Goldman Sachs just delivered a reality check at their biggest tech conference of 2025: if AI spending decelerates sharply in Q4 2025 and 2026, tech stocks could drop 20%. That's not a casual warning from some random analyst - this is Goldman's top tech team looking at the numbers and seeing problems.
Here's what has them spooked: companies are burning through AI budgets at an unsustainable rate. Microsoft, Google, Amazon, and Meta are each spending tens of billions on AI infrastructure, and most of it isn't generating revenue yet. OpenAI's CFO Sarah Friar said revenue will triple to $13 billion, which sounds impressive until you realize they're spending way more than that on compute and talent.
The math doesn't work. I've seen this movie before - dotcom companies spending money like water while promising future profits that never materialized. The difference is that AI infrastructure is way more expensive than websites. Each H100 GPU costs $40,000, and training a large language model burns through thousands of them.
Goldman's analysts think the current spending pace is unsustainable. Companies are building data centers for AI workloads that don't exist yet, betting that demand will catch up to supply. Maybe it will, or maybe we're building the 2025 equivalent of dark fiber networks that never got used.
The warning signs are already there. Several AI startups have quietly shut down after burning through Series A funding in months instead of years. The cost of training competitive AI models keeps increasing faster than revenue. Even OpenAI, the golden child of AI, is reportedly burning through $7 billion a year while making a fraction of that in revenue.
What really caught my attention: Goldman expects stock buybacks to rise 12% to $1.2 trillion next year, but warned that growth could be lower if AI spending continues at current levels. Translation: companies might have to choose between buying back stock to prop up share prices or betting everything on AI.
The smart money is starting to get cautious. Nvidia's stock has been volatile despite record earnings, and other AI infrastructure companies are trading at insane multiples based on future promises. When Goldman Sachs, who made billions from the AI boom, starts warning about a correction, it's time to pay attention.
This doesn't mean AI is bullshit - the technology is real and transformative. But the current spending levels assume exponential growth in AI adoption that might not happen as fast as everyone expects. If companies start pulling back on AI investments in late 2025, the whole sector could crash harder than crypto in 2022.