Anthropic reports $104 billion loss as it warns of recursive self-improvement

Massive losses and infrastructure spend

Anthropic's latest financial disclosures reveal a staggering $104 billion loss since the beginning of the year. According to the company's prospectus, the majority of this figure stems from specific balance sheet adjustments rather than daily operations. The company's actual operating loss for the period was $567 million.

Despite these heavy losses, the company's revenue has seen a significant spike in the first nine months of 2026. The firm reported that its revenue climbed from $2.5 billion to $33.2 billion during this timeframe. This growth reflects the increasing demand for its AI models even as the costs of development remain high.

Looking ahead, Anthropic plans to invest $895 billion into data centers and infrastructure over the coming years. However, it remains unclear if all of this capital will actually be spent as planned. This massive spending comes at a time when some investors are beginning to question if these enormous expenditures will ever yield a profit.

The risk of recursive self-improvement

In the same prospectus, Anthropic issued a warning regarding the potential for AI to accelerate its own development. The company notes that its current model, Claude, already assists in building newer versions by writing code and performing tests. As models become more powerful, they may take on a larger share of this development work.

Experts refer to this phenomenon as 'recursive self-improvement.' In extreme scenarios, an AI could potentially develop its own successors entirely without human intervention. Anthropic acknowledges that it has only noticed major leaps in AI capabilities in hindsight, which makes it difficult to predict or control every step of the process.

While the warning highlights a technical risk, it also serves a marketing purpose as the company prepares for an initial public offering. The more dangerous a technology sounds, the more significant it appears to potential investors. The company is balancing the need to highlight real risks with the desire to demonstrate the importance of its technology.

Local resistance to data centers

While Anthropic plans massive infrastructure spending, local communities in the United States are increasingly pushing back. During recent local elections, voters in 41 different areas—particularly in Ohio—voted on the future of AI data centers. In all 37 cases where the results were counted, the opposition won.

These victories did not result in a total ban on data centers in every location. Instead, some towns established a moratorium, while others required local governments to undergo a consultation process before making decisions. In some small communities, only a few hundred people voted, but the opposition was often clear and decisive.

Residents cite several concerns regarding the growth of these facilities. Beyond the high consumption of electricity and water, neighbors are complaining about the constant noise produced by the plants day and night. This local friction could create significant hurdles for companies trying to build the physical infrastructure needed for AI.

The state of AI in Germany

The debate over AI infrastructure and sovereignty also reached the Digital Summit in Berlin. Chancellor Friedrich Merz emphasized the importance of developing independent capabilities in both hardware and software. He noted that while domestic capabilities are vital, the country still requires trustworthy partners, specifically naming Google as a key collaborator.

German intelligence services are currently working with the U.S. company to build a new offline cloud. Merz defended this choice by stating that the Google Cloud is sufficiently sovereign for their needs. However, many experts in politics and industry disagree with this stance and argue that authorities should prioritize booking domestic cloud services.

Advocates for domestic services argue that only by choosing local providers can Germany invest in European alternatives. They worry that relying on U.S. giants limits the growth of a sovereign digital ecosystem. This tension highlights the ongoing struggle to balance the need for high-performance tools with the desire for national and regional independence.