|
|
|
Apr 10, 2026
|
|
|
|
|
|
TGIF! Senior OpenAI executives on the Stargate initiative depart. Meta is pulling top engineers into its new Applied AI Engineering division, whether they like it or not. Jerome Powell and Scott Bessent summon major U.S. banks to discuss Anthropic’s new model Claude Mythos.
|
|
|
|
Three senior OpenAI executives who helped launch the company’s original Stargate data center initiative have left or are preparing to depart the ChatGPT maker in the coming days, The Information reported Thursday. Peter Hoeschele, an OpenAI executive who played a key role in getting the Stargate effort off the ground, has already left the company. Two others—Shamez Hemani, who worked on compute strategy and business development, and Anuj Saharan, another leader in OpenAI’s compute organization—have announced their departures to colleagues. The departing executives are all going to the same new company, according to one of the people, though the name of that firm could not be learned. Late last year, OpenAI hired a new head of compute and infrastructure,
former Intel chief technology and AI officer Sachin Katti, who reports to OpenAI President and co-founder Greg Brockman. Several executives who previously reported to Brockman, including Hoeschele, were moved under Katti.
|
|
|
|
Meta Platforms is pulling top engineers from across the company into its new Applied AI Engineering division—whether they like it or not—as part of its aggressive push to step up its competitiveness in the AI model arms race. In an internal memo sent to employees on Wednesday and seen by The Information, Maher Saba, the Meta vice president who leads the new AAI organization, said the company has begun tapping staffers from other parts of the company to scale the team, plans for which were announced last month. He added that Meta will hold hackathons over the next two weeks to generate data and accelerate training for upcoming AI models. Saba described the effort as “P0” (priority zero), a Meta designation for the company’s highest priorities, and said it is aimed at helping Meta bring its models to
state-of-the-art performance more quickly and “compete in the AI race.” The AAI team will support Meta Superintelligence Labs, which is building new AI models. Saba pointed to Wednesday’s release of that group’s first new AI models, saying the AAI team “has a critical role in that work” and that the company had identified top software engineers to join it. “This ensures that we have the very best data quality from the start,” he wrote, adding that the team would generate tasks, data, and feedback to improve the models. In comments under the post, also seen by The Information, an employee asked Saba if a staffer selected to join the new org must accept the role. Saba responded that because AAI is a top company priority, “the transfers aren’t optional.” The AAI team reports up to Meta’s
chief technology officer, Andrew Bosworth. Meta Superintelligence Labs will remain responsible for model architecture, training, and research and is headed by Alexandr Wang, whom CEO Mark Zuckerberg appointed last year as Meta’s chief AI officer.
|
|
|
|
Federal Reserve Chair Jerome Powell and Treasury Secretary Scott Bessent on Tuesday summoned leaders of the major U.S. banks, including Citigroup, Bank of America and Wells Fargo, to discuss the cybersecurity risks from Anthropic’s new model, Claude Mythos, Bloomberg reported. The gathering underscored concern from U.S. regulators of the risks that advanced AI models pose to the banking system. The same day, Anthropic announced it would provide access to Mythos to only 40 organizations, including Apple, JPMorgan Chase and the Linux Foundation, so they can test their software for security vulnerabilities. At the time, it said it had been holding “ongoing discussions with US government officials” about the model and its offensive and defensive cyber capabilities. Mythos made waves even before this week’s
announcement after a draft blog post from Anthopric, mistakenly made public, said the new flagship model that could be used by hackers to “exploit [security] vulnerabilities in ways that far outpace the efforts of defenders.”
|
|
|
|
Mercor hit over $1 billion in annualized gross revenue earlier this year, up from a $500 million pace in September 2025, according to a person with direct knowledge of the company’s financials. The company pays out 60% to 70% of its top-line revenue to contractors, The
Information reported last year, meaning its net revenue is between $300 million to $400 million. The company is profitable on a free cash flow basis. The growth occurred before a group called Lapsus$ claimed to have hacked Mercor and offered to sell large swaths of data allegedly belonging to Mercor. The startup said it was “one of thousands of companies” impacted by a supply chain attack involving open-source project LiteLLM. The three-year-old startup—initially an AI-powered hiring platform—provides AI labs with specialized training data generated by contract workers with domain expertise in fields such as physics, finance and more. Its CEO Brendan Foody said in September 2025 that Mercor was “the fastest growing company of all time” after it scaled from a $1 million to a $500 million run rate in 17 months. A month later, Mercor raised $350 million at a $10 billion valuation led by Felicis. (This brief has been updated with information on Mercor’s gross and net revenue figures.)
|
|
|
|
OpenAI said Thursday it was making several changes to how it bills companies that use its AI coding agent, Codex, due to rising use of the tools. The changes are geared towards matching the amount customers spend more closely with how much they use the products. OpenAI on Thursday announced a new $100 per user per month subscription, which offers more access to Codex than its $20 per month tier, but less than its most expensive $200 per month tier. And last week OpenAI said it would start billing companies based on the number of tokens—or units of text processed by Codex—rather than the number of requests that users send to Codex. Sellers of AI have been constantly tweaking their pricing models in recent years as they try to balance price against the high cost of running the software. That’s been especially true
of the creators of AI coding agents, which developers often use to work on coding projects for hours or days on end. Anthropic recently said that users paying a flat fee for its Claude Code agent would need to pay extra if they connected Claude Code to third-party agent tools like OpenClaw, which tended to use Claude heavily. And last year, other AI coding startups like Replit and Cursor had to overhaul their pricing models because the high cost of running the coding agents was compressing their margins.
|
|
|
|
Meta Platforms has agreed to spend an additional $21 billion renting artificial intelligence chips from cloud provider CoreWeave from 2027 through 2032, bringing its total potential spending on CoreWeave to $35 billion. The latest commitment is on top of an earlier deal Meta had to spend $14 billion with CoreWeave through 2031. CoreWeave said in a filing with the Securities and Exchange Commission that the deal hinges on CoreWeave’s “satisfaction of delivery and availability of service requirements,” meaning it must provide Meta with reliable compute to fulfill the contract terms. Either party can terminate the deal “for cause,” the filing said. Earli | | | |
