This week in AI: $302B across 12 deals

What happened: Anthropic's public S-1 was posted on SEC EDGAR on October 26, listing the company under ANT on the Nasdaq Global Select Market.

• Reuters describes investor talk of a valuation near $1.5 trillion.
• Financial Times and Bloomberg report third-quarter revenue of $16.8 billion.
• The SEC must still declare the registration statement effective before pricing, and Anthropic closed a $65 billion Series H at a $965 billion post-money valuation on May 28.

Why it matters: A public listing will set a market price for a leading model lab and a reference point for AI valuations. Teams that depend on Claude should watch for pricing or capacity changes during the offering period.

What happened: DeepSeek's Harness team put dsh-superagent on GitHub under an MIT license on October 31, labeled a research prototype and not an official product.

• The agent watches one GitHub repository per installation and can also listen to Feishu, the workplace chat app from ByteDance.
• Approved requests open a task in its own session, often ending in a pull request, and auto-merge stays off unless a prompt names another account.
• Setup requires Node.js 22.19 or later and a dedicated GitHub account, and DeepSeek suggests a separate OS account because sessions run with full access.

Why it matters: Bloomberg reports the release targets Anthropic's Claude Code, so a free agent like this puts pressure on paid coding tools. Founders should test it in a sandbox before giving it access to any real repository.

What happened: On October 30, OpenAI published a system card addendum for GPT-Rosalind-6, a reasoning model for biology, drug discovery and translational medicine, offered only after review.

• OpenAI says the model runs on the same weights as GPT-6 Astra, with no added training data and no extra post-training for this release.
• Approved organizations must restrict use to authorized people, apply least-privilege and insider-risk controls, and require qualified human review of outputs.
• According to Forkast, the model is available in Work and Codex, API use needs separate approval, and it is not in the Chat tab.

Why it matters: Gated access shows how sensitive capabilities may be sold, with review and governance attached to the product. Founders in regulated fields should expect similar requirements and check whether their own products need access controls.

What happened: Nvidia and AMD will put a combined $4 billion into Meta while the company raises $97.5 billion to build out its AI infrastructure.

• Coverage from several outlets places the announcement around October 27, so it predates this run.
• Meta launched its Muse Wonder frontier models in mid-October, and a raise of this size signals that its compute plans are still expanding.
• Vendor investment ties supply commitments to the investor's own revenue, which is worth noting when reading the stakes.

Why it matters: Frontier compute is being financed through equity and supplier relationships, not only purchase orders. Startups that rely on GPU capacity should watch whether large buyers lock up allocations, since that can tighten supply and pricing for smaller customers.

What happened: State Senator Andrew Gounardes released RAISE Act 2.0 on October 29, an amendment to the RAISE Act he sponsored with Assemblymember Alex Bores.

• According to PYMNTS, large developers would have to keep a state-licensed frontier examiner who audits compliance with their safety frameworks each year.
• Broader evaluations and catastrophic risk assessments would run every 90 days.
• The attorney general could ask a court to stop hazardous activity, including limits on where a model is deployed in New York.

Why it matters: If passed, large labs and their customers will face licensed outside review in New York. Founders building frontier-scale models should track the compute threshold, which one analysis reads as moving from 10^26 to 10^25 FLOPs.