Tech company announces layoffs, blames "AI efficiencies." The implication of these headlines is that AI made all these jobs redundant. It's a bit cheeky for no one to mention the part where the AI tools doing the replacing, actually cost more than the people they replaced.
Bryan Catanzaro, VP of applied deep learning at NVIDIA, said it straight out, "the cost of compute is far beyond the costs of the employees." Right now, the robots cost more than the humans did! Maybe that's blowout infrastructure spend, or maybe it's "tokenmaxxing" (burning tokens for the sake of looking busy). Either way, it turns out people were the cheaper option all along.
Uber's CTO, Praveen Neppalli Naga, torched the entire annual AI budget in four months flat. The culprit was AI-assisted coding, and Uber's not even pretending otherwise. Swapping basic chatbots for agentic coding tools sent costs sideways in a way nobody saw coming.
Startups have it worse. Amos Bar-Joseph, CEO of Swan AI, is paying $113,000 a month in API costs for a team of four. That's a chunky bill for a startup, especially since you still don't have any colleagues to play foosball with.
If you've used any of these tools yourself, you've probably felt it already. Flat subscriptions are turning into usage limits, with enterprise pricing being token-based, so the more you lean on it, the more it costs you. The free lunch is ending with AI companies not able to eat the difference forever.
Agentic AI chews through tokens at a completely different rate to a basic chatbot, mostly because it's working autonomously around the clock instead of waiting for you to ask it something. Coding agents are even worse, they need huge volumes of tokens to write, check and rewrite code before they hand you anything usable.
So AI is expensive to build and expensive to run. Inference costs have dropped, sure, but token usage has gone up just as fast thanks to all this extra reasoning happening under the hood. Next time a company says AI made them do the layoffs, it's worth digging a little deeper. Plenty of these companies were making bad calls long before AI could be blamed.
Tech company announces layoffs, blames "AI efficiencies." The implication of these headlines is that AI made all these jobs redundant. It's a bit cheeky for no one to mention the part where the AI tools doing the replacing, actually cost more than the people they replaced.
Bryan Catanzaro, VP of applied deep learning at NVIDIA, said it straight out, "the cost of compute is far beyond the costs of the employees." Right now, the robots cost more than the humans did! Maybe that's blowout infrastructure spend, or maybe it's "tokenmaxxing" (burning tokens for the sake of looking busy). Either way, it turns out people were the cheaper option all along.
Uber's CTO, Praveen Neppalli Naga, torched the entire annual AI budget in four months flat. The culprit was AI-assisted coding, and Uber's not even pretending otherwise. Swapping basic chatbots for agentic coding tools sent costs sideways in a way nobody saw coming.
Startups have it worse. Amos Bar-Joseph, CEO of Swan AI, is paying $113,000 a month in API costs for a team of four. That's a chunky bill for a startup, especially since you still don't have any colleagues to play foosball with.
If you've used any of these tools yourself, you've probably felt it already. Flat subscriptions are turning into usage limits, with enterprise pricing being token-based, so the more you lean on it, the more it costs you. The free lunch is ending with AI companies not able to eat the difference forever.
Agentic AI chews through tokens at a completely different rate to a basic chatbot, mostly because it's working autonomously around the clock instead of waiting for you to ask it something. Coding agents are even worse, they need huge volumes of tokens to write, check and rewrite code before they hand you anything usable.
So AI is expensive to build and expensive to run. Inference costs have dropped, sure, but token usage has gone up just as fast thanks to all this extra reasoning happening under the hood. Next time a company says AI made them do the layoffs, it's worth digging a little deeper. Plenty of these companies were making bad calls long before AI could be blamed.