This Week’s Top Five Stories in AI

The latest:

Anthropic could float at $2tn or more when it lists in October, according to the Financial Times, which said it spoke to half a dozen of the company’s backers. That would make it the second record-breaking debut of the year, after SpaceX listed at roughly $1.77tn. Anthropic leads most frontier-model benchmarks, and the timing lands as warnings of an AI financial bubble intensify.

Details:

• The valuation: Investors expect Anthropic to float in October at $2tn or above, the Financial Times reported, citing half a dozen backers. No pricing range, bank syndicate or filing date has been made public, and the company itself has not confirmed a listing timetable.
• The benchmark: SpaceX’s debut valued the aerospace firm at about $1.77tn, displacing Aramco’s 2019 listing at $1.7tn as the largest on record. An Anthropic float at $2tn would push the record higher again within the same year.
• NVIDIA’s financing push: NVIDIA has signed memorandums of understanding with KKR, Goldman Sachs, Brookfield, Blackstone, BlackRock and Apollo to build autonomous compute financing platforms, an initiative aimed at mobilising more than $500bn in third-party capital for AI infrastructure deployment.
• How it works: The structures let the financial institutions independently underwrite, structure and syndicate pools of capital, giving NVIDIA’s customers access to competitive debt and hybrid financing rates. The memorandums are non-binding instruments, and no individual commitments per institution were disclosed.
• New models: xAI released Grok 4.6, building on Grok 4.5 with a focus on long-running agents and more ambitious interactive and visual work. Google DeepMind released Gemini 3.7 Flash, billed as its most intelligent workhorse model yet for coding and agents.
• The competitive squeeze: Both firms are chasing Anthropic and OpenAI, which lead across most benchmarks, while competing on price with Chinese developers such as Moonshot’s Kimi K3 for enterprise customers. Pricing details for the two new models were not announced.
• Meta goes local: Meta Superintelligence Labs launched Muse Glimmer, a 30-billion-parameter open-weight model engineered to run complex multi-step agentic workflows offline on a single consumer GPU on a Mac or PC, released under a permissive Apache 2.0 licence.
• The policy ask: The rollout coincided with Meta CEO Mark Zuckerberg urging US policymakers to lower regulatory barriers for domestic open-source models so they can compete with Chinese rivals. No specific rules or legislation were named in his call.
• OpenAI departures: Head of Ethics Chloé Bakalar left OpenAI last month, less than a year after joining, the Financial Times reported first. Safety Systems head Johannes Heidecke and Chief Futurist Joshua Achiam have also departed recently. OpenAI has not stated a reason for the exits.
• Her position: Speaking to IE Insights in March, Bakalar argued nobody at the forefront expected the technology to expand as fast as it did, and that decisions on global systems affecting billions “shouldn’t simply be one person who has a multi-billion-dollar company” setting the terms.

Background:

Business Insider reported that OpenAI has reorganised its safety, product and research teams numerous times since ChatGPT launched in 2022, as the company grew from a small research lab into a large technology firm.

Between the lines:

The week’s items point in one direction: capital is moving faster than governance. A $2tn listing and a $500bn compute financing effort scale up the money, while OpenAI loses its ethics, safety-systems and futurist leads. Meta’s open-weight push and Zuckerberg’s deregulation appeal frame Chinese competition as the argument for fewer constraints, not more.

What’s next

Watch for an Anthropic filing and pricing range ahead of the expected October listing, whether NVIDIA’s memorandums convert into binding commitments, and whether OpenAI names replacements for its departed ethics and safety leads.