Anthropic's Revenue Soars to $16.3 Billion in First Half, Net Loss of $80.7 Billion
Artificial intelligence company Anthropic achieved revenue of $16.3 billion in the first half of 2026, an increase of nearly 15 times from $1.1 billion in the same period last year, with a net loss of $80.7 billion during the same period.
Of the net loss, $72 billion came from accounting expenses related to the fair value changes of convertible notes and other financing instruments. Excluding this non-cash item, the operating loss was approximately $8.7 billion; infrastructure spending for the first half was close to $9 billion, surpassing the full-year figure of $7.3 billion for 2025.
The company is obligated to pay approximately $450 billion under signed cloud computing agreements over the next 7 to 10 years, with the vast majority of contracts being non-cancelable, requiring payment regardless of actual usage; this includes approximately $111 billion to Google, about $110 billion to Amazon Web Services, and around $31 billion to Microsoft.
Total future spending related to computing power could reach $750 billion, including agreements still under negotiation; about one-fifth of revenue comes from just two clients, and these clients are not bound by long-term contracts.
The company is preparing for an initial public offering, with a target valuation of up to $2 trillion; third-quarter revenue has reached $16.8 billion, a quarter-over-quarter increase of about 45%.
The Amodei brothers have committed to donating 80% of their shares to charity; the company also warns of risks that models may cause real-world harm, as well as political resistance faced by data centers.
In market mechanisms, funds mainly flow to cloud service providers like Google, Amazon, and Microsoft to secure computing power, representing event-driven long-term irrevocable commitments; beneficiaries include cloud providers and chip-related parties, while Anthropic itself faces pressure from the need for continuous financing to cover fixed expenses, as well as risks related to customer concentration and usage uncertainty.
Source: Public Information
ABAB AI Insight
Anthropic was founded by former core members of OpenAI and has historically secured large-scale financing from investors like Google and Amazon, while simultaneously signing large computing power procurement agreements; previous public documents indicated revenue of about $4.6 billion and a net loss of $42 billion for 2025. The data for the first half of this year continues the trend of high growth alongside high losses, accompanied by fluctuations in the fair value of valuation tools amplifying book losses.
The capital path converts investor funds and customer prepayments into long-term irrevocable commitments for cloud services and equipment leasing, locking in hundreds of billions of dollars in computing power with Google, Amazon, and Microsoft, while distributing Claude through cloud market channels; the motivation is that the supply of computing power becomes a hard constraint for the iteration of cutting-edge models, ensuring the pace of training and inference by pre-locking capacity rather than waiting for the spot market.
This is similar to the capital expenditure cycles of early cloud providers or semiconductor manufacturers locking in wafer capacity through long-term agreements; Anthropic is currently in an expansion phase from model development to reliance on large-scale infrastructure, with revenue growth far exceeding traditional software, but the fixed expense structure is closer to that of heavy asset operators.
Essentially, this represents a restructuring of the industry chain and capital concentration: through non-cancelable contracts, the pricing power and delivery certainty of the computing power supply side are concentrated among a few cloud providers, with the mechanism being that demand side exchanges long-term payments for priority capacity, while suppliers gain visible income and investment recovery, simultaneously constraining the expansion speed of AI laboratories by their own financing capabilities and contract performance, rather than purely technological breakthroughs.
ABAB News · Cognitive Law
Soaring revenue cannot mask fixed expenses; locked contracts equate to an advance on future earnings.
Non-cash losses can be adjusted, but non-cancelable computing power bills cannot be adjusted.
Customer concentration combined with long-term commitments means that the faster the growth, the higher the rigid costs.