Whalesbook
### The Compute Capacity Gambit
Anthropic is set to receive a substantial influx of computing power, with SpaceXAI finalizing a deal to grant access to its Colossus 1 facility. This agreement will inject 300 megawatts of new capacity into Anthropic's operations within a month, utilizing over 220,000 Nvidia GPUs housed at the Memphis mega data center [5, 9]. This critical capacity boost directly addresses Anthropic's well-documented constraints and aims to expand usage limits for its Claude Code and Claude API services, which have been a bottleneck for development and deployment velocity throughout 2026 [5].
The partnership highlights Anthropic's ongoing strategy of securing vast compute resources from established technology providers. This approach contrasts with the years-long, capital-intensive process of building proprietary infrastructure. Anthropic's existing compute deals are significant, including an agreement for up to 5 gigawatts (GW) with Amazon Web Services, nearly 1 GW of which is expected by the end of 2026, and a 5 GW agreement with Google slated to commence in 2027 [4, 25]. Furthermore, the company has secured substantial Azure capacity commitments from Microsoft and NVIDIA, valued at $30 billion [4, 5]. These collaborations underscore the immense scale of compute required for frontier AI models and reflect Anthropic's estimated valuation, which surged to approximately $350 billion by November 2025 and reached $380 billion by February 2026, with recent pre-IPO market indicators suggesting an implied valuation around $1.2 trillion by May 2026 [4, 20, 25, 30]. The AI infrastructure market itself is projected to expand from $158.3 billion in 2025 to $418.8 billion by 2030, indicating a robust growth environment driven by demand for specialized hardware like GPUs [46].
### Musk's Pragmatic Pivot and SpaceX's AI Ambitions
This collaboration represents a notable shift for Elon Musk, who had previously expressed public criticisms of Anthropic on X [5]. By providing essential compute infrastructure, Musk’s SpaceXAI is now a key enabler for the AI startup, signaling a pragmatic approach to capitalizing on the AI boom. For SpaceX, the deal offers revenue and utilization for its Colossus 1 infrastructure, while its own AI division focuses on training Grok models [5]. The agreement also includes an expression of interest in developing multiple gigawatts of orbital AI compute capacity, a concept Musk has championed [5, 7]. SpaceX is reportedly targeting a mid-2026 IPO with valuation estimates ranging from $1.5 trillion to over $1.75 trillion, with its AI and space infrastructure ambitions playing a significant role in its valuation narrative [6, 31]. The company's own filings acknowledge that its orbital AI initiatives are in early stages and involve significant technical complexity and unproven technologies, though it projects orbital data centers could become more cost-effective than terrestrial facilities within years [5, 31].
### THE FORENSIC BEAR CASE: Dependency, Interdependence, and Infrastructure Risk
Anthropic's strategy of leasing compute power, while enabling rapid scaling, introduces inherent dependencies. The company’s valuation has grown astronomically, with annual revenue projected to reach $26 billion by 2026 [4], yet its reliance on third-party infrastructure providers like SpaceX, Amazon, Google, and Microsoft exposes it to potential price fluctuations, supply chain disruptions, or strategic shifts by its partners. The immense capital expenditure by hyperscalers—companies like Google (market cap ~$4.8T), Amazon (market cap ~$2.96T), and Microsoft (market cap ~$3.07T)—to build out their own AI data centers, estimated at $7.6 trillion cumulatively between 2026 and 2031, underscores the sheer cost of foundational AI infrastructure [44]. Anthropic's leasing model, while agile, may face higher long-term operational costs compared to owning dedicated, depreciated infrastructure.
Furthermore, the technical viability and economic practicality of SpaceX's ambitious orbital AI compute plans remain uncertain. Despite Musk's optimism, SpaceX itself has admitted these initiatives are in their early stages, involve unproven technologies, and may not achieve commercial viability, citing risks from the harsh space environment causing hardware wear and potential malfunctions [5, 31]. While NVIDIA (market cap ~$5.05T) dominates the AI hardware market with over 80% share due to its CUDA ecosystem, the geopolitical landscape and concentration risk associated with relying so heavily on a single vendor for critical components like GPUs present another layer of concern [2, 21].
### The Evolving AI Infrastructure Frontier
The global AI infrastructure market is experiencing exponential growth, projected to reach $418.8 billion by 2030 with a compound annual growth rate of 21.5% [46]. The AI computer market is expected to see similar expansion, reaching $281.67 billion by 2030 at a CAGR of 34.4% [12]. This surge is driven by the insatiable demand for compute power from large language models (LLMs) and generative AI applications. The trend towards specialized, purpose-built AI data centers, often featuring dense configurations of GPUs and high-speed interconnects, is becoming the norm, transforming how compute resources are managed [8, 24]. The current deal between SpaceXAI and Anthropic is a microcosm of this larger trend, showcasing the intense competition and strategic alliances forming to secure the foundational compute capabilities necessary for AI advancement.