Unprompted: The Hidden Cost of Being Early

Unprompted is an occasional opinion column from Kunal Gupta for Pivot 5 readers.

The Hidden Cost of Being Early

Over the past week, I listened to each of the earnings calls from Microsoft, Meta, Google, and Amazon. What I was most curious about was not how they were making money, but how they were planning to spend it.

Capital expenditures at this scale are rarely an accident. When companies commit tens of billions of dollars to infrastructure, it is a clear indication of where they believe the world is headed.

Amazon, Microsoft, Google, and Meta are increasing their capex spending at an unprecedented rate, pouring over $300 billion into AI-related infrastructure investments this year alone. Compared to 2023, this represents a 50% increase—not just incremental growth, but acceleration. Acceleration suggests urgency.

The stock market punished them for missing their cloud revenue targets, interpreting it as a sign of insufficient demand. In reality, the opposite is true. Demand is outpacing supply, and their capacity constraints are forcing them to leave money on the table.

These companies are not investing in the hope that AI will take off. AI has already taken off. They are investing because, even at this early stage, they cannot meet demand fast enough.

The models themselves are becoming cheaper and more widely available, as seen with DeepSeek’s recent release. The technology is proliferating. But the true bottleneck is no longer in the model layer—it is in compute, energy, and data center capacity.

What these companies seem to understand, and what markets may not yet fully appreciate, is that the AI boom is no longer about improving models. It is about securing control over the scarce infrastructure that those models depend on.

The Business of Scarcity

Scarcity has always been a source of power in business. Owning what is rare—whether land, oil, bandwidth, or distribution—has historically been more valuable than simply competing on a product level.

For years, AI’s scarcity was in the models themselves—the cutting-edge research, the training techniques, the algorithmic breakthroughs. That era is ending. Models are becoming commoditized, available at lower costs and with fewer restrictions. The defensibility of AI is shifting downward into the physical layer of compute.

AI infrastructure is not infinitely scalable in the way software is. The supply of GPUs is constrained. Power grids have limitations. Data centers require real estate, permitting, cooling, and long-term planning. The barriers to entry in AI are no longer about intelligence alone; they are about logistics, supply chains, and physical constraints.

If AI is going to be as transformative as it promises to be, then compute will become a form of leverage—not just determining who can run models, but who can afford to build and deploy them at scale.

The Cost of Being Early

Investing ahead of demand is expensive. The companies leading the AI capex race are absorbing the immediate costs of infrastructure expansion without immediate returns. For public markets, which are often impatient for clear revenue gains, this can look like overreach.

But this is how infrastructure shifts always unfold. Large-scale technological transitions—from railroads to electricity to the cloud—require enormous upfront investment, often before there is clear revenue to justify the spending.

The cost of being early is capital-intensive. It requires patience and, in some cases, invites skepticism. But the cost of being late is fatal.

Amazon’s cloud business was unprofitable for years before it became the backbone of enterprise computing. Meta’s pivot to mobile advertising was initially seen as risky before it became its primary growth engine. Microsoft’s shift to enterprise cloud services required years of investment before it redefined the company’s trajectory.

AI infrastructure is shaping up to be a similar transition. The companies investing now are not just placing bets on the future; they are securing their ability to participate in it at all.

Signals of the Future

The lesson in all of this is not just about AI. It is about how industries evolve, and where power consolidates as they scale.

Scarcity is shifting. The defensibility of AI will not be determined by who builds the best model, but by who owns the infrastructure it runs on.

Capital at this scale is never an accident. The companies pouring billions into AI infrastructure today are not gambling; they are positioning themselves for inevitability.

Most businesses—whether in AI or not—face a similar choice. Invest ahead of the curve, before the returns are obvious, or wait until it is too late to catch up.

History does not reward hesitation. It rewards those who move first.

Kunal

Unprompted is an occasional opinion column I’ll share. Let me know what you think by email: [email protected].