The four largest US technology companies have locked in nearly $2.4 trillion in purchase commitments, contractual obligations, and leases tied to artificial intelligence, according to new reporting. Alphabet leads the group with roughly $900 billion in total commitments, a figure about nine times higher than a year earlier. Meta follows with around $700 billion in future spending, while Amazon has set aside $220 billion in capital expenditure for this year alone. Microsoft’s own obligations also contribute to the total, making this the largest wave of technology infrastructure spending ever recorded by the private sector.
These numbers go far beyond ordinary annual capital expenditure. They include long-term purchase orders for advanced chips, power-purchase agreements, and data-center leases that stretch decades into the future. Much of this financial weight currently sits in footnotes rather than balance sheets, but it represents real obligations that will eventually come due. About half of Meta’s total is tied to data-center leases running as long as 30 years, roughly eight times what the company carried only a year ago. Alphabet’s commitments have grown at a similar pace, with much of the expansion structured through off-balance-sheet vehicles that keep headline debt figures looking tidy.
Cash flow pressure intensifies
Two of the four giants are already feeling the strain. Alphabet reported negative free cash flow of nearly $6 billion in the second quarter, its first quarterly outflow in close to two decades. The company continues to generate substantial operating profit, but the pace of AI infrastructure spending has overwhelmed that cash generation. Amazon’s free cash flow swung to negative territory on a trailing-twelve-month basis earlier this year, and its second-quarter results showed the cash position remains under pressure even as AWS revenue grew 37 percent.
Meta is widely expected to follow a similar path. The company narrowed its 2026 capital expenditure forecast to between $130 billion and $145 billion by raising the low end of the range. In the second quarter, Meta’s free cash flow fell 91 percent year over year. The pattern across all four companies is consistent: revenue is still growing, but capital spending is growing faster, and the gap first shows up in cash flow.
A familiar analogy with an unfamiliar scale
Amazon CEO Andy Jassy recently compared the current moment to the first AWS build-out, when the company spent years investing before demand caught up. The analogy is intended to reassure investors that the spending will eventually pay for itself, much as AWS grew from a side project into a business generating more than $40 billion per quarter. The difference is one of scale. The original AWS build-out cost a fraction of what the industry is now committing in a single year, and the infrastructure being constructed today is far more specialized, expensive, and energy-intensive than standard cloud data centers.
The AI boom has created a unique set of demands. Training frontier-level models requires clusters of tens of thousands of accelerators, each consuming large amounts of electricity and generating enormous heat. Deploying these models at scale requires additional computing capacity around the world. That has led hyperscalers to sign long-term contracts for nuclear, solar, wind, and natural-gas power, as well as for next-generation memory, networking gear, and cooling systems. These are not optional purchases; they are commitments that may take years to be fully utilized.
Demand signals remain strong
The question hanging over earnings season is whether demand is keeping pace. AWS answered in the affirmative this week with its fastest growth in four years, while Alphabet’s cloud unit grew 82 percent in the second quarter. Microsoft also reported strong cloud and AI services demand in recent results, with Azure growth accelerating as customers move from experimentation to production workloads. Enterprise buyers are increasingly treating AI capacity as a strategic resource, similar to how they once moved to cloud databases and productivity suites.
Still, the combined capital expenditure across the largest cloud operators is on track to overtake the cash their core businesses generate, a threshold the industry has never crossed before. That means the companies are effectively borrowing from future cash flows, or from debt markets, to fund projects whose returns are not yet proven. The recent wave of AI infrastructure is not limited to the four companies mentioned. Oracle, OpenAI, and various data-center REITs have announced massive construction plans, and governments in the US, Europe, and the Middle East have begun subsidizing or facilitating new facilities.
How accounting smooths the cost
For now, all four companies remain profitable because capital spending is depreciated over years rather than booked as an immediate expense. When a company buys a $1 billion server farm, it does not subtract $1 billion from profit in that quarter. Instead, it spreads the cost over the useful life of the equipment, typically three to six years for servers and much longer for buildings and land improvements. This accounting treatment smooths the earnings impact and gives executives room to frame the investments as forward-looking bets rather than current losses.
But depreciation does not eliminate the cash outflow. The cash is spent when the equipment is purchased or when lease payments come due. Investors have begun paying closer attention to free cash flow and to the total commitments disclosed in financial statements for this reason. Alphabet’s negative free cash flow was an important symbol even if the company remained profitable on an accrual basis. Meta’s 91 percent drop in free cash flow was another warning sign. These numbers reveal that the scale of betting is not merely a line item in a capital budget; it is beginning to change the financial profile of the most valuable companies in the world.
A decade of infrastructure in a few years
The commitments announced to date imply that hyperscalers are trying to build a decade’s worth of data-center capacity in a few years. Supply chains for advanced chips, particularly graphics processing units from Nvidia and custom accelerators from Google, Amazon, and Microsoft, are stretched. Power grids are becoming a bottleneck, with some projects waiting years for grid connections. In response, companies are co-locating facilities near substations, building their own renewable generation, and even negotiating long-term agreements with nuclear plant operators.
There is also a competitive dynamic at play. No company wants to fall behind in AI infrastructure because the perceived cost of under-building is higher than the cost of over-building. If demand continues to grow, early movers will have an advantage in training models and serving customers. If demand does not materialize as expected, the surplus capacity could be repurposed for cloud computing or other workloads. But the sheer size of the commitments makes repurposing less straightforward than it was in previous cycles.
The $2.4 trillion in commitments is real money that will come due whether AI revenue materializes at the scale these companies are betting on or not. Some of those obligations can be renegotiated, and some lease agreements include termination options, but the broad trajectory is fixed. The companies have essentially placed a massive wager on the future of artificial intelligence. They are doing so with the confidence of firms that have seen previous technology bets pay off. The difference is that the current bet is larger than any before it, and the timeline for returns remains uncertain.
The next few quarters will be watched closely. Free cash flow trends, data-center utilization rates, AI revenue disclosures, and management commentary will all provide clues about whether the spending is rational or euphoric. What is already clear is that the biggest technology companies have committed an extraordinary portion of their future cash flows to one technology, and the consequences of that choice will shape the industry for years to come.
Source: TNW | Amazon News