Anthropic has signed a $35 billion cloud-computing agreement with Lambda, a San Jose-based provider backed by Nvidia, according to reporting first published by the Wall Street Journal and confirmed by Bloomberg. The compute is destined for Hut 8's Beacon Point campus in Nueces County, Texas, where Nvidia will reportedly hold the facility lease while Lambda supplies the capacity to Anthropic.
The arrangement covers roughly 350 megawatts, according to Bloomberg, on a campus with 1,000 MW of utility capacity already under an AEP Texas interconnection agreement. Hut 8 expects initial energization in the first quarter of 2027, with the first Phase 2 data hall following in mid-2028 — meaning the compute is a bet on future capacity rather than an immediate boost.
Neither Anthropic nor Lambda has confirmed the figure publicly, and none of the reports disclose GPU count, contract term details beyond "six years," or how obligations split across Anthropic, Lambda, Nvidia and Hut 8. The deal lands one week after Anthropic agreed to spend $45 billion renting capacity from Nscale's West Virginia campus — part of a broader pattern of AI labs locking in multi-billion-dollar compute well ahead of when the power actually comes online.
The European Commission designated ChatGPT a Very Large Online Search Engine on August 31, after OpenAI disclosed roughly 159.1 million average monthly EU users for ChatGPT search — more than triple the 45 million threshold that triggers the DSA's strictest tier. Reddit and Roblox were designated Very Large Online Platforms in the same announcement.
The Commission's reasoning is capability-based rather than category-based: ChatGPT qualifies because it can search the web and return information gathered from online sources, not because it's a popular AI product. That template gives regulators a path to bring Google's Gemini, Anthropic's Claude, or Perplexity under the same rules once their EU usage crosses the same line.
OpenAI has roughly four months from notification to comply with obligations including annual systemic-risk assessments, independent audits, minor-safety protections, a public ad repository, and data access for vetted researchers. Non-compliance risks fines of up to 6% of global annual turnover — the same ceiling that produced a €120 million fine against X in December 2025.
AIR, founded by Yair Saban (CEO) and Niv Hoffman (CTO) — both veterans of Israel's Unit 8200 intelligence corps — has raised $50 million across two seed rounds closed within weeks of each other: $10 million led by Sequoia, followed by $40 million led by Greenoaks.
The platform discovers agents already running inside a company, continuously vets the skills, tools and components those agents use, and blocks interactions that fail its security criteria. It also offers a marketplace of pre-vetted add-ons and skills for AI agents, aiming to become the supply-chain checkpoint that agentic AI currently lacks.
Google is preparing to launch Gemini 3.8 Flash, internally known as "Skimaki," with reports suggesting a possible release as early as Wednesday, September 2 — today. Google employees reportedly began internal testing of the model last month.
Coding remains one of the most consequential enterprise AI battlegrounds, and one where Anthropic and OpenAI have held a clear edge. The Flash launch arrives as Google's flagship Gemini 3.5 Pro continues to face delays, reportedly because it has yet to meet the company's internal coding-performance bar — making a strong Flash showing important proof that Google's models can compete on code, not just scale.
From September 2, Gemini Notebook — formerly NotebookLM — moves from fixed daily feature-generation caps to a compute-based budget that factors in prompt complexity, chat length, number of sources, and which features are used. The change rolls out to consumer accounts on web and mobile.
Limits now refresh every five hours instead of daily, with a weekly ceiling replacing the old monthly one. If a request would exceed the remaining budget, Notebook will suggest a lighter-weight alternative output, or let users queue heavy jobs — like Video Overviews or Slide Decks — with a "Generate Later" option that completes automatically once the quota resets. Google AI Plus subscribers get roughly double the standard allowance.