Big tech's $650bn roll of the dice on artificial intelligence

The four largest US technology companies will collectively spend roughly $650 billion on capital expenditures in 2026, almost entirely on data centres and AI infrastructure. The figure, compiled from recent earnings forecasts, marks a sharp acceleration in a spending cycle that is now entering its third year, with no clear signs of slowing.

The scale is staggering. To put it in perspective, $650 billion exceeds the GDP of most countries. And unlike previous investment cycles in tech, this one is being driven not by proven revenue streams but by a shared conviction that artificial intelligence will reshape the global economy. The question investors are now asking is whether this conviction will be vindicated before the bills come due.

The rebuild of the digital infrastructure layer

Lei Qiu, CIO of thematic innovation equities at AllianceBernstein, described the spending as a wholesale "rebuild of the digital infrastructure layer." Speaking on Bloomberg Technology, Qiu argued that the investment is not simply about chasing AI hype. Large-cap companies are spending aggressively for two reasons: growth and defence.

On the growth side, companies see AI as a generational platform shift, similar to the rise of cloud computing but potentially far larger in scope. On the defensive side, they fear that failing to invest now will leave them exposed to disruption from faster-moving competitors, including well-funded startups that are scaling at extraordinary speed.

Some private AI companies have reached $100 million in revenue in a fraction of the time it took their predecessors. The problem, Qiu noted, is that reaching that milestone quickly tells you very little about whether a company can sustain it. The velocity of growth in AI is real, but so is the velocity of disruption.

Early signs of payoff in unexpected places

There is evidence, albeit early, that AI investment is producing tangible returns. Qiu pointed to productivity gains in traditional manufacturing, where AI tools are already being used to optimise processes, reduce waste and accelerate decision-making. Retail is another sector showing early results.

These are not the headline-grabbing applications that dominate conference keynotes. They are quieter, more incremental, and arguably more significant. If AI can deliver meaningful productivity improvements in sectors that account for large portions of GDP, the case for $650 billion in infrastructure spending becomes much easier to make.

The challenge is timing. Productivity gains in manufacturing do not show up in quarterly earnings the way a new subscription product might. Investors looking for rapid proof of return on investment may not have the patience to wait for these slower-burn payoffs to materialise.

The terminal value problem

The AI spending surge is forcing a fundamental rethink of how investors value technology companies. According to Qiu, the traditional approach of projecting terminal revenue and applying a terminal multiple is breaking down.

The reason is straightforward: the probability that any given company can hold onto its revenue and profit over a five or 10-year horizon is changing rapidly. When a startup can appear from nowhere, raise billions and capture market share in months, the moat that once justified premium valuations looks far less durable.

This is showing up in compressed multiples. Investors are paying less for future earnings because the confidence in those earnings persisting has diminished. It is a rational response to a market where the competitive dynamics are genuinely uncertain.

A market caught between fear and opportunity

The recent sell-off in tech stocks reflected this tension. On one hand, $650 billion in annual spending with limited near-term revenue to show for it looks like a classic case of over-investment. On the other, the companies making these bets are among the most profitable in history, with balance sheets that can absorb years of heavy spending.

Qiu argued that concerns about return on investment have actually decreased as the evidence of AI's utility has grown. The worry is no longer whether AI works. It is whether the companies spending the most will be the ones that capture the value.

For investors with a long-term horizon, the current market dislocation may represent an opportunity. Valuations in some corners of the AI ecosystem have been marked down significantly, even as the underlying technology continues to improve at a rapid pace. New model releases, including Anthropic's Claude Opus 4.6, are pushing the boundaries of what AI can do in areas like financial research and complex reasoning.

The $650 billion question no one can answer yet

The honest assessment is that nobody knows whether this investment cycle will produce returns commensurate with its scale. History offers mixed precedents. The fibre-optic buildout of the late 1990s bankrupted the companies that laid the cables but created the infrastructure that made the modern internet possible. The cloud computing boom of the 2010s rewarded patient investors handsomely but punished those who bet on the wrong platforms.

AI's trajectory could follow either path, or chart an entirely new one. What is clear is that the largest technology companies in the world have made their bet. Six hundred and fifty billion dollars says they believe this road leads somewhere. The next 12 months will start to reveal whether they are right.