China Presses Its Case for the AI Lead
The day's loudest signal came from China, where two moves suggested the gap with U.S. labs is narrowing fast. Moonshot AI's new Kimi K3 drew global attention for writing GPU code 14.82x faster than PyTorch, according to Crypto Briefing — a coding feat that challenges assumptions about America's lead. Demand was so intense that Moonshot paused new subscriptions entirely; in a post from its official @kimi_moonshot account, the company said sign-ups had surged beyond what its GPUs could handle.
Alibaba pushed just as hard. Its Qwen team previewed Qwen3.8-Max-Preview, a 2.4-trillion-parameter flagship, with a bold claim about its ranking: per Bloomberg (via Techmeme) and MarkTechPost, Alibaba says only Anthropic's Fable 5 — its Claude line — beats it. Separately, Futu (富途牛牛) reported that Alibaba, listed in Hong Kong under ticker 09988.HK, is preparing an open-source Qwen3.8, underscoring how China's biggest players are racing to set the pace.
Apple Takes OpenAI to Court
Apple filed a lawsuit against OpenAI, accusing the AI company of stealing trade secrets through its hiring practices, according to reports from multiple outlets including WION and Ukraine's Межа. One notable detail: the complaint reportedly leaves out the name of Jony Ive, the designer now tied to OpenAI's hardware ambitions — a conspicuous omission in a case built around talent and secrets.
The Money Question: Is AI Paying Off?
As spending soars, the industry spent the day arguing over how to measure return. Nvidia CEO Jensen Huang framed the upside in the largest possible terms, putting a $4 trillion market opportunity on AI — a projection now fueling speculation about how big the chipmaker could become.
OpenAI offered a sharper yardstick. CFO Sarah Friar argued the right metric isn't buying the cheapest compute but maximizing 'useful intelligence per dollar,' reframing the payoff debate for companies pouring money into the technology.
The business models underneath are shifting, too. Anthropic is overhauling how people pay for its top model, according to Firstpost, scaling back subscription access and moving many customers to API-based billing. It's a reminder that even the leaders are still figuring out how to price frontier intelligence.
Building Public Infrastructure — and a Compute Chief
Not all the ambition is commercial. A new nonprofit, Current AI, launched with $400 million in funding commitments and a sweeping goal: to build 'the World Wide Web of AI' — open, shared, public infrastructure that anyone can use for free, per TechCrunch.
OpenAI, meanwhile, moved to shore up the hardest constraint in the business: compute. The company named Uday Ruddaraju, a former xAI engineer, as its Chief Technology Officer, Compute, according to Firstpost, PTI and MSN. The role puts him in charge of the infrastructure that everything else depends on — fitting, on a day when demand for GPUs forced one rival to shut its doors to new users.
The Takeaway
The throughline is pressure. Chinese models are posting numbers that command attention, capacity is the binding limit, and the leaders are simultaneously litigating over talent, rewriting their pricing, and staffing up for a compute war that Huang thinks is worth trillions.