Capital & Industrial Strategy
Top Line
CoreWeave has completed a $4.2 billion convertible bond offering — upsized from an initial $3 billion target — confirming robust institutional appetite for pure-play AI infrastructure debt even as bond buyers signal growing selectivity on the broader wave of AI paper.
Anthropic is in early talks to lease up to 1 gigawatt of compute from Apollo-owned Stream Data Centers, a strategic move to reduce cloud-provider dependency and lock in dedicated TPU capacity — the clearest signal yet that frontier AI labs are pursuing vertical integration of infrastructure.
OpenAI and Anthropic have simultaneously released cheaper, higher-performing models — GPT-6 Sol/Luna and Claude Opus 5.5 respectively — intensifying a price war that is visibly cannibalising traditional software budgets, with AI providers now accounting for 8% of enterprise software spend versus 1.4% a year ago.
SoftBank has launched one of the largest-ever junk bond sales to fund its AI investment push, while Alibaba unveiled a new AI chip and announced data centre expansion across Europe and the Middle East, together illustrating how Asian capital is aggressively financing the AI infrastructure buildout.
Multiple AI-related IPOs are being postponed as investor caution grows around concentrated or smaller bets, even as Nscale tests Wall Street appetite and Accelevation targets a $5.4 billion valuation — creating a bifurcated public market where tier-one names clear and secondary plays stall.
Key Developments
Anthropic's Infrastructure Play: Leasing Direct Compute to Break Cloud Dependence
Anthropic is in early-stage negotiations to lease up to 1 gigawatt of capacity from Stream Data Centers, a company majority-owned by Apollo Global Management, according to The Information. The facilities would be provisioned with tensor processing units designed by Broadcom and Google — not Nvidia GPUs — a significant architectural choice that signals Anthropic's intent to diversify its silicon supply chain while deepening its relationship with Google, which remains a key investor and TPU developer. These are early talks, not a signed agreement, and terms remain unconfirmed.
The strategic logic is clear: frontier AI labs that remain entirely dependent on hyperscaler cloud procurement are exposed to pricing leverage, capacity constraints, and competitive disadvantage, given that Microsoft, Google, and Amazon are simultaneously their landlords and rivals. Securing direct tenancy in third-party data centres — funded in part through private credit from firms like Apollo — represents a new financing model for AI compute that bypasses both equity dilution and hyperscaler dependency. This mirrors the path CoreWeave took earlier, and it positions Anthropic to control its cost structure ahead of a potential IPO, which the WSJ notes has already been delayed.
The AI Model Price War Accelerates — and It Is Eating Traditional Software Budgets
OpenAI launched GPT-6 Sol and GPT-6 Luna — cheaper, faster variants within the GPT-6 Astra family — on the same day Anthropic released Claude Opus 5.5 at 40% lower cost than predecessor models, with performance claims matching or exceeding Fable 5.1 and Mythos 5.1, according to TechCrunch and The Information. The simultaneous releases are not coincidental — both labs are competing for enterprise API wallet share at a moment when procurement data shows AI spending surging.
Data from procurement software firm Zip, reported by The Information, shows AI providers including OpenAI, Anthropic, Cursor, and Sierra now account for 8% of enterprise software spend — up from 1.4% in the prior 12-month period. This is direct budget displacement, not incremental spend. In response, incumbents including Amazon, Microsoft, Figma, and Workday are discounting their own AI products to prevent churn, while Microsoft has separately authorised 30–50% discounts on Copilot subscriptions for large seat commitments, per The Information. The price compression dynamic is now structural: frontier labs cut prices to capture enterprise share, incumbents respond with discounts, and the entire software pricing stack deflates.
China's AI Infrastructure Surge: Alibaba's Chip and Data Centre Offensive
Alibaba used its annual Apsara conference to unveil the Zhenwu V900, an AI training chip it claims delivers three times the performance of its predecessor, and announced accelerated data centre expansion across Europe and the Middle East as part of a 20-gigawatt global infrastructure buildout, according to The Information and the Wall Street Journal. The Zhenwu V900 announcement is the most direct evidence yet that Chinese hyperscalers are closing the semiconductor gap created by US export controls, with domestic chip performance now credibly competing on training workloads.
The BBC's reporting from Inner Mongolia — where dozens of data centres are under construction at what workers describe as 'China speed' — contextualises these announcements within a state-backed infrastructure mobilisation that is qualitatively different from commercial buildout cycles in the West. The Information provides the harder read: four years after US export controls were designed to kneecap China's AI infrastructure, Chinese models have gained global token share, US tech giants are now eyeing Chinese chips, and America's AI industrial strategy is underperforming its original ambitions. This is not a symmetric competition — China's state-directed capital allocation and domestic chip progress are compounding simultaneously.
AI Infrastructure Capital Markets: CoreWeave's Bond Success Meets IPO Caution
CoreWeave completed a $4.2 billion convertible bond offering — upsized from an initial $3 billion target — confirming that institutional debt markets remain open for established AI infrastructure names at scale, per The Information. Separately, SoftBank launched what is shaping up to be one of the largest-ever corporate junk bond sales at record yields to fund its AI investment programme, according to Bloomberg. The Reuters observation that corporate bond buyers are growing 'picky with a flood of AI debt' suggests investors are differentiating by quality rather than retreating wholesale.
The equity IPO market tells a more cautious story. Semafor reports that multiple AI-related IPOs have been postponed as investor appetite softens for smaller or more concentrated bets. Nscale — a British AI data centre developer whose revenue is heavily concentrated in Microsoft and Anthropic — is testing that appetite with its own IPO, per TechCrunch. Meanwhile, AI infrastructure firm Accelevation is targeting a $5.4 billion valuation in a US IPO. The pattern is bifurcation: debt markets remain liquid for tier-one names, while equity markets are demanding clearer customer diversification and proven unit economics before granting premium valuations to smaller infrastructure plays. CME Group's GPU futures contract — designed to make AI compute an investable asset class — has also stalled at the CFTC, which is extending its regulatory review, per The Information.
Healthcare AI Attracts Serious Growth Capital; Industrial AI Embeds Pre-IPO
Melbourne-based Heidi Health closed a $100 million fundraising round at a $900 million valuation, with CEO Thomas Kelly outlining plans to expand into additional markets and deepen AI integration into clinical workflows, according to Bloomberg. Separately, Anthropic announced a partnership with OpenEvidence to bring medical AI to practitioners globally, per Reuters — an arrangement that positions Claude as a clinical-grade model without Anthropic needing to build the healthcare distribution layer itself. Healthcare remains one of the clearest sectors where AI is moving from pilot to embedded workflow, driven by documentation burden, diagnostic support, and consultation efficiency rather than speculative use cases.
In industrial AI, Barrick Gold's North American unit has agreed to a five-year partnership with predictive maintenance startup Avathon ahead of a planned IPO, per The Information. The deal illustrates two simultaneous dynamics: AI vendors are securing long-duration enterprise contracts to build revenue predictability, and industrial companies are using AI partnerships to signal technology sophistication to public market investors. Snorkel AI's $350 million Series E at a $3.5 billion valuation — triple its prior mark — reflects parallel demand for the data infrastructure layer that makes domain-specific AI training possible across sectors including healthcare and industrials, per TechCrunch.
Signals & Trends
AI Compute Is Becoming a Sovereign Asset Class — With Fragmented Control
Three simultaneous developments point to a structural shift in who controls AI compute. Anthropic is negotiating direct data centre tenancy to escape hyperscaler dependency. Alibaba is building a 20-gigawatt global infrastructure stack with its own chips. SoftBank is raising junk debt at record yields to fund AI positions. CME's GPU futures are stalled at the CFTC. What's emerging is not a single global compute market but a fragmented set of overlapping infrastructure stacks — US hyperscaler, Chinese state-linked, and private-credit-funded independent — each with different cost structures, regulatory exposures, and customer bases. Investors who assumed a single AI infrastructure trade are now navigating a multi-polar compute landscape where the question is not just which chips win but which infrastructure stack captures which workloads by geography and regulatory jurisdiction.
The AI Agent Commerce War Has Opened a Second Front in Retail Distribution
Amazon's decision to block Meta's Muse AI agent from shopping on its platform while Shopify actively welcomed it is the first concrete instance of retailers using platform access as a competitive weapon in the AI agent era. This is not a privacy dispute — it is a distribution battle. If AI agents become the dominant purchase interface, whoever controls agent access controls the point-of-sale. Amazon's move is defensive: it cannot afford to let a Meta-controlled agent intermediate its own customer relationships. Shopify's countermove is offensive: it positions the platform as the agent-friendly alternative to closed marketplaces. For investors in commerce infrastructure, payments, and retail technology, the question is now which platforms embed agent access as a feature and which treat it as an existential threat — because that strategic choice will determine traffic flows and margin structures in the next e-commerce cycle.
US AI Industrial Strategy Is Underperforming Its Own Ambitions — and the Market Is Pricing It In
The confluence of signals today is pointed: China has achieved credible chip performance despite export controls, Chinese models have gained global token share, US AI IPOs are stalling, GPU futures face regulatory friction, and Treasury Secretary Bessent is being eyed as a potential AI czar rather than a dedicated technology policymaker. The Information's analysis that America's AI dream is 'failing to launch' as an industrial strategy — four years after export controls were positioned as the decisive intervention — is a significant reassessment. Meanwhile, investors are described as 'playing both sides of the AI divide' as Trump and Xi meet, which implies capital markets have already internalised a more competitive landscape than the policy community has publicly acknowledged. The gap between US political rhetoric on AI dominance and the actual competitive dynamics now represents a meaningful risk for portfolios built on the assumption of sustained American structural advantage in AI infrastructure and models.
Explore Other Categories
Read detailed analysis in other strategic domains