Accounting effect of financing, compute costs, and infrastructure saturation risk: Anthropic's financial x-ray
Artificial intelligence startup Anthropic recorded net losses exceeding $80 billion during the first half of 2026. The figures come from a draft prospectus for its Initial Public Offering (IPO) cited by The Wall Street Journal. In parallel, Ken Griffin, founder of hedge fund Citadel, warned of the risk that the supply of AI infrastructure and processors could end up exceeding market demand.
Accounting charges for convertible notes and operating revenue
The company's finances reflect the accounting complexity associated with capital-intensive financing rounds:
- Non-cash impact of liabilities: Around $72 billion of the total losses corresponded to accounting adjustments resulting from changes in the valuation of convertible notes and other financial instruments previously issued to raise capital.
- Annual net loss comparison: In the first half of the previous year, the company had reported net losses of $13 billion.
- Operating performance and sales: Although Anthropic's revenue multiplied by 12 in 2025 to reach nearly $4.6 billion, the firm recorded operating losses of over $8 billion (without taking into account liability accounting write-downs).
Citadel's warnings: The risk of GPU oversupply
Ken Griffin, founder of Citadel, analyzed the investment outlook and structural risks facing the artificial intelligence sector during a Citadel Securities conference:
- Parallel with the dot-com bubble: Griffin compared the current environment with the internet boom of the late 1990s. He noted that major economic transformations are often accompanied by bubbles where many valuations prove unsustainable, although the investments ultimately build the infrastructure for enduring companies.
- Semiconductor overcapacity (GPUs): The investor warned that the greatest risk to profitability is that the installed capacity of graphics processing units (GPUs) will exceed actual compute demand.
- Risk of infrastructure price deflation: Griffin compared this scenario to the overbuilding of fiber optics in the telecommunications sector around the year 2000. If the supply of GPUs outpaces market needs, compute capacity rental prices could plummet, impacting return on investment timelines for data centers.
At next+ we consider that Anthropic's financial report evidences that the gap between revenue growth and AI operating costs remains the sector's primary structural challenge. The magnitude of accounting losses tied to convertible notes foreshadows the pressure language model startups will face when listing on public markets, where investors will demand visibility into actual operating profitability. For chief financial officers, investment committees, and technology leaders, Citadel's warnings regarding GPU oversupply suggest that strategic value will begin shifting from the raw accumulation of compute infrastructure toward inference cost efficiency and the effective monetization of enterprise applications.