Capital & Industrial Strategy

Sam Altman told OpenAI staff the company is open to slowing development of cutting-edge AI and wants peers to do the same, according to Bloomberg. The statement is notable precisely because OpenAI is the market leader with the most to gain from a pause that freezes the competitive landscape in its favour — a dynamic that cuts directly against the framing of this as a safety measure. Wired reports that AI leaders are themselves questioning whether any coordinated slowdown would survive antitrust scrutiny, given that agreements among competitors to limit output or competition are among the most scrutinised categories under US antitrust law.

For investors, the signal has two readings. The optimistic interpretation is that OpenAI is genuinely concerned about safety-capability misalignment at the frontier, and a managed slowdown would reduce systemic risk. The strategic interpretation is that a freeze benefits entrenched incumbents with large model advantages, while disadvantaging capital-efficient challengers like DeepSeek and Mistral who have competed on doing more with less. The antitrust question also creates a material IPO risk disclosure issue — Semafor notes that existential risk framings are now appearing in prospectus risk factor discussions for both OpenAI and Anthropic.

Blackstone's compute financing vehicle, built around a May joint venture with Google to acquire $5 billion worth of TPUs for resale to AI firms, is now expected to purchase substantially more chips than initially disclosed, according to The Information. Separately, the Pentagon is in active negotiations to provide a $5 billion loan to Fluidstack, an AI cloud infrastructure startup advised by Anduril founder Palmer Luckey's bank, per The Wall Street Journal. Note that both the expanded Blackstone deal and the Pentagon loan are reported as active negotiations, not closed transactions.

The structural significance here is the shift in who is financing AI infrastructure. Traditional equity and venture capital are being supplemented — and in some cases replaced — by structured finance instruments, government lending facilities, and asset-backed deals secured against compute contracts. The Information reports that OpenAI, Anthropic, and neoclouds including Nscale are actively building out capital markets and structured finance teams to handle deal volumes that don't fit standard corporate financing templates. This institutionalisation of AI financing means the capital stack for AI infrastructure is beginning to resemble project finance or real asset financing more than traditional tech venture rounds — with commensurate implications for return profiles, leverage, and who ultimately controls capacity.

Oracle reported cloud sales that more than doubled analyst estimates, with 30% top-line growth for the quarter ending August 31, driven by AI data center demand. Critically, The Information notes Oracle is managing capex discipline by requiring customers to pay upfront — effectively using customer prepayments to fund expansion rather than balance sheet leverage, a structurally advantaged model compared with peers burning cash on speculative capacity. RBC's Rishi Jaluria characterised the AI trade as more durable than markets are currently pricing, per Bloomberg.

Adobe's picture is the inverse. Its Q3 revenue beat estimates, but a forward guidance miss renewed investor concern that generative AI competitors are eroding its core creative software moat, per Bloomberg and The Information. Adobe is now breaking out AI product revenue separately, suggesting it is trying to reframe its story around AI monetisation — but the market's 2% post-earnings drop indicates scepticism about whether AI product revenues will replace what generative AI upstarts are taking from its existing customer base. The Oracle-Adobe divergence is a clean empirical illustration of the infrastructure-versus-application layer dynamic: compute infrastructure is capturing AI spend directly, while application-layer incumbents face substitution risk.

Anthropic has published a detailed report alleging that Alibaba, Moonshot AI, and DeepSeek conducted systematic 'distillation' campaigns against Claude — using thousands of fake accounts and millions of real user queries to extract model knowledge and train competing systems, per TechCrunch, The Wall Street Journal, and CNBC. The decision to publish a named report, rather than pursue quiet legal action, suggests Anthropic is seeking to establish a public record — likely as groundwork for regulatory action, export control advocacy, or both. Moonshot is simultaneously reported by Semafor to be eyeing a Hong Kong-Shanghai dual listing, creating an awkward juxtaposition between its capital-raising ambitions and its legal exposure in the US.

The strategic implication for capital allocators is that the competitive moat of frontier US AI models is being systematically tested via distillation rather than direct R&D competition — a capital-efficient attack vector that compressed the gap between US and Chinese model capabilities faster than raw investment differentials would predict. This makes IP protection, usage monitoring, and access controls increasingly material factors in valuing US AI lab assets.

Jensen Huang has guided for approximately 70% revenue growth in the coming year, with TechCrunch reporting he explicitly addressed concerns about circularity in Nvidia's ecosystem deals — a reference to the perception that Nvidia's investments in AI companies create customers who then buy Nvidia chips. Simultaneously, the DOJ is investigating Nvidia's $20 billion deal to license chip startup Groq's technology and absorb most of its employees, which the companies described as a licensing arrangement rather than an acquisition to avoid standard merger review thresholds, per The Information. The Groq investigation is confirmed as active; its outcome is not yet determined.

Nvidia is also deepening its partnership with inference chip startup d-Matrix through NVLink Fusion integration, per The Information, extending its networking stack into third-party inference hardware. Separately, Positron — an inference-focused chip startup — closed an $875 million round at a $5 billion valuation led by NEA, Atreides, and Jim Clark, per The Wall Street Journal. China's Enflame, backed by Tencent, tripled on its Shanghai debut, signalling strong domestic investor appetite for AI chip plays in China's parallel stack. The cumulative picture is a chip layer in active consolidation, with Nvidia extending control through partnerships and licensing deals that the DOJ is now scrutinising for structural market foreclosure.