Cost of compute > employees

Companies are spending more on AI compute than on the humans using it. That's not a prediction, it's a line item on real budgets right now. And the implications run deeper than most people realize. Bryan Catanzaro, Nvidia's VP of Applied Deep Learning, put it bluntly in an April 2026 interview with Axios: "For my team, the cost of compute is far beyond the costs of the employees." At roughly $28,000 per month in AI compute per team member, the machines have quietly overtaken the people operating them. He's not alone in noticing. Uber's CTO reportedly blew through the company's entire 2026 AI budget before March, largely due to token costs from tools like Claude Code. Swan AI's CEO burned through their full AI budget in just two months. These aren't cautionary tales from startups with no financial discipline. These are well-funded, sophisticated organizations hitting the same wall.

The numbers keep climbing

The scale of spending is staggering. According to Morgan Stanley, tech firms committed roughly $740 billion to AI-related spending in 2026, a 69% increase from the year before. Meta raised its capital expenditure forecast to between $125 billion and $145 billion. Google boosted its projection to at least $180 billion and signaled it would be "significantly" higher the following year. Gartner forecasts worldwide IT spending will hit $6.31 trillion in 2026, up 13.5% from 2025. McKinsey estimates that total AI-related spending, including data centers and IT hardware, could reach $5.2 trillion by 2033. The GPU server market alone is projected to grow from $50 billion today to over $455 billion by 2040. Meanwhile, AI software fees have climbed 20% to 37% over the past year, according to spending management company Tropic. This isn't a one-time infrastructure investment. It's a recurring and growing operational cost.

The productivity paradox

Here's where it gets uncomfortable. Despite these massive investments, the evidence that AI is actually cheaper than human labor remains thin. A widely cited 2024 MIT study found that AI automation is economically viable in only about 23% of jobs. In the remaining 77%, humans are still the more cost-effective option. That finding sits awkwardly next to the wave of tech layoffs that have swept the industry. Companies are cutting headcount and pouring money into compute, but the math doesn't always support the swap. Keith Lee, an AI and finance professor at the Swiss Institute of Artificial Intelligence's Gordon School of Business, described it as a "short-term mismatch." Companies are betting on a future where compute costs decline and AI capabilities improve, but right now they're paying premium prices for tools that don't yet deliver premium economics.

Why companies keep spending anyway

If AI compute costs more than employees, why are companies accelerating their investments? A few reasons stand out. First, there's a genuine belief that costs will come down. Hardware gets cheaper over time, models get more efficient, and inference costs have already dropped dramatically over the past two years. Companies are positioning themselves for a future where today's expensive bets become tomorrow's competitive advantages. Second, there's a fear of falling behind. Mark Zuckerberg captured this sentiment on a recent earnings call: "Every sign that we're seeing in our own work and across the industry gives us confidence in this investment." When your competitors are spending hundreds of billions, standing still feels riskier than overspending. Third, AI spending has become decoupled from traditional ROI calculations. S&P Global research found that roughly 80% of the growth in final private domestic demand in the first half of 2025 came from AI data centers and high-tech spending. Strip out AI, and U.S. business investment was actually declining. AI spending isn't just a corporate strategy, it's propping up parts of the broader economy.

The hidden cost structure

What makes AI compute costs particularly tricky is their variable nature. Traditional employees have predictable costs: salary, benefits, overhead. AI compute, on the other hand, scales with usage in ways that are hard to predict and even harder to control. Token costs, the fees charged every time an AI model processes or generates text, are the new line item that's catching CFOs off guard. An engineer using an AI coding assistant might generate thousands of dollars in token costs per month without anyone noticing until the bill arrives. Multiply that across an engineering organization and you're looking at compute budgets that rival or exceed payroll. This is pushing companies to rethink how they manage AI spending. Compute is starting to resemble a core operating expense, subject to the same kind of cost controls as headcount. Some firms are already implementing usage caps, choosing cheaper models for routine tasks, and evaluating AI providers based on token efficiency rather than raw capability.

Complements, not substitutes

The most interesting shift happening right now is in how companies frame AI's role. The early narrative was straightforward: AI replaces workers, companies save money. The emerging reality is more nuanced. Some firms are beginning to reevaluate AI not as a clear cost-saving substitute for labor, but as a complementary tool. The value isn't in replacing a $150,000 engineer with $28,000 in monthly compute. It's in making that engineer dramatically more productive, able to ship in days what used to take weeks. This reframing matters because it changes the math entirely. If AI is a complement, then the relevant question isn't "is compute cheaper than people?" It's "does the combination of people plus compute produce enough additional value to justify the total cost?" For many teams, the answer is clearly yes. For others, especially those that scaled AI usage fast without strong limits, it's becoming a balance-sheet headache.

What this means going forward

We're in a peculiar moment. The companies building AI infrastructure are openly acknowledging that it costs more than the people using it. Yet investment is accelerating, not slowing down. This can't last forever in its current form. Either compute costs need to come down significantly, AI needs to deliver measurably higher productivity to justify the premium, or companies will start pulling back. The most likely outcome is some combination of all three. In the meantime, the smartest companies are treating AI budgets with the same rigor they apply to headcount. They're measuring cost per task, tracking token usage, negotiating with providers, and making deliberate choices about which problems are worth throwing compute at. The cost of compute may be far beyond the cost of employees today. But the companies that figure out how to make that equation work, or at least make it sustainable, will be the ones that define the next era of technology.