Sam Altman admits OpenAI "screwed up" GPT-5 rollout, warns of AI bubble

OpenAI CEO Sam Altman just broke his silence on the messy GPT-5 rollout, the ongoing AI bubble, and the company's next multi-trillion-dollar bets.

Over a rare, hours-long dinner with reporters in San Francisco, Altman candidly admitted that OpenAI "totally screwed up some things" when launching GPT-5 last week. The biggest mistake was pulling the plug on GPT-4o, which sparked an immediate user revolt over its missing warmth. Just an hour before sitting down to dinner, Altman personally made the call to bring 4o back as an option for paid subscribers.

Crushed GPUs and 700 million users

Despite the backlash, OpenAI's infrastructure is buckling under sheer demand. API traffic doubled in just 48 hours following the launch, and the company is once again out of GPUs.

Altman revealed that ChatGPT now reaches over 700 million weekly active users, roughly quadrupling in size over the past year. It currently ranks as the fifth-largest website globally. Altman believes passing Instagram and Facebook for the number three spot is well within reach, though taking down Google remains a far steeper climb.

He also touched on the controversial topic of emotional attachment. While Altman estimates users with genuinely unhealthy relationships make up way under 1 percent, internal meetings on the issue are constant. He took a subtle jab at rivals like Grok, stating OpenAI has no interest in building anime sex bots and intends to keep ChatGPT's default stance firmly in the center.

Hardware, Chrome, and the bubble

Altman was remarkably blunt about financial realities. He believes investors are currently overexcited and that AI is in a classic bubble, though the underlying technology remains transformative.

To keep scaling, Altman warned that OpenAI will need to spend trillions of dollars on data center construction in the near future. He claimed the company would already be highly profitable today if not for its astronomical training costs.

Beyond data centers, Altman confirmed plans to back a brain-computer interface startup to take on Neuralink. He teased that the upcoming consumer hardware device co-designed with Jony Ive represents a rare computing paradigm shift. He even hinted at an appetite to buy Google Chrome if antitrust regulators force a sale.

As for his own future, Altman admitted he is not naturally suited to run a public company, jokingly wondering if an AI might replace him as CEO within three years.

Why it matters

Altman's candid remarks highlight the immense growing pains of running a product with hundreds of millions of daily habits attached to it.

We think the immediate restoration of 4o proves that AI has crossed the line from a raw software utility into an emotional companion. Users do not treat major model upgrades like silent backend patches anymore; changing an AI's conversational tone now triggers genuine social backlash.

For developers and the broader industry, the severe GPU crunch means compute shortages will continue to constrain model access and dictate product roadmaps for the foreseeable future. Altman's talk of spending trillions on data centers is an aggressive signal to rivals that survival requires massive, unprecedented capital expenditure. At the same time, expanding into hardware, brain chips, and browser acquisitions looks like an ambitious attempt to own the entire user stack before the current hype cycle cools.