Anthropic's IPO Prospectus Reveals Massive Spending, Ken Griffin Calls AI a Bubble

Anthropic's first-half 2026 revenue was reportedly $16.3 billion, a substantial increase from $1.1 billion in the first half of 2025, according to a new draft of its IPO prospectus seen by the Wall Street Journal. However, the company also reported a net loss of $80.7 billion, largely attributed to accounting related to convertible shares and notes.

In the first half of 2026, the company's compute and infrastructure spending reached nearly $9 billion, up from $2.3 billion in the same period last year. Anthropic is projected to spend up to $750 billion on compute capacity in the coming years as it seeks to acquire the resources it needs.

Anthropic has made significant long-term financial commitments to major cloud providers. This includes $111 billion to Google, $110 billion to Amazon Web Services (AWS), and $31 billion to Microsoft, with these contracts spanning a 7- to 10-year period.

A critical aspect of these agreements is their non-cancellable nature, meaning Anthropic is obligated to pay these amounts regardless of actual usage. This highlights a substantial financial risk for the company.

In addition to cloud infrastructure, Anthropic has committed $86 billion to rent Nvidia chips from Nscale and Lambda, with these contracts extending through 2033 and 2034. Furthermore, $160 billion is allocated for leasing Google chips through a deal with Broadcom. The company's total commitments currently stand at $600 billion, with ongoing negotiations that could potentially increase this to $750 billion.

The S1 prospectus reportedly indicates that just two customers account for roughly one-fifth of Anthropic's first-half revenue. This significant customer concentration poses a risk to the company's financial stability.

Any potential reduction or instability in capital expenditure (capex) spending by these major hyperscalers could severely impact Anthropic's revenue. If these key customers, likely two of the hyperscalers, were to pull back, it would create a substantial challenge for the company's financial projections.

Anthropic's prospectus highlights various risk factors, including safety concerns and growing opposition to data center construction. The President has suggested that data center builders should consider offering direct payments to communities rather than solely focusing on building schools or other local infrastructure. This proposition reflects a broader conversation about how these large-scale projects can benefit local populations.

This approach of direct payments is a tactic that has been previously associated with certain political figures, raising questions about its feasibility and broader implications for community engagement and regulatory frameworks within the AI industry.

Ken Griffin, who heads the hedge fund Citadel, one of the largest and most successful hedge funds globally, recently stated at a Bloomberg forum that AI is 'almost certainly a bubble'. Griffin was notably associated with the fallout of the hedge fund Situational Awareness.

His comments come as the AI sector sees massive investments and rapidly escalating valuations, prompting discussions about the sustainability of its current growth trajectory.

Ken Griffin explicitly referred to AI as a bubble but qualified his statement by noting that major technological progressions often lead to bubbles. He emphasized that these cycles inevitably produce winners and losers in the market.

Griffin believes that transformative technologies that change the world tend to emerge from the aftermath of these bubbles. Companies that successfully navigate and survive the market correction are those that ultimately perform exceptionally well in the long term.

Market analysts anticipate a potential downward rerating for AI-focused companies as the market cools off and moves beyond the initial IPO hype. Investors are expected to shift their focus from speculative enthusiasm to demanding concrete financial data.

The market will eventually scrutinize the actual worth of these companies based on real numbers, leading to a re-evaluation of current valuations. This suggests a period where the true financial performance will dictate market sentiment rather than projected growth.

Jensen Huang, CEO of Nvidia, acknowledged the incredible modern demand for AI, citing companies like Coreweave and Scale as examples. He highlighted a significant 'float' period, which is the time between when computers are shipped and when they begin generating revenue.

This delay can last for many months and necessitates substantial upfront capital investment from companies. Huang emphasized Nvidia's goal to streamline the supply chain, making it easier for companies to deploy computing power faster and thereby reduce this 'float' period.

Analysts from Gartner and Bain have raised concerns about the financial sustainability of current AI spending patterns. They point to a significant discrepancy, or 'lumpiness', in the supply chain between the massive upfront investment in infrastructure and the actual generation and recognition of end-market revenue.

Companies are making substantial upfront investments in chips, assuming that these will eventually yield returns. However, there are questions about whether the revenue growth will materialize quickly enough to justify these considerable capital costs, particularly given deployment delays. The concern is not whether revenue will eventually appear, but whether it will happen at a pace sufficient to validate the current level of expenditure.

Concerns are emerging regarding 'circular financing' in the AI industry, where companies backing IPOs reportedly use funds to purchase chips from the same entities. This practice raises questions about the transparency and genuine market valuation of these AI firms.

Additionally, the industry faces 'force majeure' risks due to challenges in building necessary physical infrastructure. For instance, an Australian data center company called Next DC is reportedly encountering local opposition and potential construction delays for its projects, highlighting the broader issue of community acceptance and the ability to rapidly expand data center capacity.