The $800 Billion Revision

In February 2026, OpenAI informed investors that it had slashed its total compute spending target through 2030 to approximately $600 billion. This represents a staggering 57% reduction from the $1.4 trillion, 30-gigawatt infrastructure ambition previously circulated by CEO Sam Altman just months prior. [1]

This $800 billion downward revision is arguably one of the most significant infrastructure forecast corrections in modern technological history. It does not represent a failure of the technology, but rather an industry-wide capitulation to the uncompromising physical and financial laws of “inference economics.” [2]

Training vs. Inference: The Economics Shift

In the lifecycle of artificial intelligence, model training requires immense, episodic capital expenditure. However, model inference—the actual computational process of running the AI to generate answers, analyze documents, and execute agentic workflows for end-users—introduces a continuous, compounding operating expense. [4]

As mass-market adoption scales, inference rapidly overtakes training as the dominant cost driver. OpenAI’s internal metrics illustrate this crisis of success: in 2025, the company generated $13.1 billion in revenue, significantly surpassing its $10 billion projection, while managing to keep overall spending to $8 billion, under its $9 billion target. [1][3]

However, the underlying unit economics revealed a structural vulnerability: the expenses associated with running its AI models (inference costs) increased fourfold in 2025. Despite tripling revenue, the exploding costs of inference caused OpenAI’s adjusted gross margins to plummet from 40% in 2024 to just 33% in 2025. [3]

The 900-Million-User Cost Problem

The sheer mathematics of serving hundreds of millions of weekly active users mandates a total shift in corporate focus. The ChatGPT user base alone rebounded to over 900 million, making continuous inference the dominant operational cost. [5]

Success is no longer defined strictly by headline model capabilities or parameter counts, but by rigorous unit economics: cost per inference, token generation speed, power efficiency, and total system utilization. [4]

Under these updated forecasts, OpenAI expects cumulative cash burn to reach an astonishing $665 billion through 2030, with the company not anticipating positive cash flow until the end of the decade. [1]

Fundraising at Scale: The $100 Billion Round

To bridge this massive cash flow gap until profitability, OpenAI is finalizing a historic fundraising round seeking over $100 billion, which would value the firm at $830 billion. [1]

A critical component of this round is a near-finalized $30 billion investment from Nvidia, reflecting the symbiotic reliance between the primary model developer and the primary hardware supplier. [1]

Despite the margin compression, OpenAI remains bullish on its top-line potential, projecting total revenue to exceed $280 billion by 2030, divided equally between consumer subscriptions and enterprise software sales, as it lays the groundwork for a potential initial public offering (IPO) in the fourth quarter of 2026. [1]