The Information AM

Save 25% on an annual subscription to read the most important news about technology and business first. For even more, Save $250 on The Information Pro for unlimited access to our proprietary org charts, databases and surveys.

1.	Andrej Karpathy Joins Anthropic By Rocket Drew Source: The Information

Andrej Karpathy, a founding member of OpenAI and the former director of AI at Tesla, is joining Anthropic, he posted on X on Tuesday. “I think the next few years at the frontier of LLMs will be especially formative,” he wrote.

The hire is a coup for Anthropic, given Karpathy’s legendary status in the AI industry—among other feats, he helped launch the first course on deep learning at Stanford and coined the term “vibe coding.” High-profile AI researchers have recently been leaving AI labs like Anthropic to start their own “neolabs,” but Karpathy’s move shows that some prominent researchers are still keen to work on the leading models.

In recent years, Karpathy has turned his attention to education, creating tutorials and YouTube videos about AI through his company Eureka Labs, which describes itself as “building a new kind of school that is AI native.” Karpathy said in his announcement that “I remain deeply passionate about education and plan to resume my work on it in time.”

2.	Google Unveils New Video Model, Search Upgrades By Erin Woo Source: The Information

Google on Tuesday announced a new video model, upgrades to search and a slightly streamlined coding agent offering at its annual conference.

The Google I/O event is officially a developer conference, but in practice Google has also used it in recent years to showcase new consumer AI offerings and to try to draw attention away from rivals OpenAI and Anthropic.

Tuesday’s announcements highlight where Google is making strides, particularly in multimodal AI and consumer growth, and where it’s playing catch-up, especially on coding. The new video model, Gemini Omni, is similar to its successful Nano Banana image model, which used Gemini as a base to make it better at creating graphics and keeping images consistent across edits. Gemini Omni aims to do the same thing for video creation, and comes as OpenAI has retreated from video by sunsetting its Sora tool.

Google also announced a new “agents” feature for search that can continuously scan for apartment listings or new product launches. It will start rolling out for paying subscribers this summer.

And Google announced new features for its Gemini consumer app, which it said now has 900 million monthly users (OpenAI said in February that ChatGPT had reached that milestone). The features included a “Daily Brief” similar to ChatGPT Pulse, personalized updates that OpenAI announced last fall and then said in December it was deprioritizing.

In December, OpenAI was focusing on ChatGPT as part of a “code red” response to Google’s Gemini 3 model, which was then considered to be the leading model. Since then, OpenAI has pivoted to focus more on enterprise as both it and Google chase Anthropic’s coding prowess, particularly after Anthropic’s release of its Mythos model.

Google on Tuesday announced a new desktop app for its Antigravity coding agent and said that it was merging another of its coding tools, Gemini CLI, into Antigravity.

3.	Goldman Sachs to Lead SpaceX IPO By Valida Pau and Cory Weinberg Source: The Wall Street Journal

SpaceX tapped Goldman Sachs for the top role on its initial public offering next month, according to the Wall Street Journal.

The “lead left” position would put Goldman in the driver’s seat for what will likely be the largest IPO ever. Morgan Stanley, Bank of America, Citi and JPMorgan are the other top banks on the deal.

Elon Musk’s company is expected to make public its IPO prospectus for the first time as soon as Wednesday.

4.	Sam Altman Offers YC Founders $2 Million in OpenAI Tokens For Equity By Jemima McEvoy Source: The Information

OpenAI cofounder and CEO Sam Altman late Tuesday offered to invest $2 million in every startup currently in the Y Combinator startup accelerator program—not in cash, but in OpenAI tokens.

“I am excited to see what will happen with tokenmaxxing startups, both for how they work internally and the products they can build,” Altman posted on X after unveiling the offer in a talk to YC founders late Tuesday. The offer is available to “every startup in the current YC batch,” he said.

Altman’s offer is a part of a pilot program launched by OpenAI and YC for the spring and summer participants in the program, which will collectively encompass hundreds of early-stage startups, according to a person close to OpenAI.

The offer is structured as a Simple Agreement for Future Equity or SAFE, meaning the startups would receive the tokens now while OpenAI’s ownership stake would be determined in a future financing round.

This new program follows the emergence of a trend known as “tokenmaxxing,” in which employees at big tech companies have sought to maximize their consumption of tokens and credits as a way to show off how much they’re using AI. For startups they are an increasingly expensive resource, though the cost depends on which model the person is using and how much they cache, or repeat, inputs.

For example, $2 million tokens would buy around 1 trillion worth of GPT-5 tokens. Some super users at Meta have used 73.7 trillion in tokens over a 30-day period, in comparison. Meanwhile, Peter Steinberger, the founder of OpenClaw who was hired by OpenAI in February recently revealed he spent $1.3 million to buy 603 billion OpenAI tokens in a single month.

5.	Alibaba Unveils New AI Chip as China Accelerates Adoption of Domestic Chips By Juro Osawa Source: The Information

Alibaba Group’s semiconductor design unit on Wednesday unveiled a new chip that can be used to train and run AI models. The move comes as the Chinese government pushes to accelerate the adoption of homegrown AI chips to reduce the country’s dependence on Nvidia.

Alibaba said the new AI chip, Zhenwu M890, delivers three times the performance of its predecessor. It is “exceptionally well-suited” for complex workloads that involve multiple AI agents working together, the company said.

Domestic AI chip suppliers like Alibaba’s T-Head chip unit have an increasingly important role to play, as China tries to become more self-sufficient with key technologies such as semiconductors. The Information last year reported that Alibaba had begun using its own chips to train some of its AI models.

Alibaba announced the new Zhenwu M890 chip at the annual Alibaba Cloud Summit conference on Wednesday. T-Head vice president Gao Hui, who gave a presentation at the conference, also unveiled the unit’s product roadmap for the next two years. It plans to release another new chip called Zhenwu V900 in the third quarter of 2027, followed by Zhenwu J900 in the third quarter of 2028, according to the roadmap Gao showed.

T-Head is part of Alibaba’s broader effort to compete in all aspects of AI, from infrastructure to models to applications. On Wednesday, Alibaba also launched its latest large language model called Qwen3.7-Max, emphasizing its capabilities in coding and complex multi-step tasks.