Capital & Industrial Strategy
Top Line
Anthropic has signed a deal to use all of SpaceX's Colossus 1 compute capacity — 300 megawatts, 220,000+ Nvidia GPUs — a strategic necessity driven by 80-fold Q1 revenue growth, and a deal that hands xAI a new revenue stream as a compute supplier to rivals.
DeepSeek is entering its first external funding round at a valuation of $45–50 billion, with Chinese government-backed investors set to take a stake, formally aligning the lab with Beijing's technology self-sufficiency agenda.
SpaceX filed plans for a 'Terafab' vertically integrated semiconductor and compute fabrication facility in Texas costing up to $119 billion, signalling Musk's intent to control the full AI hardware stack from chip to data centre.
Samsung crossed a $1 trillion market cap on AI chip demand, South Korea surpassed Canada as the world's seventh-largest equity market, and AMD hit a record high — the AI infrastructure trade is reshaping global capital market rankings.
Snap confirmed its $400 million Perplexity integration deal has ended, a rare signal that not all AI distribution partnerships survive contact with commercial reality.
Key Developments
Anthropic-SpaceX Compute Deal: Competitive Necessity Creates an Unlikely Alliance
Anthropic has signed an agreement granting it access to all compute capacity at SpaceX's Colossus 1 data centre in Memphis, Tennessee — 300 megawatts powered by more than 220,000 Nvidia GPUs, to be fully available by end of May. The deal is confirmed and closed. CEO Dario Amodei publicly attributed the move to 80-fold revenue growth in Q1 2026, framing the compute constraint as the binding bottleneck on Anthropic's expansion. CNBC WSJ
The strategic geometry here is striking. Anthropic — backed by Google and Amazon — is now a paying customer of xAI's compute infrastructure, funnelling revenue to Elon Musk's AI empire at the precise moment Musk is litigating against OpenAI. For SpaceX/xAI, this validates the emerging thesis that xAI's real business is neocloud infrastructure, not just frontier model development. TechCrunch The deal also includes a space-based computing development component, pointing toward orbital data centre ambitions that give SpaceX's launch infrastructure a direct monetisation path within the AI supply chain. Semafor Semafor separately notes this represents a material shift in Anthropic's capital strategy — previously more conservative on compute spend relative to hyperscaler norms. Semafor
DeepSeek's State-Backed Fundraise Formalises China's AI Industrial Strategy
DeepSeek is in the process of closing its first external investment round at a valuation reported at $45 billion by TechCrunch TechCrunch and $50 billion by the WSJ WSJ, with capital coming from government-backed investors. This is an announced intention, not a closed deal — the valuation discrepancy between sources reflects ongoing negotiations. The funding explicitly aligns DeepSeek with Beijing's technology self-sufficiency push, transforming what was a commercially independent lab into a strategic national asset.
The valuation places DeepSeek — a lab that achieved frontier performance at a fraction of US compute costs — at a significant premium reflecting its geopolitical as much as commercial value. For Western AI investors, this raises the competitive stakes: a Chinese lab that can train competitive models at low cost, now with state capital backing and insulation from commercial pressures, is a structurally different competitive threat than a pure commercial rival. The round also signals that Chinese AI capital formation, which had been led by internet giants like Meituan (leading Moonshot AI's $2 billion round at a $20 billion valuation Bloomberg), is now supplemented by direct state intervention at the frontier.
SpaceX's Terafab: Vertical Integration Ambition at the Scale of a National Programme
SpaceX has filed plans for a 'Terafab' facility in Texas described as a multi-phase, vertically integrated semiconductor manufacturing and advanced computing fabrication facility, with a capital cost of up to $119 billion. TechCrunch CNBC This is a filed proposal, not a committed capital expenditure — the multi-phase structure means actual spending will depend on regulatory approvals, technology milestones, and financing. The stated intent is to manufacture chips for Tesla, SpaceX, and xAI, i.e., to internalise semiconductor supply across the Musk industrial complex.
The strategic logic is clear: Musk's enterprises are currently dependent on Nvidia and TSMC for the compute that underpins Tesla's autonomy stack, xAI's model training, and SpaceX's guidance systems. A vertically integrated fab eliminates that dependency and, if successful at scale, creates a potential third-party supply business. At $119 billion, the project would rival TSMC's US investment commitments in scale. The Anthropic compute deal simultaneously demonstrates that Musk already has a paying customer for his existing data centre capacity — a proof-of-concept for the broader infrastructure monetisation thesis.
AI Infrastructure Capital: From Data Centres to Optical Fibre, the Build-Out Deepens
Nvidia has committed up to $3.2 billion in investment in Corning to fund three new US advanced manufacturing plants dedicated to optical fibre for AI infrastructure. CNBC This is a confirmed deal. Simultaneously, Hut 8 — a crypto miner pivoting to AI compute — signed a Texas data centre lease worth at least $9.8 billion with an undisclosed investment-grade counterparty. Bloomberg Arm guided to $2 billion in sales of its new in-house AI chip from next year, while warning of smartphone market weakness — confirming the data centre AI segment is now the primary growth driver for the chip architecture business. FT
The Nvidia-Corning deal is analytically significant beyond its dollar value: it shows Nvidia extending its supply chain control downstream into the physical infrastructure layer, ensuring optical interconnect capacity keeps pace with GPU shipment volumes. This mirrors the pattern of hyperscalers investing in subsea cable — the bottleneck migrates and the incumbent integrates to control it. The WSJ separately noted that investors are repricing glass and specialty materials companies as AI infrastructure plays, reflecting how broadly the capital is now flowing through the supply chain. WSJ Samsung crossing $1 trillion on chip demand, with SK Hynix also at record highs, confirms memory as a sustained high-margin beneficiary of the AI cycle, though WSJ analysts flagged margin sustainability questions. WSJ
Enterprise Adoption: From Pilots to P&L Trade-offs
Two data points this week clarify the enterprise adoption dynamic. Match Group confirmed it is slowing hiring to fund increased AI tool expenditure, explicitly citing cost as the driver. TechCrunch Meanwhile, Starbucks and Lowe's are integrating into OpenAI's ChatGPT interface to build customer-facing AI applications — a distribution play that uses OpenAI's consumer footprint rather than building standalone AI products. WSJ Apple's R&D spending has crossed 10% of revenue for the first time in 30 years, growing twice as fast as sales — a signal that even hardware-centric incumbents feel existential pressure to accelerate AI investment. CNBC
The Match Group case illustrates a trade-off that will recur across mid-cap enterprises: AI tools are expensive enough that adoption requires explicit headcount substitution rather than pure additive investment. This is the mechanism by which AI spending translates into labour market impact, and it is now appearing in public company disclosures. The Starbucks/Lowe's ChatGPT integration pattern suggests that OpenAI's consumer distribution network is becoming a platform for enterprise customer acquisition — a dynamic that threatens both pure-play enterprise AI vendors and the Anthropic consumer push announced this week. Bloomberg
Signals & Trends
xAI Is Becoming an Infrastructure Business — and the Market Has Not Fully Priced the Implications
The convergence of three developments this week — the Anthropic compute deal, the Terafab filing, and analyst commentary about xAI's neocloud positioning — suggests that Musk's AI empire is being restructured around infrastructure ownership rather than model competition. The Colossus 1 capacity being rented to Anthropic is a proof of commercial demand at scale. Terafab, if partially executed, extends that logic to chip manufacturing. The strategic implication for the broader market is that xAI may be less of an OpenAI competitor and more of an AWS-style infrastructure layer that charges AI labs — including rivals — for access. This would make xAI's revenue model more defensible and less dependent on winning the model quality race. Investors valuing xAI purely on Grok's model performance relative to GPT-5 or Claude 4 are likely underweighting the infrastructure monetisation thesis.
Chinese AI Capital Formation Is Accelerating Through Two Parallel Channels — and They Are Converging on Frontier
This week saw Moonshot AI raise $2 billion at $20 billion from Meituan (private internet capital) and DeepSeek preparing to close at $45–50 billion with government-backed investors. These are distinct channels — commercial internet giants backing consumer AI applications, and the state backing efficiency-focused frontier research — but both are now operating at scale simultaneously. The convergence risk for Western competitors is significant: Chinese labs will have access to both commercial distribution infrastructure (via Meituan, Alibaba, Tencent) and state-subsidised compute and research capacity (via the DeepSeek investment). Alibaba's outperformance over Tencent driven by semiconductor exposure adds a third dimension — Chinese internet companies are also competing to become AI infrastructure providers domestically. Western AI investors should monitor whether the DeepSeek state investment comes with compute access provisions that would partially circumvent existing export controls.
The 'Compute Tax' Policy Debate Is Entering the Mainstream Just as Infrastructure Spend Becomes Undeniable
The WSJ published analysis of the 'compute tax' concept this week — a levy on AI compute consumption to fund displaced worker programmes — at exactly the moment when the AI infrastructure investment cycle is hitting visible scale: $119 billion chip fabs, $10 billion data centre leases, Samsung at $1 trillion. The timing is not coincidental. As AI capex becomes a dominant driver of equity market performance and macro capital flows, the political salience of who captures versus who loses from AI productivity gains will increase. The Trump administration's stated position — Susie Wiles confirming the government will not pick winners — rules out US industrial policy intervention on the model selection side, but leaves open infrastructure subsidies (CHIPS Act) and potentially does not foreclose revenue-based levies. Strategy professionals should track whether the compute tax concept moves from op-ed to legislative proposal in any G7 jurisdiction in the next 12 months, as it would directly affect the economics of every AI infrastructure investment currently being underwritten.
Explore Other Categories
Read detailed analysis in other strategic domains