“OpenAI and Anthropic, the world’s two most valuable AI companies, are aiming to go public in the fourth quarter of 2026. Until now, what was known about their finances was based on estimates and rumors. Now, the *Wall Street Journal* has analyzed confidential financial documents that both companies presented to investors prior to their most recent funding rounds.
For the first time, these documents provide a direct comparison showing just how much money is consumed by training new artificial intelligence (AI) models, where the break-even point lies, and how different the paths to reaching it are. OpenAI plans to spend around $121 billion on computing power in 2028 alone. Even if revenue were to double, this would still result in a loss of $85 billion. Anthropic’s forecasts are less extreme but follow the same logic: massive upfront investment today to secure market share in a technology that could transform entire economies.
The fundamental premise shared by both companies can be summed up in a single formula: every leap in an AI model’s performance capabilities costs more than the last. Yet, falling behind in the race means losing access to the most lucrative applications. Consequently, OpenAI and Anthropic are releasing new model versions at increasingly short intervals while simultaneously pouring more resources into training runs. The AI arms race shows no signs of slowing down.
Training costs are so immense that both companies now report two profitability metrics: one including training costs and one excluding them. If these expenses are left out, both are on the verge of a small operating profit. When they are factored in, however, OpenAI does not expect to break even until 2030, with Anthropic reaching that milestone only slightly earlier.
The reason venture capitalists accept losses of this magnitude is the unprecedented revenue growth. Yet, beneath this shared pace of growth lie two fundamentally different economic bets. OpenAI is banking on the reach of a consumer platform. 900 million people use ChatGPT, yet only a small fraction pay for it. The strategy mirrors that of social networks: first build a massive user base, then monetize it through advertising and subscriptions.
Anthropic focuses on paying business clients who integrate AI models into their workflows. This business model generates less public attention but yields more stable revenue. The coding assistant Claude Code, in particular, has proven to be a key driver of growth. Consequently, Anthropic’s projected annual revenue has surged from $9 billion to $30 billion over the past four months—surpassing even the figures cited in its investor documents.
When Anthropic released Claude Code last autumn, OpenAI was caught off guard and invested heavily in its own assistant, Codex. OpenAI is currently refocusing on enterprise clients, as investors view this segment as more stable and less risky. While the number of private users paying for subscriptions is steadily rising, it is apparently not increasing fast enough to cover rapidly growing costs or to guarantee the company a compelling narrative for an IPO. The reasoning of OpenAI’s leadership team is reminiscent of Amazon’s early days, when Jeff Bezos prioritized growth over profit and the stock price rose only after years of stagnation.
The fundamental economic question facing both companies is this: Will artificial intelligence become a foundational technology for a wide range of applications—comparable to the steam engine or electricity? Or, better yet, will it become a foundational technology that accelerates its own pace of innovation, driving ever-faster economic growth? Economists refer to this as the "invention of a method of invention," the impact of which would far exceed the one-time step-changes brought about by the steam engine or electricity.
If AI does indeed produce these effects—a possibility supported by recent technical breakthroughs, though not yet a certainty to guarantee this—then today’s investments would be more than justified, as the technology would enable rising productivity gains across virtually every economic sector. However, if the scaling laws—whereby increased computing power leads to better AI models—hit their limits, those astronomical valuations could quickly come under pressure.
Both AI companies are banking on initial public offerings (IPOs) to secure long-term funding. OpenAI closed the largest private funding round in history at the end of March: $122 billion at a valuation of $852 billion. Anthropic raised $30 billion in February at a valuation of $380 billion.
To handle the massive amount of capital that needs to be raised through these IPOs, investment banks are changing known rules: Nasdaq will grant newly listed companies faster access to its indices in the future, allowing index funds to purchase the shares sooner.
Despite record-breaking official funding rounds, the secondary market takes a more nuanced view. According to Bloomberg, investors are holding OpenAI shares worth around $600 million for which there are currently no buyers, as the bet on OpenAI CEO Sam Altman is deemed too risky. In contrast, there is strong demand for Anthropic shares. At least a dozen investors who had previously bet exclusively on OpenAI recently participated in Anthropic’s funding round. That round is also considered ambitious, yet less risky.
The momentum clearly lies with Anthropic: Claude Code was a milestone that continues to electrify the tech world. The new "Mythos" model is reportedly so advanced that access is initially being restricted to select tech companies tasked with identifying security vulnerabilities. OpenAI is currently unable to keep pace, even if Altman refuses to admit it. His CFO, Sarah Friar, recently felt the consequences of this; she, too, considered the planned fourth-quarter IPO overly ambitious and has since been stripped of her direct reporting line to Altman. Criticism of the boss appears unwelcome at OpenAI.” [1]
1. Die teuerste Wette der Technikgeschichte. Frankfurter Allgemeine Zeitung; Frankfurt. 09 Apr 2026: 19. Von Holger Schmidt, Frankfurt
Komentarų nėra:
Rašyti komentarą