Capital & Industrial Strategy
Top Line
Anthropic is in talks to acquire AI startup Decart for approximately $6 billion — a move that, if completed, would represent one of the largest acqui-hires in AI history and signals Anthropic's intent to compete on real-time simulation and world-model capabilities ahead of a widely anticipated IPO at a potential $2 trillion valuation.
CoreWeave's Q2 revenue doubled year-on-year, driving a 19% single-day stock surge and reigniting the AI infrastructure trade — but the company's own disclosures reveal dangerous single-vendor concentration risk on Nvidia hardware, a structural vulnerability now flagged to investors.
AI coding and developer-tooling valuations are detaching from conventional multiples: Cognition is reportedly in talks to raise at $40 billion just months after a $26 billion round, Lovable hit $13 billion, and CodeRabbit closed at $1.5 billion — compressing the time between funding rounds to weeks rather than years.
Tencent's Q2 results showed AI spending more than doubling while profit growth stalled, underscoring a China-wide dynamic where hyperscalers are prioritising market position over margins — a pattern that is beginning to alarm equity investors even as revenue beats estimates.
Enterprise AI adoption is bifurcating sharply: OpenAI's customers are migrating from chatbot to agentic deployments at scale, while a Reuters poll confirms the majority of Japanese firms and a growing number of Western CIOs are actively throttling AI spend as cost discipline reasserts itself.
Key Developments
Anthropic's Decart Acquisition Talks: Capability Land-Grab Before IPO
Anthropic is in active talks to acquire Israeli-founded AI startup Decart AI for approximately $6 billion, according to Bloomberg and confirmed by a separate source cited by Reuters. This is an announced intention, not a closed deal, and remains subject to finalisation. Decart is known for real-time AI simulation work, including high-speed generative world models — capabilities that sit adjacent to, but distinct from, Anthropic's core Claude language model franchise. The strategic logic is clear: as the frontier model race increasingly demands not just language reasoning but grounded, real-time environmental modelling, acquiring Decart would expand Anthropic's technical surface area ahead of what the Financial Times reports are investor expectations of a record-breaking IPO targeting a $2 trillion valuation.
The $6 billion price tag is notable relative to Decart's stage and revenue, suggesting Anthropic is paying a significant premium for the team and IP rather than current cash flows. This acquisition posture — buying forward capability before public markets constrain capital allocation — mirrors the strategic logic seen in Google's DeepMind consolidation. If Anthropic's IPO proceeds on the timeline implied by FT reporting, this may be the last major M&A move the company can execute with the speed and valuation flexibility of a private company.
AI Developer Tooling: Valuation Velocity Is Accelerating, Not Decelerating
The AI coding and developer tooling segment is generating funding rounds at a pace and valuation multiple that has no recent precedent. Cognition — maker of the Devin autonomous coding agent — is reportedly in talks to raise at a $40 billion valuation according to TechCrunch, just months after closing $1 billion at $26 billion. This is unconfirmed and still in negotiation. Separately, vibe-coding startup Lovable closed a round valuing it at $13 billion per WSJ, making it one of Europe's most valuable startups. CodeRabbit, an AI code review platform, closed at $1.5 billion per Reuters, and Blacksmith — focused on AI-powered code testing — saw its valuation jump nearly 10x in under a year on tenfold revenue growth per TechCrunch.
The common thread across these deals is the software development lifecycle as the primary enterprise ROI vector for AI. Enterprises spending on AI are overwhelmingly starting with developer productivity — it offers measurable output, contained blast radius, and referenceable ROI. This is driving a concentration of venture capital into the coding layer. The Thrive Holdings $2 billion raise at $12 billion, backed by SoftBank, D1, and Altimeter per TechCrunch, extends this logic to broader enterprise AI deployment with OpenAI backing, suggesting a distribution play layered on top of the tooling bets.
AI Infrastructure Trade Reignites — With Structural Risks Visible
CoreWeave's Q2 earnings — revenue doubling year-on-year — triggered a 19% single-session stock move and a broader AI infrastructure rally, with Nebius surging 34%, Super Micro rallying, and South Korean semiconductor giants SK Hynix and Samsung lifting the Kospi into bull market territory per CNBC and Reuters. Foxconn reported a 35% Q2 profit rise on AI demand per Reuters, and ASUS raised its AI server growth target citing sustained demand and memory shortages per Bloomberg.
However, Cisco's guidance of $7.5 billion in AI data centre revenue disappointed investors who had extrapolated from $9.3 billion in accumulated orders — implying either order-to-revenue conversion friction or margin pressure on fulfilment per Bloomberg. More structurally, CoreWeave's own investor disclosures warned of significant operational and financial difficulty if it were ever required to shift away from Nvidia chips per Bloomberg. Private capital firms are simultaneously wagering that Nvidia hardware will hold collateral value for years, per FT — a bet that is only rational if Nvidia's architectural dominance is durable. China's YMTC surpassing Kioxia in NAND shipments for the first time per Bloomberg adds a supply chain dimension: Chinese memory is gaining share under conditions of AI-driven shortage, complicating the hardware dependency picture.
Enterprise Adoption Bifurcation: Agents Versus Pilots, Cost Discipline Versus Spend
Two conflicting signals are emerging from enterprise AI deployment data. On the demand side, OpenAI's enterprise customers are demonstrably shifting from chatbot deployments to autonomous agentic workflows that execute tasks, according to Semafor — a qualitative leap that implies deeper integration, higher switching costs, and larger contract values. AT&T has publicly articulated a strategic commitment to open-weight models specifically to control token costs and protect proprietary data per WSJ, representing a coherent enterprise architecture decision rather than a pilot.
Simultaneously, Fortune reports that CIOs and CTOs who spent years building AI business cases are now imposing active usage caps as costs overrun projections, and a Reuters poll confirms that the strong majority of Japanese firms have not yet meaningfully deployed AI per Reuters. The divergence between early adopters scaling into agents and the majority still in evaluation or throttling spend is the defining enterprise dynamic of the current period — and it means AI revenue is concentrating in a smaller number of large accounts rather than broadening across the enterprise base.
China's AI Capital Intensity Mounts as Hyperscalers Absorb Margin Pressure
Tencent's Q2 results crystallise the strategic bind facing Chinese AI hyperscalers. Revenue grew 11% on AI-driven advertising gains, but profit fell short of estimates after AI spending more than doubled, per Reuters and Bloomberg. Tencent's stock is down 26% year-to-date as investors price the risk that AI investment cycles in China are longer and more capital-intensive than initially modelled, particularly given US chip export restrictions forcing reliance on less efficient domestic alternatives. Management defended the spend as necessary to avoid falling behind Alibaba and to position for what it characterised as superior long-term returns — a familiar refrain that equity markets are treating with increasing scepticism.
The competitive dynamic in China's AI market is structurally different from the US: firms cannot rely on the same foundation model infrastructure, face regulatory constraints on data use, and compete in a domestic market where DeepSeek's open-weight releases have commoditised baseline capability. The result is that Chinese hyperscalers must spend heavily on proprietary infrastructure while simultaneously defending against margin compression from open-model competition — a dual squeeze that has no clear resolution on the current earnings horizon.
Signals & Trends
Open-Weight AI Is Becoming an Enterprise Infrastructure Default, Not a Cost-Cutting Fallback
AT&T's articulated strategy of building on open-weight models to control token economics and protect proprietary data — reported by WSJ — is a signal that large enterprises are making durable architectural commitments to open models rather than treating them as temporary substitutes for proprietary APIs. The White House's reported consideration of open models in its updated AI policy framework per Wired, combined with Meta and Nvidia's explicit push into the open-weight race per CNBC, suggests a convergence between enterprise procurement logic, US industrial strategy, and platform competition dynamics. The strategic implication for closed-model providers like OpenAI and Anthropic is that the enterprise contract value they currently command is under structural pressure from below — not from a single competitor, but from an ecosystem effect.
Private Capital Is Becoming Structurally Embedded in AI Hardware Finance — With Untested Risk
The FT reports that private capital firms are making large directional bets on Nvidia GPU hardware retaining collateral value over multi-year horizons, underwriting financing structures that depend on hardware depreciation curves behaving like commercial real estate rather than consumer electronics. This is being done at scale to fund the neocloud buildout. CoreWeave's own warning about the difficulty of migrating away from Nvidia chips — disclosed formally to investors per Bloomberg — validates the asset-specificity assumption embedded in these financing bets. But the scenario that breaks this model — an architectural shift to alternative silicon, a demand air pocket, or a change in hyperscaler procurement patterns — has not yet been stress-tested by a real credit event. The financial system is building concentrated exposure to a single hardware vendor's product roadmap in a way that has not previously been formalised or regulated.
AI Valuation Rounds Are Compressing Into Months, Not Years — Flagging Reflexivity Risk
The pattern emerging across Cognition, Legora, and Lovable is that funding rounds are following one another within four to six months at step-changes of 50-100% in valuation, per TechCrunch and FT. The Legora case is instructive: a $5.6 billion valuation four months ago is now being replaced by a fundraise targeting over $10 billion. This compression is not driven by revenue multiples converging to a new norm — it is driven by a combination of strategic investor FOMO, the benchmarking of AI company valuations against Anthropic's IPO trajectory, and the wealth effect among AI equity holders now attracting private banking capital at scale per FT. When valuation rounds are driven partly by the IPO expectations of a single anchor company rather than independent fundamental analysis, the risk of correlated repricing across the segment is elevated.
Explore Other Categories
Read detailed analysis in other strategic domains