OpenAI, Anthropic Financials Reveal a $121 Billion IPO Problem

Financial projections for OpenAI and Anthropic, shared with investors ahead of recent funding rounds, reveal a stark reality: the cost to build the next generation of artificial intelligence is soaring faster than their explosive revenue growth. OpenAI, for instance, expects to spend a staggering $121 billion on computing power in 2028, projecting an $85 billion loss that year despite nearly doubling its sales, according to the documents.

"We are prioritizing growth over profits and could dial back spending on training but expect a strong return on investment," an OpenAI spokesperson said in a statement, acknowledging the aggressive spending strategy.

The documents paint a picture of an escalating arms race for AI supremacy. Both companies are pouring billions into training new models, with each leap in capability costing more than the last. To make their financials more palatable, both firms present two profitability metrics: one including the massive "compute for research" costs, and one without. Excluding these costs, both are near profitability. Including them, OpenAI doesn't foresee breaking even until the 2030s.

This cash burn creates immense pressure for blockbuster initial public offerings. The documents show both OpenAI and Anthropic will have negative free cash flow in the billions for years to come, meaning they are counting on public market investors to fund their colossal operational costs. The need is so great that bankers are reportedly lobbying index providers to change rules to allow these cash-intensive firms faster access to inclusion in major indexes like the Nasdaq.

The Soaring Cost of Intelligence

The core of the financial challenge lies in the runaway costs of training AI models. As these systems become more powerful, the computational resources required increase exponentially.

According to the financial projections, OpenAI's AI model training costs are forecast to eclipse $100 billion annually by the end of the decade. Anthropic's spending is projected to be lower but follows the same steep upward trajectory. This spending on computing power, primarily for GPUs from companies like Nvidia, represents the single largest expense and the primary obstacle to near-term profitability.

The documents show these training costs will consume a massive portion of revenue. For OpenAI, training costs are projected to be over 300 percent of revenue in 2024. While this percentage is expected to decrease, it highlights the fundamental business model challenge: selling access to AI models must eventually cover the enormous fixed cost of creating them.

Revenue Rises, But Can It Keep Pace?

Despite the costs, revenue growth for both companies is among the fastest in tech history. Both OpenAI and Anthropic expect to more than double revenue this year as business customers adopt their AI tools. OpenAI's revenue streams are diversified across enterprise clients, consumer subscriptions for ChatGPT, and new products, which include hardware.

However, the revenue comparison isn't a direct one. Anthropic's projections count sales of its technology through cloud partners like Amazon as revenue, a standard accounting practice that inflates its top line relative to OpenAI, which does not.

Beyond training, the cost of running the models—known as "inference"—also consumes more than half of the revenue for each company. While this is expected to become more efficient over time, it remains a significant cash drain, particularly for OpenAI, which supports millions of non-paying ChatGPT users. An OpenAI spokesperson said this is a strategic choice to drive adoption, with opportunities to monetize those users later.

The financial documents underscore a critical question for investors as these AI leaders march toward the public markets: can revenue growth and efficiency gains outrun the colossal, ever-growing cost of building true artificial intelligence? The answer will determine if they become the next great tech titans or cautionary tales of unsustainable spending.

This article is for informational purposes only and does not constitute investment advice.