The trillion-dollar arms race
Five technology giants are about to dump over a trillion dollars into AI infrastructure this year. It's not just an eye-watering figure. It's a collective capital expenditure binge that dwarfs the output of entire nations, and the people holding the checkbooks at the Bank for International Settlements (BIS) are starting to sound terrified. In its 2026 annual report, the BIS compared this frenzy to historical investment manias that ended with sudden reversals and economy-wide recessions.
The scale of this building boom is hard to overstate. According to forecasts, Amazon plans to spend $200 billion in 2026, Microsoft is projecting $190 billion, Google is earmarking $180 billion, and Meta is sitting at up to $140 billion. When you add Oracle's aggressive cloud buildout into the mix, the collective spending for these top five hyperscalers is set to exceed $1 trillion. Hyperscalers, as defined by industry terms, are the massive cloud companies that own and operate the data center networks underpinning the modern web, as detailed in recent industry listings.
This investment is outpacing the current earnings and free cash flow of these firms. To keep up, they're issuing debt. They're driven by a winner-take-all mentality, betting that only a tiny handful of dominant players with superior technology will control the future market share. This forces everyone to overcommit to speculative projects with highly uncertain returns. As spending climbs, the net economic surplus for the whole industry falls, and it could slip into negative territory if returns disappoint.
History repeating itself
Throughout history, technological breakthroughs have triggered manic investment phases. The BIS report draws direct parallels to the canal and British railway mania of the 1800s, the electrification exuberance of the 1920s, and the dot-com boom of the late 1990s.
These episodes shared a common cycle: a genuine breakthrough attracted capital far in excess of what commercial returns could support. Investors bought into a future that couldn't possibly pay for itself in the short term. When returns fell short, financing pulled back suddenly, turning the capex boom into a protracted investment bust. Safe to say, the economic fallout didn't stay confined to tech; it sparked broader recessions.
As a labor economist, I see the same capital-labor misalignment happening today. Silicon Valley is pouring hundreds of billions into physical assets—silicon chips, fiber optic cables, and massive server farms. They're betting these physical assets will run the workplace of tomorrow without humans. But they're ignoring the reality of how work gets done. They're building a massive technological superstructure without a stable economic foundation. When the capital flows reverse, it's the workers and suppliers who will pay the price.
Circular financing and rising debt
The fragility is magnified by how opaque this boom actually is. The BIS report warned about the lack of transparency, pointing to circular financing deals and undisclosed datacenter lease terms that hide the real risks.
We see this paper game everywhere. A tech giant invests in a hot AI startup, and the startup immediately turns around and spends that cash to lease compute power back from the investor's cloud footprint. On paper, it looks like revenue growth. In reality, no new value has been created. This dynamic mirrors the warnings we saw when Satya Nadella spoke about soaring costs hollowing out corporate margins as they chase the hype.
To fund these massive buildouts, some tech giants are cutting human operations. They're laying off workers to direct more dollars toward microchips. This is the exact paradox we saw when Oracle cut 21,000 jobs to fund its massive buildout. They're trading human capital, which generates steady operational value, for depreciating hardware assets that have yet to prove their utility. It's a bad trade, and it's unsustainable.
Roadblocks on the energy grid
Physical constraints are also waiting to derail the train. The BIS highlights a growing list of supply-side roadblocks that are driving up costs.
Electricity availability is a major bottleneck. Data centers are putting heavy pressure on power grids and energy prices, creating inflationary pressures that spill over into other sectors. Combine that with grid connection delays, chip shortages, and rising memory prices, and the operational costs become staggering. To lock in capacity, firms are signing long-dated contracts for power and space. But these long-term commitments lock them in; if demand doesn't materialize, they'll be stuck with billions in liabilities.
The risk also extends to their supplier ecosystem. The contractors building these facilities—engineering and construction firms—have weak balance sheets compared to the hyperscalers. They're completely exposed to a capital pullback. If a tech giant freezes construction next month, these builders go under first.
The empty promise of productivity
The big promise of this infrastructure boom is productivity. Yet, the returns aren't showing up.
Enterprises running pilots report minor efficiency gains at the individual employee level—like writing emails slightly faster. But very few report discernible productivity gains once projects enter production at scale. The gains are small, while the costs are enormous.
This mismatch is the core of the reverse centaur paradigm. Instead of using AI to augment human creativity, companies are using it to monitor workers and turn them into disposable appendages to a machine. This doesn't drive genuine growth; it makes the labor force more fragile.
If returns don't path up, the investment bust will be painful. With central banks keeping policy rates tight to fight inflation, a sudden pivot in sentiment could trigger sharp asset price drops. The resulting debt-fueled feedback loops could trigger a severe recession, similar to the tech stock volatility we saw after the recent Oracle market recoil. When the bubble pops, the fallout will hit the credit markets, the supply chain, and ultimately the global workforce.
