Why OpenAI Missed Its User and Revenue Targets

TL;DR

OpenAI missed its target of 1 billion weekly active ChatGPT users before end of 2025 and fell short on 2025 revenue, per a Wall Street Journal report, even as it carries $600 billion in compute spending commitments. Its consumer business is weak, but enterprise coding demand is surging with ChatGPT 5.5, while power and supply-chain constraints now cap the whole industry.

Transcript

Jason, do you want to tell us about your new favorite podcast? Oh, it's so good. My feed is now because, you know, since cancel culture ended, Sachs, everybody uses the R word and the f word right now. My entire feed on Instagram is either gay or down syndrome or bulldogs. It's one of those three. And then I stumbled upon the Miss Thing pod, Miss T... Read More

Key Insights

• OpenAI expected to reach 1 billion weekly active ChatGPT users before the end of 2025 but missed that milestone and still had not hit it four months into 2026, according to a Wall Street Journal investigative report.
• OpenAI has $600 billion in spending commitments for compute, roughly equal to the entire value of the enterprise on secondary markets, against a reported $20-30 billion revenue run rate for ChatGPT.
• CFO Sarah Frier is reportedly worried revenue is not growing fast enough to keep up with expenses, and she does not think OpenAI is ready for public reporting standards even though Altman wants to IPO later this year.
• ChatGPT 5.5 has drawn strong reviews from Silicon Valley developers and coders, and it is built on a new base model called Spud, the first base-model upgrade OpenAI has done in over a year.
• Anthropic's Opus 4.7 appears to be a bust, with users rolling back to 4.6 and complaining it rations compute and reduces thinking time, a decision Sacks calls prudent given Anthropic being token constrained.
• The single choke point constraining AI, per Chamath, is access to the power needed to drive tokens, meaning missed forecasts stem from supply of compute capacity rather than any shortfall in demand.
• Less than half of the gigawatts announced for AI data centers are actually being built, with most stuck in red tape and supply-chain delays for transformers, gas turbines, and grid infrastructure.
• Power constraints hurt Anthropic and OpenAI most while benefiting hyperscalers like Oracle, Amazon, Meta, Microsoft, and Google, opening a lane for Grok and SpaceX, which hold excess capacity.

Questions & Answers

Q: How many users did OpenAI expect ChatGPT to reach and did it hit the target?

According to a Wall Street Journal investigative report published Tuesday, OpenAI expected to hit 1 billion weekly active users for ChatGPT before the end of 2025. It missed that target and still had not reached the milestone four months into 2026. OpenAI also missed its 2025 revenue target for ChatGPT, though the exact number was not specified, against a reported $20-30 billion run rate.

Q: Why does OpenAI's $600 billion compute commitment matter?

OpenAI has $600 billion in spending commitments for compute, an amount roughly equal to what the entire enterprise trades for on secondary markets. In other words, the value of the whole company equals its spend commitments in the coming year. Against a run rate of only $20-30 billion, the CFO is reportedly worried that revenue is not growing fast enough to keep up with expenses, calling the affordability of these data-center commitments into question.

Q: What is the conflict between Sarah Frier and Sam Altman about?

CFO Sarah Frier, who is coming into liquidity, is reportedly worried that ChatGPT revenue is not growing fast enough to keep up with expenses. OpenAI wants to IPO later this year, but Frier does not think the company is ready for public reporting standards, while Altman wants to move faster. According to the Wall Street Journal this created tension between them, prompting the two to release a joint statement dismissing the report.

Q: Why does David Sacks have a contrarian take on OpenAI's bad week?

Sacks argues that at the product level the past couple of weeks were actually good for OpenAI. They released ChatGPT 5.5, which drew strong reviews from Silicon Valley developers and coders. By contrast Anthropic's Opus 4.7 appeared to be a bust, with users rolling back to 4.6 and complaining it rationed compute and reduced thinking time. Sacks sees coding-usage momentum shifting from Opus toward GPT 5.5.

Q: What is the Spud base model and why is it significant?

GPT 5.5 is based on a new base model called Spud, which Sacks describes as the first base-model upgrade OpenAI has done in over a year. Having a new base model is significant because it paves the way for future product improvements as well. Sacks believes OpenAI is feeling optimistic about its product right now and that these improvements will continue, helping shift developer momentum toward GPT 5.5.

Q: Why does Chamath say the AI industry is power constrained rather than demand constrained?

Chamath argues there is one specific choke point constraining everything: access to the power needed to drive tokens. When companies like OpenAI miss a number or forecast, he says it reflects insufficient compute capacity today, not weak demand. The reason firms miss forecasts is entirely due to the supply of power necessary to generate output tokens, and that problem is only getting worse as backlogs build.

Q: What supply-chain problem could make AI capacity shortages worse?

Chamath notes backlogs are building not just for power access but for the componentry needed, including transformers, natural gas turbines, resiprocating engines, and tactical grid infrastructure. Looking at gigawatts under construction, less than half of announced projects are actually being built, with most stuck in red tape and supply-chain delays. As a result there is no credible strategy to turn much of this announced capacity on.

Q: Who benefits and who is hurt by AI power constraints?

Chamath says the power and compute constraints will hurt Anthropic and OpenAI the most, since they must negotiate how much equity and control to give up to secure compute. The constraints benefit the hyperscalers, specifically Oracle, Amazon, Meta, Microsoft, and Google. He adds this opens a big lane for Grok and SpaceX, which have excess capacity, framing the Grok-Cursor deal as just the appetizer and suggesting Elon and Dario could do a deal.

Summary & Key Takeaways

• A Wall Street Journal report revealed OpenAI missed its target of 1 billion weekly active ChatGPT users before end of 2025 and fell short on 2025 revenue. This matters because OpenAI has $600 billion in compute spending commitments while running at roughly $20-30 billion in revenue, raising questions about affordability ahead of a planned IPO.

• David Sacks offered a contrarian take: while the consumer business is weak, the product side had a strong stretch. ChatGPT 5.5, built on a new base model called Spud, drew praise from developers, while Anthropic's Opus 4.7 disappointed and pushed users back to 4.6. Coding usage is shifting from Opus toward GPT 5.5, so Sam may be right for the wrong reason.

• Chamath argued the real constraint is power to generate output tokens, not demand. Less than half of announced gigawatts are actually being built due to red tape and supply-chain backlogs in transformers and turbines. This hurts OpenAI and Anthropic while benefiting hyperscalers and leaves excess-capacity players like Grok and SpaceX room to run through the market.