The $700 Billion Question — What Tech Is Spending in 2026

There is a moment in every major technological transition when the capital being deployed becomes so large it strains ordinary comprehension. The railroad boom of the 19th century had it. The telecom buildout of the late 1990s had it. The internet infrastructure wave of the early 2000s had it.

We are in that moment again — and by most measures, the current AI infrastructure buildout is already the largest technology capital expenditure cycle in history.

The five largest U.S. cloud and AI infrastructure providers — Microsoft, Alphabet, Amazon, Meta, and Oracle — have collectively committed to spending between $660 billion and $700 billion on capital expenditure in 2026 alone. This figure nearly doubles their 2025 spending levels. Following the latest round of earnings calls in late April 2026, Wall Street analysts from Evercore and Bank of America revised their 2027 estimates above $1 trillion for the first time.

To understand what is happening, you need to understand why.

Every major hyperscaler is operating from the same strategic premise: AI workloads will consume every unit of compute capacity that can be brought online, and the company that falls behind on infrastructure will fall behind on AI capability, cloud revenue, and competitive positioning — with consequences that could take a decade to recover from.

The result is a collective action dynamic that is self-reinforcing. As Amazon announces $200 billion in annual capex, Google responds with $185 billion, Microsoft commits $190 billion, and Meta raises its guidance to $135 billion. Each new announcement sets a new floor for the others. All of them report that their markets are supply-constrained rather than demand-constrained — meaning they could sell more capacity than they can build.

Amazon CEO Andy Jassy defended his company's $200 billion commitment — the largest of any company in the group and substantially above consensus expectations that had been around $147 billion — by pointing out that AI capacity is being monetized as quickly as it is installed. AWS reached a $142 billion annualized revenue run rate with growth accelerating. Google Cloud grew 63% year over year to $20 billion in revenue in its latest quarter, with a cloud backlog of $462 billion — a figure that nearly doubled in a single quarter.

The breakdown of capex spending reveals the full supply chain being built.

Roughly two-thirds of hyperscaler capex is going directly to GPUs, CPUs, and AI accelerators — predominantly Nvidia chips, though custom silicon from companies like Google's TPUs and Amazon's Trainium is growing. The remaining third goes to data center construction, networking infrastructure, cooling systems, and power infrastructure. This is not abstract spending — it is physical infrastructure being built at a pace the global construction and manufacturing supply chain is straining to keep up with.

The component pricing environment has also become a factor. Multiple hyperscalers flagged that a portion of their capex increases in 2026 reflect higher component prices — particularly memory pricing, which Meta CEO Mark Zuckerberg specifically cited. Microsoft attributed approximately $25 billion of its capex increase to higher component pricing. This creates a secondary dynamic where the buildout itself is inflating the cost of the buildout.

Beyond the hyperscaler headlines, global IT spending is expected to grow 9.8% to more than $6 trillion in 2026. This broader number is significant because it establishes a floor beneath the more speculative AI bets. Even if specific AI applications disappoint, the infrastructure layer keeps expanding — data centers need chips, clouds need capacity, and software needs platforms regardless of which AI use cases ultimately win.

The comparison to historical technology investment cycles is instructive. AI capex currently equates to roughly 0.8% of GDP. During the telecom infrastructure boom of the late 1990s, infrastructure spending reached peaks above 1.5% of GDP — a figure that, applied to today's economy, would imply potential annual capex well above what is currently being spent. Goldman Sachs has noted that by this comparison, the current cycle has substantial runway remaining.

The question that investors are asking more aggressively in 2026 is when the spending translates into commensurate returns. The short answer is that evidence is beginning to emerge — but it is uneven, and markets are punishing companies that can't demonstrate it clearly.

Alphabet was the clearest winner in the most recent earnings cycle. Google Cloud's 63% revenue growth, accelerating enterprise adoption of Gemini models, and a massive backlog gave investors a concrete narrative connecting capex to revenue. Alphabet's stock jumped roughly 10% on its results and had its best month since 2004 in April 2026.

Meta faced more skepticism. Unlike the hyperscalers with cloud businesses that directly monetize compute capacity, Meta's AI spending is directed at internal products — advertising systems, content recommendation, and its open-source Llama model ecosystem. The connection between $135 billion in capex and incremental advertising revenue is harder to model, and JPMorgan downgraded the stock following its earnings call, citing a challenging path to generating returns.

Microsoft, meanwhile, guided that it expects to remain capacity-constrained through 2026 even as it increases spending — a signal that demand is real, but that the company is not yet generating the margin improvement some investors had hoped to see.

Every dollar of hyperscaler capex flows through a supply chain that extends far beyond the headline names. Nvidia is the most visible beneficiary, but the buildout requires an ecosystem: custom chip designers like Broadcom, whose AI semiconductor revenue is forecast to double year over year; semiconductor fabrication at TSMC; server manufacturing at companies like Quanta Computer; power management from Eaton; data center cooling from companies like Vertiv; and networking from Arista and others.

The Stargate project — a $500 billion infrastructure initiative involving OpenAI, SoftBank, and Oracle — is the clearest expression of how the buildout is scaling beyond individual company balance sheets into multi-party, government-adjacent infrastructure investments.

The $700 billion AI capex wave is the defining capital allocation story of our era — comparable in scale and consequence to the railroad infrastructure buildout that defined the 19th century American economy. Like that era, the infrastructure itself is likely to be durable and transformative regardless of which specific companies and applications emerge as winners. And like that era, the companies supplying the picks and shovels — chips, cooling, power, networking — may end up among the most consistently profitable players across the full cycle.

The risk is real: history shows that technology infrastructure buildouts often overshoot near-term demand before demand catches up. The late 1990s telecom buildout left mountains of stranded fiber capacity. But the secular demand for AI compute appears more demand-driven than that cycle — every enterprise, every government, and every individual consumer is now a potential customer for AI capabilities. The question is timing, not direction.

The money is being spent. The infrastructure is being built. The only debate is whether the returns arrive fast enough to justify the investment — and on that, Q1 2026 earnings were more encouraging than skeptics expected.

Educational content only. Not financial advice. Brezco Analytics is an independent research and media platform.

Sources: