Bill Gates warns generative AI has hit a plateau

While the rest of the tech world is hyping up the next quantum leap in AI, Bill Gates is hitting the brakes on the hype train.

In an interview with German business outlet Handelsblatt, the Microsoft co-founder dropped a reality check: he believes the current generative AI technology has hit a plateau. Gates stated that, contrary to the optimism coming out of OpenAI, he does not expect GPT-5 to be significantly better than its predecessor. He did, however, admit he could be wrong.

The leap is over, and chips are still expensive

Looking back at the benchmarks, Gates called the jump from GPT-2 to GPT-4 incredible. But as far as he is concerned, GPT-4 marks a boundary, and the industry should brace for an initial period of stagnation.

Then there is the sheer cost of keeping the lights on. Gates pointed out that specialized Nvidia AI chips cost around $30,000 each, chewing up massive amounts of computing power and electricity. While training frontier models remains prohibitively expensive, operational costs are improving. Gates noted that the cost per query has dropped from around ten cents to roughly three cents today, though the overall infrastructure bill remains staggering.

Practical gains over raw intelligence

Even if model intelligence hits a wall, Gates still sees massive short-term value in practical deployments. Over the next two to five years, he expects AI accuracy to climb noticeably alongside steep cost drops, finally yielding dependable commercial applications.

He highlighted healthcare and climate resilience as prime examples. Developing nations could benefit from basic medical diagnostics delivered through mobile phones, and drug and vaccine development could accelerate rapidly. On the sustainability side, Gates mentioned he is investing in nearly 100 companies utilizing AI to modernize power grids, alongside efforts to engineer heat-resistant crops.

As for the elusive black box problem, Gates believes deciphering how AI encrypts and processes information will be a major scientific milestone. When it comes to AGI, he made it clear that nobody truly knows when it will land, even if its ultimate arrival reshapes humanity.

Why it matters

We think Gates is delivering a dose of practical realism that builders desperately need right now.

For developers and founders, the takeaway is straightforward: stop designing business roadmaps around the assumption that GPT-5 will magically solve all your hallucination and reasoning issues. If foundation models are plateauing, the competitive edge shifts entirely to execution, cost management, and reliable domain integration. If query costs are hovering around three cents, your job is figuring out how to deliver ten cents of value on top of that today.

For the broader industry, Gates pointing out a plateau puts pressure on model providers to justify their massive capital expenditures. Buying $30,000 chips only makes sense if each generation delivers exponential capability. If diminishing returns set in early, the real winners will not be the companies chasing raw model scale, but the ones quietly applying existing models to dull, massive problems like electricity grids and pharma pipelines.