The Signal — May 2, 2026
A courtroom admission that undermines an entire lawsuit, a chip empire growing in Nvidia's shadow, and Meta buying its way into the robotics race.
Musk Admits xAI Distills OpenAI's Models
Under cross-examination in Musk v. Altman, Elon Musk conceded that xAI "partly" trains Grok by distilling OpenAI's models. OpenAI attorney William Savitt pressed the point during testimony on April 30, and Musk characterized it as standard industry practice before acknowledging his own company's participation.
The admission lands differently when you remember what the lawsuit is about. Musk is suing OpenAI for abandoning its nonprofit mission and hoarding powerful AI behind commercial walls. Meanwhile, his own lab is siphoning knowledge from those same commercial models to build a competitor. It's the plaintiff helping himself to the defendant's product while arguing the defendant shouldn't have a product.
Distillation itself isn't unusual. Smaller labs routinely learn from larger models' outputs. But admitting it under oath, in the middle of a trial premised on OpenAI's closed-source sins, creates a credibility problem that no amount of "everyone does it" can paper over. Musk's legal team now has to argue that OpenAI's commercialization harmed him while his company directly benefits from that commercialization.
Sources: WIRED · TechCrunch · CNBC · MIT Technology Review
Huawei's AI Chip Revenue on Track to Hit $12 Billion
Huawei expects its AI chip business to pull in roughly $12 billion this year, a 60% jump from $7.5 billion in 2025. The growth is almost entirely demand-driven: Chinese tech companies have nowhere else to go. U.S. export controls on Nvidia forced domestic buyers toward Huawei's Ascend processors, and the Ascend 950PR entering mass production in March gave them something worth buying.
The numbers tell a straightforward supply-and-demand story. Cut off access to the global leader, and local alternatives absorb the demand regardless of whether they match the incumbent's performance. Huawei doesn't need to beat Nvidia's H100 or B200 on benchmarks. It just needs to be available, and right now, it's the only game in town for Chinese hyperscalers building out AI infrastructure.
Whether the Ascend line can close the performance gap doesn't matter yet for revenue purposes. Orders are flooding in faster than Huawei can fill them.
Sources: Financial Times · The Deep Dive · IndexBox
Meta Makes Its First Robotics Acquisition
Meta acquired Assured Robot Intelligence, a startup focused on building AI models that help robots interpret and adapt to human behavior. It's Meta's first known acquisition in the robotics space, placing the company alongside Figure, 1X, Tesla, and Apptronik in the growing humanoid robotics field.
Humanoid robotics went from research curiosity to venture capital magnet over the past 18 months, and the major tech companies are now positioning themselves before the market consolidates. Meta's angle appears to be the perception layer: how robots understand what people are doing and respond appropriately, rather than the hardware or locomotion side.
No financial terms were disclosed. The acquisition signals intent more than immediate product plans, but it puts Meta on the board in a field that barely existed as a commercial category two years ago.
Sources: Bloomberg · TechCrunch · Business Insider
On the Editor's Desk
Held today: the Five Eyes (CISA, NSA, and partner agencies from Australia, the UK, Canada, and New Zealand) released joint guidance on agentic AI security, warning that organizations deploying autonomous AI tools are granting more access than they can realistically monitor. It's the first coordinated international framework addressing agent-specific risks. Important for the governance beat, but the guidance itself reads more as a warning than a watershed, and with three stories already carrying real consequence, it sits better as a flag for future coverage than a lead.
