrate Claude into IBM's development toolset, with over 6,000 early users reporting an average productivity increase of approximately 45%. The same technology could become a tool for its own transformation, or it could be seen as a weapon to disrupt IBM.

But on the day when sell-off sentiment dominated, the market chose the latter.

IBM's plunge wasn't actually the beginning of this turmoil. Back on February 20th, the cybersecurity sector had already experienced a collective flash crash, a day later dubbed "Black Friday" for the software industry by some analysts.

The catalyst was Anthropic's announcement of the launch of Claude Code Security.

This tool can automatically scan for security vulnerabilities in the codebase, generate patches for manual review, and directly integrate functions such as vulnerability scanning and software composition analysis, which were originally handled by professional security tools, into the developer's daily workflow.

The market reaction was equally strong.

JFrog plunged 24.61% that day because its core business is highly dependent on the control of the software supply chain, which happens to be the easiest entry point for AI agents. Companies like CrowdStrike, Cloudflare, Okta, and Zscaler, whose main business is not code scanning, also experienced significant declines due to the "contagion" of sentiment. The entire sector lost tens of billions of dollars in a single day, and the Bug Fund, which tracks cybersecurity ETFs, fell to its lowest point in over two years.

Of course, many rational analysts have come forward to refute this, saying that such a sell-off is completely illogical.

For example, JPMorgan Chase believes that investors' concerns are exaggerated. Analysts at Wedbush Securities stated vehemently that this was "the most illogical trade" they had ever seen in their careers.

Their reasons are somewhat convincing.

While AI helps defenders identify vulnerabilities, it also helps attackers find them. By 2026, hackers have begun using large models to launch more precise phishing attacks, develop automated attack agents, and even implement "cue word injection" and "memory poisoning" against AI models within enterprises.

Even more problematic is the fact that a large number of unauthorized AI agents deployed within enterprises, once they gain access to internal systems and sensitive data, pose a security risk in themselves. This means that the need for security has not disappeared, but rather is expanding.

CrowdStrike's endpoint protection, Zscaler's zero-trust network access, and various identity and access management tools are precisely the infrastructure that must be relied upon to deal with these new AI threats.

What SaaS is most reluctant to admit, but is happening right now.

As APPSO has previously reported, AI has had a significant impact on the entire SaaS industry.

The turmoil surrounding IBM and security stocks is one facet of a larger crisis facing the entire enterprise software industry. The SaaS industry's prosperity over the past two decades has been built on a simple logic: the more employees an enterprise has, the more software seats it buys, and the more stable the vendor's subscription revenue becomes.

The emergence of AI agents has disrupted this equation.

Taking Claude Cowork as an example, it can navigate across systems autonomously and complete tasks such as data entry, lead scoring, and multi-step business approvals independently, eliminating the need for companies to maintain a large-scale frontline operations team.

A finance department that previously required five people to log in with their own accounts may now only need one manager using AI tools, thus eliminating the subscription fees for the other four positions. This phenomenon represents the most direct threat currently facing SaaS vendors.

In this transformation, the most vulnerable are SaaS vendors that provide general functionality and primarily serve small and medium-sized enterprises, such as HubSpot, Atlassian, and Asana. The workflows of these products are relatively standardized and can be easily imitated and replaced by AI.

The more general the functions and the more standardized the processes of software (such as scheduling and task management), the easier it is for AI to directly replace them.

This impact is not limited to large companies; it is also a devastating blow to ordinary entrepreneurs.

An entrepreneur named Ira Bodnar posted an article titled "Claude Kills All Startups" on the X platform, which quickly garnered over 2.92 million views. Just two months prior, he had developed an AI product that automates the management of Google and Meta advertising accounts for businesses, boasting several hundred paying clients and a conversion rate as high as 70%, showing strong growth momentum.

She woke up one morning to find that the transaction rate had dropped to 20%.

The reason is simple: Anthropic launched the Meta Ad Connector around the same time. Its functionality was incomplete; it could only perform analytics and couldn't directly manage accounts, but customers were already observing. Bodnar wrote in a post, "It will be able to do that in a few months, so continuing to develop it here doesn't seem very meaningful."

IBM lost its valuation, while Bodnar lost the very reason for its entire product category to exist. In contrast, what's even harder to factor into stock prices and harder to include in reports are the ordinary jobs that are disappearing.

Stock prices may fall but then rebound, but changes in the job market, though slow and painful, are far more real. When companies no longer need as many people to generate workloads, the first to suffer are the young people who haven't even started working yet.

Research from Stanford University's Digital Economy Lab indicates that among the professions most significantly impacted by AI, the employment rate for young graduates aged 22 to 25 has declined by 13%, while experienced senior architects remain secure in their jobs.

This differentiation shows that AI has not eliminated all human jobs like in science fiction movies, but has quietly eliminated them from the bottom up. In the past, newly graduated programmers always relied on writing simple basic code and fixing minor bugs to accumulate experience and climb up the ranks step by step.

Now, these "dirty and tiring jobs" are all dominated by AI agents.

Many large companies now have a tacit understanding: I will not lay off employees on a large scale, but if someone resigns, I will definitely not hire a new person and will directly fill the vacancy with AI.

The students' choices already demonstrate their assessment of the market. Data from MIT shows that its introductory programming courses peaked in the 2022 academic year and have been declining ever since, with enrollment in traditional computer science majors falling from 823 to 672, a drop of over 18%.

Meanwhile, MIT's newly established "Artificial Intelligence and Decision Making" program saw enrollment surge from 37 to 372 in 2022, an increase of approximately tenfold. Duke University's introductory CS course enrollment declined by about 20% during the same period, and Princeton's CS program experienced a similar decline in the number of junior and senior students.

Around the same time that IBM's stock price plummeted, a blog post titled "The Global Intelligence Crisis of 2028," published by Citrini Research, went viral on the X platform, garnering over 20 million views.

The author pretends to be looking back from 2028, and the picture he paints is not optimistic. Companies use AI to lay off employees, and the money saved is used to buy more AI. More powerful AI makes the next round of layoffs possible. The laid-off employees consume less, corporate revenue declines, and they have to rely on AI to further reduce costs.

Round and round, without end.

The author himself stated that this was merely a thought experiment, not a prophecy. However, after reading this article and then looking at the news of the day, it's difficult to say that what it describes is entirely imagined. SaaS companies' subscription revenue is shrinking, the necessity of the IT outsourcing industry is being questioned, and it's becoming increasingly difficult for recent graduates to find jobs—the first few links in this chain are already vaguely visible.

The International Monetary Fund estimates that about 40% of global employment is exposed to the impact of AI, with some jobs being enhanced and others facing contraction. The ultimate outcome will depend on the complementarity of different industries and policy responses.

The World Economic Forum predicts that new technologies will create approximately 78 million net jobs by 2030, but this redistribution process will inevitably be accompanied by a great deal of friction and pain.

We often use sensational terms like "disruptive," "explosive," and "game-changer" to describe the impact of AI. In my opinion, these terms are misused—the changes brought about by AI are more like a rising tide.

The rising tide doesn't discriminate, it doesn't discriminate, whether you're a century-old giant like IBM or an independent entrepreneur like Bodnar, the water level rises uniformly. Some stand on high ground, some on the mudflats, and only when the tide recedes do you discover who's been swimming naked—the problem is, this time the tide doesn't seem to be receding.

MIT students are already quietly changing their career paths, and big companies are also quietly waiting for employees to leave naturally and then not hiring replacements. Citrini's article has garnered over 20 million views, not because it is particularly well-written, but because so many people see in it the kind of unease they vaguely feel but haven't yet expressed.

Anxiety itself is not a bad thing. It at least shows that the tide has not completely submerged people, and there is still time to figure out where they stand and how to find their irreplaceable place in front of the machines.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

ifanr | Original Link · View Comments · Sina Weibo