Inflationary Spiral of AI Infrastructure: Behind the Surge in Capital Expenditure of Tech Giants

When Capital Expenditure Becomes an Inflation Indicator

With the upcoming earnings season for big tech companies starting next week, Wall Street's focus is no longer on profits or revenue—but on how much these giants are willing to spend on AI data centers. Google, Amazon, Microsoft, and Meta have already announced total spending this year will exceed $700 billion, but a thorny question is emerging: how much computing power can this money actually buy?

The answer is not optimistic. Memory chip prices have surged, power equipment, building materials, and skilled labor have become increasingly difficult to obtain, and grid access is more expensive than ever. Morgan Stanley estimates that the cost of building 1 gigawatt of AI capacity has risen by about 20%. For a common NVIDIA architecture system, the cost per gigawatt has climbed from about $29 billion to $35 billion; for a newer version, the cost has risen from $41 billion to $49 billion.

This creates a dangerous cycle: tech giants order more AI data center equipment → shortages intensify → prices rise → companies are forced to raise spending forecasts → further stimulates demand → prices continue to climb. Brad Gastwirth, head of research at Circular Technology, estimates that about 20% to 30% of the next AI capital expenditure increase will come from inflation, with 70% to 80% still being real expansion. But the problem is that higher spending does not always mean faster construction.

The Inflation Component Cannot Be Ignored

Investors need to be wary of the "water" in capital expenditure growth. A previous RBC study found that this year's memory price surge may explain about 45% of the capital expenditure growth of major cloud companies. Cantor Fitzgerald analysts expect that the established spending plans for 2026 will see little change this quarter, but forecasts for 2027 will surge significantly: Google at $283 billion, Amazon at $271 billion, and Meta at $200 billion.

No one wants to appear cautious in this AI race, so no company will actively cut spending for now. However, Gastwirth suggests that when executives discuss capital expenditure on earnings calls, one should listen carefully to whether they simultaneously mention metrics such as power capacity, GPU deployment, memory procurement, network construction, and new data center campuses. "If capital expenditure rises in tandem with these metrics, it is genuine expansion; otherwise, a bigger number may just be tech companies paying more to stay in the same place."

The True Cost of the Computing Power Arms Race

This capital expenditure spiral essentially reflects the imbalance between supply and demand in AI infrastructure. On one hand, competition among large models forces tech companies to hoard computing power at any cost; on the other hand, supply chain bottlenecks—especially in memory and power—constrain actual delivery capacity. While NVIDIA's GPUs remain in short supply, other links in data center construction—from transformers to cooling systems—are becoming new choke points.More profound is that the relentless rise in capital expenditure is reshaping the financial structure of tech giants. When annual spending on the scale of $700 billion becomes the norm, these companies must find new revenue streams to justify the investment. AI cloud services, API calls, and enterprise subscriptions are the focal points, but in the short term, it is difficult to cover the high hardware costs.

Outlook: When Will the Spiral Peak?

In the short term, there is no sign of slowing capital expenditure. As long as AI model parameters grow several times annually, the demand for computing power for training and inference remains insatiable. However, cost pressures may force structural adjustments in the industry: for instance, more efficient chip architectures, denser liquid cooling solutions, or reducing computing power demands through software optimization.

For investors, the upcoming earnings season will be a critical verification point. If the giants can provide specific infrastructure metrics (rather than just dollar amounts), the market may be able to more clearly distinguish real growth from inflationary bubbles. Otherwise, this capital expenditure spiral could ultimately end in the classic pattern of "overinvestment → overcapacity → value destruction."