Seeing the data organized by JPMorgan, the four cloud giants—Microsoft, Oracle, Google, and Amazon—now have accumulated outstanding orders worth $2.1 trillion. Even more interestingly, the compute capacity procurement commitments from the two AI labs, OpenAI and Anthropic, account for almost $1 trillion—close to half.
This structure tells the story well: the rate and scale at which cutting-edge AI labs are burning money have reached a level that can support roughly half of cloud providers’ business. From a capital allocation perspective, this is a classic cyclical bet—betting on whether the AGI narrative can deliver commercial value in the coming years.
Historically, there have been many similar cases: fiber network buildouts in the late 1990s, telecom equipment procurement in the early 2000s, and shale-oil capital expenditures in the 2010s. Each time involves massive upfront investment, with the bet that future demand will be able to absorb the capacity. Some turned out well; others left behind excess capacity and bad debts.
In essence, this $1 trillion in compute orders is about locking in cash flows from the next few years in advance. For cloud providers, it means stable revenue expectations. For AI labs, it creates the pressure to get their business model to work. Whoever is first to roll out large-scale, real-world applications can absorb these costs; those who can’t, face a classic case of capital misallocation.
From the macro cycle perspective, this concentration of capex is high (the two labs account for half), and the risk isn’t small. If the speed of AI monetization falls short of expectations, or if financing issues arise for one of the labs, the entire chain’s cash flow and valuation logic will be reassessed. This isn’t a technical problem—it’s a matter of the capital cycle and liquidity management.
As the old saying goes: order backlogs are a good thing—provided customers can keep paying. In this game, the one who can first turn AI applications into real cash wins.
This structure tells the story well: the rate and scale at which cutting-edge AI labs are burning money have reached a level that can support roughly half of cloud providers’ business. From a capital allocation perspective, this is a classic cyclical bet—betting on whether the AGI narrative can deliver commercial value in the coming years.
Historically, there have been many similar cases: fiber network buildouts in the late 1990s, telecom equipment procurement in the early 2000s, and shale-oil capital expenditures in the 2010s. Each time involves massive upfront investment, with the bet that future demand will be able to absorb the capacity. Some turned out well; others left behind excess capacity and bad debts.
In essence, this $1 trillion in compute orders is about locking in cash flows from the next few years in advance. For cloud providers, it means stable revenue expectations. For AI labs, it creates the pressure to get their business model to work. Whoever is first to roll out large-scale, real-world applications can absorb these costs; those who can’t, face a classic case of capital misallocation.
From the macro cycle perspective, this concentration of capex is high (the two labs account for half), and the risk isn’t small. If the speed of AI monetization falls short of expectations, or if financing issues arise for one of the labs, the entire chain’s cash flow and valuation logic will be reassessed. This isn’t a technical problem—it’s a matter of the capital cycle and liquidity management.
As the old saying goes: order backlogs are a good thing—provided customers can keep paying. In this game, the one who can first turn AI applications into real cash wins.