Capital & Industrial Strategy

Anthropic's annualized revenue run rate crossed $65 billion in July, up from roughly $9 billion at end-2025, according to documents seen by Bloomberg, TechCrunch, and CNBC. The company added $18 billion in annualized revenue in just two months, with Q2 quarterly revenue exceeding $11.5 billion — 14 times the year-ago figure. The consistency of the reporting across multiple outlets citing independent sourcing makes these figures high-confidence, though they represent run-rate projections rather than audited results.

The strategic read here is twofold. First, Anthropic is clearly converting its enterprise API and Claude product distribution into durable recurring revenue at a pace that validates the frontier model business at scale, countering narratives that commoditisation would compress margins before revenue matured. Second, the IPO signalling embedded in these disclosures — sharing figures with prospective investors — suggests Anthropic is actively managing its pre-IPO narrative. At $65 billion annualized, a public offering could command a valuation north of $200 billion at current AI multiples, which would reprice the entire sector. The WSJ separately reported unusual options activity tied to Anthropic stake sale whispers, suggesting institutional money is already positioning ahead of any liquidity event.

Nvidia has agreed to provide up to $105 billion to backstop OpenAI's Ohio data center campus, to be developed by SoftBank's SB Energy, and has separately invested $1.5 billion directly into that development entity, according to Bloomberg, Reuters, the FT, the WSJ, and Semafor. The deal is described as signed and agreed, not merely announced as an intention, though the $105 billion figure represents a maximum backstop commitment rather than upfront capital deployed.

The strategic logic is transparent and significant: Nvidia is guaranteeing that its chips power this facility, effectively using its balance sheet as a demand-creation instrument. By underwriting the construction risk, Nvidia removes the financing uncertainty that has slowed hyperscale data center commitments, while locking OpenAI into Nvidia silicon for a generation of infrastructure. This is vertical integration achieved not through acquisition but through capital commitment — a model that smaller chip competitors cannot replicate. Combined with the $1.5 billion SB Energy investment, Nvidia is now a direct financial stakeholder in the infrastructure layer it supplies, aligning its incentives with continued hyperscale GPU deployment at a scale that dwarfs any single hyperscaler's individual capex commitment.

Stripe has agreed to acquire OpenRouter — a platform that enables developers to access and dynamically route queries across hundreds of AI models — for more than $7 billion, according to Bloomberg. The deal is described as agreed by sources familiar with the matter; regulatory approval status was not specified in available reporting. OpenRouter sits at the infrastructure layer between application developers and model providers, optimising for cost, latency, and capability across a fragmented model landscape.

Stripe's strategic intent here is not payments-adjacent AI — it is a direct move to own the enterprise AI procurement and routing layer. As enterprise AI spending matures, CFOs and engineering teams are increasingly focused on model cost optimisation rather than raw capability. A platform that intelligently routes workloads to the cheapest effective model becomes a critical savings lever, and whoever controls that routing layer gains visibility into enterprise AI consumption patterns that is commercially valuable in its own right. For Stripe, which already processes payments for a large share of the developer economy, adding OpenRouter creates a bundled offering — pay for AI, route AI, optimise AI spend — that deepens lock-in with its developer customer base and positions Stripe as an AI infrastructure company rather than a payments processor.

Groq's $350 million raise at a $3.5 billion valuation, reported by TechCrunch, marks a strategic retreat from proprietary chip development toward Nvidia-powered cloud services. The company is now building out data center capacity using Nvidia hardware — the exact ecosystem it was designed to disrupt. This is a significant signal: the custom silicon path requires capital and time horizons that most startups cannot sustain against Nvidia's entrenched software stack, customer relationships, and now its direct infrastructure financing capabilities.

Simultaneously, Bloomberg and CNBC report that Chinese models — DeepSeek, Alibaba's Qwen, and Moonshot — are approaching parity with leading US platforms on benchmark performance while undercutting on price and offering greater deployment flexibility. Alibaba's release of Qwen's open weights, explicitly framed as a challenge to Meta's Llama, escalates the open-weight model arms race. For US AI companies, the threat is not just geopolitical but commercial: cheaper, near-equivalent Chinese models available through grey-market channels or in jurisdictions without US export controls directly compete with paid API access to US frontier models.

The UK government has launched an urgent economic impact assessment after the Trump administration blocked foreign nationals from accessing Anthropic's Fable 5, according to the FT. The assessment reflects a belated recognition that dependency on US frontier AI models creates sovereign economic risk — access to frontier capability is now effectively a matter of industrial policy, not just technology procurement. The UK's position is structurally weak: it has no domestic frontier model and its cloud infrastructure is dominated by US hyperscalers.

South Korea's sovereign AI competition, reported by Bloomberg, has narrowed to three competitors after eliminating a team led by Motif Technologies, signalling that Seoul is applying genuine selection pressure rather than distributing funding broadly. The contrast between the UK's reactive scramble and South Korea's structured competition illustrates two different government postures: one managing dependency after the fact, the other attempting to build indigenous capability proactively. Alphabet's debut Australian dollar bond issuance, flagged by Bloomberg, adds another dimension — US hyperscalers are now tapping non-dollar debt markets to fund AI infrastructure expansion, effectively exporting the financing burden of AI build-out to foreign capital markets.