Anthropic files for $965B IPO, deploys 50K NVIDIA Blackwell GPUs

Anthropic filed confidential IPO paperwork on Monday, setting a $965 billion valuation that would make it the largest initial public offering in history. The filing caps a staggering capital-raising spree. The company recently closed a $65 billion fundraising round and underscores the immense financial momentum behind frontier AI development. Meanwhile, IREN Limited and BE Networks are deploying NVIDIA DSX Air to simulate and validate network architecture for over 50,000 NVIDIA Blackwell Ultra GPUs, a sign that AI infrastructure buildouts are scaling at a pace that demands digital-twin de-risking before physical construction. Why this matters now: the IPO filing and the GPU deployment represent twin signals that the AI industry is transitioning from research-stage experimentation to industrial-scale capital deployment, with implications for hyperscaler capex, chip supply chains, and regulatory frameworks.

The $965 billion valuation and its investor logic

Anthropic’s confidential IPO filing values the company at $965 billion, a figure that reflects investor conviction that frontier AI models will generate enterprise-grade recurring revenue at scale. The valuation is supported by the company’s recent $65 billion fundraising round, which included participation from major institutional investors and sovereign wealth funds. The IPO paperwork, filed confidentially with the SEC on Monday, does not disclose the number of shares or the expected price range, but the $965 billion figure makes it larger than any previous technology IPO, including Alibaba’s $25 billion debut in 2014 and Saudi Aramco’s $29.4 billion listing in 2019. The valuation implies that investors are pricing Anthropic not as a startup but as a platform company with the potential to rival the largest cloud providers and enterprise software firms. The company’s revenue model centers on Claude Code, its developer platform, and API access for enterprise customers, though Anthropic has not publicly disclosed revenue figures. The confidential filing gives Anthropic flexibility to adjust its offering based on market conditions, a common tactic for large IPOs. The $965 billion figure also reflects the premium investors place on AI companies with proprietary foundation models, as opposed to application-layer startups that depend on third-party APIs. Anthropic’s valuation is now roughly three times that of xAI, Elon Musk’s AI venture, which was valued at $300 billion in its most recent funding round.

How $15 billion in SpaceX data center costs reshapes the P&L

Anthropic’s IPO filing reveals that the company pays $15 billion per year to SpaceX for access to data center capacity, according to SpaceX’s own IPO filing. This figure is staggering: it represents a significant portion of Anthropic’s operating expenses and highlights the hidden infrastructure costs that frontier AI companies must bear. The $15 billion annual payment covers GPU clusters, power, cooling, and networking in SpaceX’s data centers, co-located with SpaceX’s rocket launch infrastructure to exploit dedicated power grids that commercial data centers cannot match. For context, $15 billion is larger than the entire annual revenue of most public cloud companies. The cost structure confirms that Anthropic is running training and inference workloads on a massive scale, using NVIDIA Blackwell Ultra GPUs across SpaceX's dedicated GPU clusters. The SpaceX arrangement also indicates that Anthropic is willing to pay a premium for guaranteed capacity in a market where GPU availability remains constrained. The $15 billion figure will be a key line item for IPO analysts assessing Anthropic’s gross margins and path to profitability. The company will need to demonstrate that its revenue can grow faster than its infrastructure costs, a challenge that has plagued every AI company from OpenAI to Google DeepMind. The SpaceX deal also raises questions about concentration risk: if Anthropic’s access to SpaceX data centers were disrupted, the company would face a severe operational bottleneck.

IREN and BE Networks rewrite the GPU deployment playbook

IREN Limited and BE Networks are using NVIDIA DSX Air to simulate the deployment of over 50,000 NVIDIA Blackwell Ultra GPUs before a single chip is physically installed. DSX Air is a digital twin platform that models network architecture, power distribution, and cooling requirements at the AI-factory scale. BE Networks provides its Verity automation platform for Day 0, Day 1, and Day 2 operations, meaning the simulation covers everything from initial provisioning to ongoing maintenance. This approach reduces deployment risk by identifying network bottlenecks, thermal hotspots, and configuration errors in the virtual environment before they cause delays or failures in the physical buildout. For IREN, a company that has pivoted from bitcoin mining to AI infrastructure, the simulation capability is critical: it allows the firm to guarantee deployment timelines to hyperscaler customers without overbuilding physical capacity. NVIDIA DSX Air integrates with NVIDIA Spectrum-X Ethernet and NVIDIA NVLink to validate that the network fabric can handle the data throughput required by 50,000 Blackwell Ultra GPUs. The simulation also models power consumption at the rack level, enabling IREN to optimize its power purchase agreements and avoid costly over-provisioning. BE Networks’ Verity platform automates the configuration of switches, routers, and GPU servers, reducing the manual labor required for large-scale deployments. The partnership between IREN, BE Networks, and NVIDIA signals that AI infrastructure is moving from bespoke, one-off builds to standardized, simulation-validated deployments. This digital-twin approach is already being adopted by other hyperscalers for their own AI factory projects, further validating the model.

Microsoft’s coding model comeback reshapes the competitive landscape

Microsoft is preparing to release a new coding model next week in a bid to reclaim ground lost to Anthropic’s Claude Code and GitHub Copilot. The Information reports that Microsoft’s AI independence day is approaching, with the company developing its own foundation models rather than relying entirely on OpenAI’s technology. Microsoft’s new coding model is designed to compete directly with Claude Code, which has gained traction among enterprise developers for its ability to generate, debug, and refactor code across multiple languages. The competitive dynamics are shifting: Microsoft has invested billions in OpenAI but is now building in-house capabilities to reduce dependency on external model providers. This move mirrors Anthropic’s strategy of developing proprietary models while also securing exclusive access to infrastructure through the SpaceX deal. The coding model market is the most contested segment in enterprise AI, with GitHub Copilot, Amazon CodeWhisperer, Google’s Gemini Code Assist, and xAI’s Grok all vying for developer mindshare. Microsoft’s advantage lies in its distribution: Visual Studio and GitHub have hundreds of millions of users. However, Anthropic’s Claude Code has built a reputation for handling complex, multi-file refactoring tasks that simpler autocomplete tools cannot manage. The outcome of this competition will determine which company captures the developer ecosystem, the most valuable entry point for enterprise AI adoption. Developer tools carry outsized strategic weight because they create sticky workflows: once a team standardizes on a coding assistant for complex refactoring, migration costs are high. Microsoft's incumbency in the IDE layer gives it an installation base that Anthropic cannot replicate purely through API superiority. Anthropic's counter-leverage is accuracy on production-grade codebases, where Claude Code's multi-file context handling outperforms autocomplete-style tools. The developer tool fight is effectively a proxy war for enterprise AI platform dominance, and whichever company wins it shapes the revenue composition of AI infrastructure spending for the rest of the decade.

Trump’s executive order and the state regulation battleground

The Trump administration issued Executive Order 14365 on December 11, 2025, establishing a national AI policy framework designed to challenge state-level AI regulations and sustain US global AI dominance. The order directly targets state laws that impose restrictions on AI development, particularly in healthcare, finance, and autonomous systems. For Anthropic, the regulatory environment is a double-edged sword: federal preemption of state laws could reduce compliance costs and accelerate deployment, but the order also signals that the administration will scrutinize AI companies’ safety practices. The executive order creates a federal AI oversight body with the authority to review large-scale model deployments, including the kind of infrastructure buildout that Anthropic is pursuing with SpaceX and IREN. The order’s impact on the IPO is significant: investors will need to assess the risk that federal regulation could impose new costs or limit the company’s ability to deploy models in certain sectors. State-level regulation has been a growing concern for AI companies, with California, New York, and Washington proposing bills that would require safety testing, bias audits, and transparency disclosures. The executive order aims to create a single national standard, but it also invites legal challenges from states that argue AI regulation falls under their police powers. For NVIDIA, the regulatory landscape directly affects Blackwell Ultra GPU demand: federal rules that slow AI deployment will force hyperscalers to delay GPU purchase orders and shift capital to compliance. The reverse is equally true — when the order removes state-level barriers, AI infrastructure demand rises in step with the freed deployment capacity.

The IPO filing and the GPU deployment are not isolated events. They are the leading edge of a structural shift in how AI companies raise capital, build infrastructure, and navigate regulation. Anthropic’s $965 billion valuation will test whether public markets share the private market’s conviction that frontier AI models are a generational investment. The $15 billion SpaceX data center cost will force the company to demonstrate a clear path to operating leverage, or face the same margin compression that has dogged every capital-intensive technology business. IREN and BE Networks’ simulation-driven deployment model will likely become the standard for AI factory construction, reducing the risk of costly delays and enabling faster scaling. Microsoft’s coding model launch will determine whether the developer ecosystem consolidates around a single platform or fragments across multiple providers. And the Trump executive order will shape the regulatory terrain for years to come, determining whether AI companies operate under a single national framework or a patchwork of state laws. For investors, the next twelve months will reveal whether the AI industry’s infrastructure spending can translate into sustainable revenue growth, or whether the capital intensity of frontier AI creates a winner-take-most dynamic that only a handful of companies can survive.