Capital & Industrial Strategy

Stripe has agreed to acquire OpenRouter for a price reported variously as $7 billion by the Wall Street Journal and $8 billion by the Financial Times, making it Stripe's largest acquisition to date. OpenRouter is a model-routing marketplace that allows developers to query multiple AI models through a single API, abstracting away vendor lock-in. The deal is confirmed and announced, though regulatory review status was not specified in available reporting.

Stripe's stated rationale invoked 'the singularity,' which TechCrunch correctly identifies as window dressing. The real strategic logic is that as enterprises run workloads across heterogeneous model providers — OpenAI, Anthropic, Google, open-source — the billing, metering, and settlement layer becomes a high-value chokepoint. Stripe is positioning to own that layer. OpenRouter's founder background in NFT infrastructure is notable primarily because it demonstrates how quickly founders can pivot platforms into critical B2B infrastructure when timing aligns with a market wave.

Google has finalised an expanded partnership with Marvell Technology valued at approximately $12 billion, which includes an option for Google to acquire up to $12.2 billion in Marvell shares, according to the Financial Times and CNBC. Marvell's stock rose 10% on the announcement. The deal centres on custom AI chip development, extending Google's established Tensor Processing Unit programme with a strategic manufacturing and design partner.

The equity option embedded in the deal is the structurally interesting element: it gives Google a path to partial ownership of a critical supplier without a full acquisition, preserving Marvell's independence and customer relationships while securing preferential access. This mirrors the logic behind Microsoft's OpenAI structure. For Marvell, the deal provides revenue certainty at a scale that justifies substantial R&D commitment to Google-specific silicon. The broader implication is that hyperscalers are no longer merely buying chips — they are co-developing and partially capitalising their chip supply chains to structurally reduce dependence on Nvidia, which CNBC notes is a shared priority across Google and its competitors.

Fractile, which develops inference-optimised AI chips and has a confirmed supply agreement with Anthropic, is in advanced talks to raise capital at a $6.5 billion valuation, according to Bloomberg. This represents more than a sixfold increase from the valuation it achieved in May. The round has not closed; the valuation reflects advanced negotiations rather than committed capital.

The Anthropic supply deal is doing substantial work here. In the current environment, a contractual commitment from a frontier lab — especially one whose annualised revenue has reportedly surged to $65 billion — functions as a de facto revenue guarantee that anchors venture pricing at levels disconnected from traditional metrics. This creates a two-tier chip startup market: those with anchor contracts from Anthropic, OpenAI, or Google commanding premium valuations, and everyone else competing on speculative future demand. The dynamic accelerates consolidation around a small number of deep relationships between model labs and their preferred hardware partners.

OpenAI CFO Sarah Friar told employees at an all-hands meeting that the company will be public in 2027 or sooner, according to CNBC. However, Semafor reports that OpenAI's sales growth has slowed in a way that is unsettling pre-IPO investors — while Anthropic's annualised revenue has exceeded $65 billion, up more than sevenfold since the end of last year. These two data points in combination are the significant signal: a planned 2027 IPO from a company whose primary competitor has dramatically outgrown it on revenue trajectory.

The enterprise privacy competition between the two firms adds further texture. OpenAI has launched new customer data protections explicitly designed to differentiate from Anthropic's data retention practices, per TechCrunch and the Wall Street Journal. This is a defensive enterprise sales move, not a product innovation — it suggests OpenAI is losing the trust argument in certain procurement conversations and is responding with contractual rather than technical differentiation. For IPO underwriters, the combination of decelerating growth, a more aggressive competitor, and a company still burning capital at scale at a planned 2027 listing is a challenging story to price.

Bloomberg sources report that SpaceX approached Cognition AI, the maker of the Devin AI coding agent, about a potential acquisition — which would have followed SpaceX's earlier acquisition of Cursor, a popular AI coding assistant. Cognition's CEO publicly denied the report, creating a direct conflict between executive statement and sourced reporting. The denial does not resolve the underlying signal: SpaceX is actively building out an enterprise AI coding capability stack, and Cognition was apparently on the target list regardless of whether discussions advanced.

The strategic logic is straightforward. SpaceX operates one of the most demanding software engineering environments in aerospace, and AI coding tools translate directly into engineering productivity. But the acquisition pattern also suggests SpaceX is positioning AI coding capability as a proprietary engineering advantage rather than purchasing off-the-shelf solutions — a preference for internalising tools that give competitors no access. Combined with SoftBank's $200 million investment in Gravis Robotics for autonomous construction machinery, the session underscores that physical-world, capital-intensive industries with specific engineering constraints are now active buyers or investment targets for domain-specific AI, not merely enterprise software recipients.