The AI buildout has an increasingly low-tech bottleneck: people. Reuters Breakingviews reported on September 14 that the U.S. construction industry faces a shortfall of roughly 439,000 workers as data centers compete with housing, transport and industrial projects for skilled labor.
Amazon.com, Inc. (NASDAQ:AMZN) and Microsoft Corporation (NASDAQ:MSFT) can finance multibillion-dollar AI campuses and secure GPUs. Neither can conjure electricians, welders and specialized contractors on demand.
Balance sheets solve the money problem, not the scheduling problem
For Amazon.com, Inc., purchasing power is the advantage. AWS can sign long-term equipment and construction contracts, pay for scarce labor and spread infrastructure across regions. Amazon’s scale also lets it fund training pipelines and repeat standardized data-center designs.
A new data center facility in Plano, Texas. Photo from CoreWeave
Paying more, however, does not create unlimited labor supply. Data-center construction wages can pull workers from other projects, while local housing shortages constrain labor mobility and permitting delays can stretch project timelines. Higher wages and slower completion push financing and carrying costs higher while delaying the point at which committed capital begins generating cloud revenue. That timing gap matters because land, equipment and financing commitments can start costing money well before a delayed campus begins producing cloud revenue.
Microsoft Corporation faces the same physical constraint but has a large commercial backlog and Azure demand waiting for capacity. Getting data centers online faster can convert that backlog into revenue. Money does not solve the physical bottleneck: customers cannot consume capacity that is still a construction site. Cost overruns and delays can weaken returns on capital even if end demand remains strong.
Professional investors still crowd both hyperscalers
Insider Monkey’s database counted 369 hedge funds holding Amazon in Q2 2026, up from 353 in Q1. Arrowstreet Capital increased its position 24% to 40,346,851 shares. Microsoft slipped to 273 holders from 282, although Arrowstreet increased its Microsoft stake 14% to 27,659,541 shares. Those filings predate the latest construction-shortage debate.
As of August 31, 74,451,155 Microsoft shares were sold short, equal to 1.00% of float, with 3.18 days to cover. The low percentage simply describes outstanding positioning.
This bottleneck cuts both ways. A sustained AI construction boom can pull more workers into higher-paying skilled trades and eventually expand supply. Before that adjustment occurs, Amazon and Microsoft are competing for labor in the same way they compete for chips and power. Investors who model AI capex only as dollars and GPUs are missing a crucial variable: how fast physical capacity can actually be built.
READ NEXT:
The Trillion-Dollar AI Capex Gap: Why Too Much Hardware Could Be Nvidia’s Trap and Microsoft’s Opportunity and Redditors Are Skeptical of SpaceX’s Orbital Data Centers. Are They Right?
Hedge Fund Resource Center