Capital & Industrial Strategy

OpenAI is in early-stage talks to raise at least $30 billion at a pre-money valuation of approximately $1.4 trillion, according to Bloomberg and The Information. Critical qualifier: no term sheet has been signed. TechCrunch characterises this as the anticipated final pre-IPO round ahead of a delayed 2027 debut. Sam Altman confirmed at OpenAI DevDay that the company will go public 'at some point' but is being patient, citing safety concerns as a factor in the delay.

The commercial backdrop justifies investor appetite: OpenAI's annualized recurring revenue is approaching $70 billion, up roughly 70% since the start of Q3, driven by aggressive enterprise sales and price cuts on its API, per The Information and Reuters. At DevDay, OpenAI also launched Dots — a premium personal AI agent competing directly with Meta's Muse — and announced features positioning ChatGPT as an alternative app distribution layer, signalling a platform land-grab strategy that extends well beyond model licensing.

Anthropic's confidential IPO prospectus, reported exclusively by Reuters, reveals the company has committed to pay SpaceX up to $84.5 billion for Nvidia-based compute capacity through 2029 — agreements that are largely cancellable on 90 days' notice, per The Information. The total planned AI buildout disclosed is $518 billion, much of which depends on deals with Big Tech partners including Amazon and Google — the same hyperscalers in which Anthropic has taken strategic investment. The prospectus discloses steep ongoing losses alongside rapid revenue growth, and flags 'catastrophic or existential risks to humanity' from its own technology.

The IPO timing is under pressure. The Information and The Wall Street Journal both flag a broad IPO market stall: Oura's postponement citing 'uncertainty in the IPO market,' followed by Amaero, Holtec Nuclear, and SB Energy all pausing listings. The FT notes Anthropic faces a 'narrowing window' to list before year-end. The prospectus's explicit risk disclosures — unusual in their candour — create a dual challenge: they satisfy regulatory expectations but may amplify investor hesitation about terminal risk scenarios.

AMD's $8.2 billion all-stock acquisition of World Labs — Fei-Fei Li's two-year-old spatial intelligence startup — confirms a pattern The Information describes as defining 2026 venture exits: M&A is generating returns while the public markets stall. The deal joins Nvidia's acquisition of Hugging Face, SpaceX's purchase of Cursor, and Stripe's acquisition of OpenRouter as major strategic consolidation moves. For AMD, the strategic logic is clear — World Labs brings foundational spatial and visual AI capabilities that can differentiate AMD's accelerator stack against Nvidia in emerging agentic and robotics workloads, not just data centre LLM inference.

The consolidation dynamic also extends to infrastructure financing. Samsung and five affiliates are investing $1 billion in KKR-backed Helix Digital Infrastructure, per The Information, while PaleBlueDot AI is seeking $600 million in private credit from lenders including Brookfield to purchase chips for a South Korean facility, per Bloomberg. Private credit is increasingly filling the gap between equity raises, signalling that infrastructure financing has matured into a distinct and growing asset class within the AI capital stack.

The White House 'Accord on Super Intelligence,' announced Tuesday, commits leading AI companies to non-binding internal controls on model alignment and external auditing, according to The Information. Trump described it as 'morally binding' and floated a 10-person industry oversight committee, but Politico and CNBC confirm the accord carries no legal force. House Speaker Johnson separately stated guardrails should remain 'voluntary,' signalling Congressional alignment with the executive posture. The WSJ reports Trump defended the approach as relying on existing DoJ authorities.

The accord's industrial policy dimension is significant: participants also backed data centre expansion commitments. The AI Infrastructure Coalition — members include Google, Meta, and Microsoft — simultaneously unveiled pledges to cover energy costs and minimise water usage for data centres, per the FT, framed explicitly as pre-emptive action to blunt local backlash ahead of US midterms. The combined picture is a regulatory environment where incumbents shape the rules, capacity expansion is politically endorsed, and legislative intervention is actively deferred.

DeepSeek has publicly released software developed jointly with Huawei to programme AI chips, per Bloomberg. The release is strategically significant beyond its technical content: it represents the first publicly available, production-tested software toolchain designed to make Huawei's Ascend chips a viable substitute for Nvidia's H-series accelerators in frontier AI training and inference workloads. CUDA's ecosystem lock-in has been Nvidia's most durable competitive moat; a credible open-source alternative, built by China's leading LLM laboratory on China's leading domestic chip, directly targets that moat. Chinese hardware stocks are simultaneously heading for their worst quarter, per Bloomberg, following a July selloff driven by concerns about AI firms' ability to justify stretched valuations — suggesting equity markets are not yet pricing in the DeepSeek-Huawei stack as a commercial inflection.