Capital & Industrial Strategy
Top Line
Anthropic commits $11.6 billion to Akamai over seven years in the largest deal in Akamai's history, signalling a strategic bet on CPU-based inference at scale and introducing an equity kicker — up to 5% of Akamai stock — that redefines how frontier AI labs structure infrastructure partnerships.
A divided federal appeals court upholds the Pentagon's designation of Anthropic as a supply-chain risk, a ruling with direct consequences for Anthropic's ability to win U.S. government contracts and for its imminent IPO, which founders are simultaneously seeking to insulate via a 50.1% founder voting-control structure.
British AI neocloud Nscale closes $3.36 billion in convertible financing ahead of a U.S. IPO, with Nvidia among the backers, reinforcing that the infrastructure layer of the AI stack continues to attract institutional capital despite crowding concerns.
Inference-layer startups Fal and Fireworks AI are exploring new funding rounds — Fal in early talks at a reported $15 billion valuation — as soaring developer demand for fast, cost-efficient model serving creates a distinct and increasingly well-capitalised sub-sector.
Meta's personal AI agent Muse is outpacing ChatGPT's early user-growth numbers, forcing Microsoft into a defensive product consolidation with its revamped Copilot super-app and marking a tangible shift in where consumer AI monetisation may actually land.
Key Developments
Anthropic's Akamai Deal Redraws Infrastructure Economics for Frontier Labs
Anthropic has signed a seven-year, $11.6 billion cloud infrastructure commitment with Akamai, with a stated potential to grow to approximately $20 billion, according to TechCrunch. The deal is confirmed and represents the largest contract in Akamai's history. The strategic logic is multi-layered: Anthropic gains access to a CPU-heavy infrastructure stack differentiated from the GPU-dominated hyperscaler offerings of AWS, Google Cloud, and Azure, suggesting the company is optimising for inference cost efficiency at scale rather than training capacity alone. The equity component — Akamai granting Anthropic warrants for up to 5% of its stock, vesting as spend increases — is structurally significant. It effectively turns a vendor relationship into a partial alignment of interests, reducing Anthropic's net infrastructure cost on a risk-adjusted basis and giving Akamai a direct stake in Anthropic's success ahead of IPO.
The deal also signals Anthropic's intent to diversify away from dependence on Amazon Web Services, which has committed up to $4 billion to Anthropic as an investor and cloud partner. Spreading $11.6 billion across Akamai over seven years, alongside existing AWS and Google commitments, suggests Anthropic is deliberately constructing a multi-cloud architecture to preserve negotiating leverage and operational resilience. Bloomberg characterised this as Anthropic going big on compute at a moment when Microsoft is rethinking its AI infrastructure posture — a direct competitive contrast.
Pentagon Blacklisting of Anthropic Upheld — IPO Governance Structure Reveals the Stakes
A divided 2-1 panel of the D.C. Circuit Court of Appeals upheld the Department of Defense's designation of Anthropic as a supply-chain risk, siding with the Trump administration in a ruling authored by Judge Gregory Katsas, a Trump appointee, according to Wired and CNBC. Anthropic had argued the designation violated its constitutional rights; the dissenting judge agreed with at least some of those arguments, indicating the legal question is genuinely contested. The practical consequence is that Anthropic remains excluded from direct DoD procurement pipelines, a significant limitation for a company positioning itself as a safety-focused, enterprise-grade AI provider. The political dimension is equally important: the ruling lands as Anthropic is preparing for a public offering, and any perception of adversarial positioning with the federal government creates material risk for institutional investors evaluating the IPO.
Simultaneously, TechCrunch reports that Anthropic's seven co-founders are seeking shareholder approval for a dual-class structure granting them 50.1% voting control on most corporate matters. This is a defensive move calibrated for the post-IPO environment: it insulates management from activist pressure and ensures the founders retain authority over strategic decisions — including any future attempts by outside shareholders to push the company toward government accommodation. The governance play and the Pentagon ruling together create a coherent, if high-risk, picture: Anthropic is going public with a confrontational federal government relationship and a founder-controlled structure designed to weather it.
AI Infrastructure Capital Continues to Flow — But the Cracks Are Appearing
Nscale, a British AI neocloud, has secured $3.36 billion in convertible financing from Third Point, Nvidia, and other investors ahead of a planned U.S. IPO, according to TechCrunch. This is confirmed committed capital, not a rumoured round. Nvidia's participation is strategically notable: the chip giant has a consistent pattern of backing GPU-hungry neoclouds to extend demand for its hardware, and Nscale's data center buildout is precisely the kind of infrastructure that absorbs Nvidia capacity at scale. The convertible structure gives investors downside protection while maintaining upside participation in what is being positioned as a pre-IPO growth vehicle. The deal underscores that the AI infrastructure layer continues to attract large institutional commitments despite saturation concerns.
Against this backdrop, Crusoe has abandoned its $1.25 billion plan to deploy Boom Supersonic's stationary turbines as power sources for AI data centers, according to TechCrunch. Boom CEO Blake Scholl confirmed the turbines are no longer in Crusoe's near-term plans. The abandonment reflects a pattern of ambitious alternative energy-AI infrastructure partnerships struggling to reach execution — the gap between announced capital commitments and operational deployments is widening. Separately, The Wall Street Journal reports that Oracle's Project Jupiter data center development in New Mexico is encountering force majeure notices, a removed development partner, and local community resistance. These are confirmed operational difficulties, not speculation, and they signal that even well-capitalised hyperscale buildouts face non-trivial execution risk.
The Application Layer Absorbs Capital as Meta's Muse Forces a Consumer AI Reckoning
Meta's personal AI agent Muse is topping app store charts and reportedly outpacing ChatGPT's early user-growth trajectory, according to TechCrunch, with Meta now actively cross-promoting Muse across its family of apps. This is the most significant consumer AI traction signal since ChatGPT's 2022 launch, and it carries major capital implications: Meta is monetising AI through its existing distribution moat rather than building a standalone subscription business, which structurally disadvantages pure-play AI consumer app companies that cannot match Meta's zero customer-acquisition-cost advantage. Microsoft has responded by consolidating its previously fragmented Copilot products — AI coding tools, Office 365 automation, and an always-on agent feature — into a unified super-app, according to The Information and CNBC. The consolidation is a defensive product move that also reflects Microsoft's continued search for a coherent AI monetisation strategy.
Accel Partner Matt Weigand, speaking to Bloomberg, articulated a view now gaining traction in venture: the infrastructure layer is maturing and the application layer is where differentiated returns will be captured. This is consistent with the inference startup funding activity — Fal reportedly in early discussions at a $15 billion valuation and Fireworks AI also exploring new rounds, per The Information — where value is accruing to the enabling layer between raw models and end applications. The Fal figure is an early-stage discussion reported by two sources; no terms are confirmed.
Signals & Trends
Google DeepMind Talent Exodus Is Seeding a New London AI Venture Cohort
Bloomberg reports that a breakfast meeting of 15 Google DeepMind employees and alumni in central London turned immediately to startup fundraising discussions, reflecting a broader pattern of technical talent departing established AI labs to found ventures. This matters for capital allocation because DeepMind alumni carry brand equity that compresses early fundraising timelines and attracts premium valuations — the VC frenzy Bloomberg describes is a rational response to the quality of the talent pool now becoming available. London is emerging as a second concentration point for this activity alongside San Francisco, and European sovereign wealth and pension capital is increasingly being positioned to capture it. Senior investors should track which DeepMind alumni form founding teams in the next six months, as these will attract disproportionate early-stage capital and could define the next generation of foundation model challengers.
The Equity-for-Infrastructure Model Could Structurally Alter How AI Labs Finance Compute
Anthropic's Akamai deal, in which the vendor grants equity warrants that vest as spend increases, is a structural innovation worth tracking beyond its immediate context. It solves a genuine tension for frontier AI labs: they require massive, long-duration infrastructure commitments that strain balance sheets pre-IPO, while infrastructure vendors need revenue certainty to justify capacity investment. The equity kicker aligns incentives without requiring the AI lab to raise additional dilutive equity purely for infrastructure financing. If this model replicates — and infrastructure vendors competing for AI lab contracts have strong incentives to offer similar terms — it would effectively transfer a portion of AI lab upside to the infrastructure layer, changing the risk-return profile of companies like Akamai, CoreWeave, and Nscale relative to pure capital-markets fundraising. The Federal Reserve's flagged concern about the AI ecosystem becoming 'too big to fail,' noted by Reuters, adds a systemic dimension: regulators are beginning to model the financial interconnectedness of these arrangements.
Chinese AI Model Adoption Is Globalising Faster Than U.S. Policy Can Respond
CNBC reports a substantial increase in global business adoption of Chinese AI models in 2026, a development that is registering as a concern in Washington. The Trump-Xi summit addressed AI safety and cooperation framing, with Xi calling for joint prevention of AI misuse, but the structural issue for capital markets is different: if Chinese models — likely DeepSeek variants and successors — are capturing enterprise market share in non-U.S. markets at scale, the total addressable market projections underpinning current U.S. AI valuations may be systematically overstated for international segments. This is a slow-moving risk that will not appear in near-term earnings but will shape long-range competitive positioning. Investors with exposure to U.S. AI companies dependent on global enterprise revenue should be modelling Chinese model penetration by geography, particularly in Southeast Asia, the Middle East, and Latin America, where U.S. export controls have less purchase and price sensitivity is high.
Explore Other Categories
Read detailed analysis in other strategic domains