The AI Buildout: What’s Actually Wrong With It — And the Case for the Other Side
The numbers behind the AI infrastructure buildout are hard to process at a human scale. Goldman Sachs now estimates the total capex for compute, data centers, and power will run to roughly $7.6 trillion between 2026 and 2031. Hyperscalers alone committed over $750 billion for 2026 and are on pace for $900 billion in 2027, a spending curve that, as a share of GDP, has already eclipsed the dot-com buildout. For advisors, this isn’t an abstract tech-sector story — a meaningful share of client portfolios now sits on the outcome of this bet. It’s worth understanding both what critics get right and where the optimists have a real case.
The disadvantages
The grid can’t keep up. The most concrete bottleneck isn’t chips, it’s power. Of the roughly 16 gigawatts of data center electricity capacity announced for 2026 delivery, only about 5 gigawatts are actually under construction. A quarter of the 140 projects slated for completion by the end of 2026 haven’t even disclosed how they’ll get power. Grid connection wait times in major U.S. markets stretch three to four years, and in Northern Virginia — the densest data center corridor in the country — the wait is now seven years.
Electricity bills are becoming a political issue. Goldman Sachs projects the buildout will push electricity costs up 6% between 2026 and 2027, with another 3% by 2028. A recent YouGov poll found more than two-thirds of Americans expect their bills to rise if a data center is built nearby, and rising community opposition is now a genuine constraint on new projects, not just a PR problem.
The financing is getting creative — and that’s a warning sign. Firms increasingly use asset-based financing to keep this spending off balance sheets. Meta’s $30 billion Hyperion project, for example, keeps only 20% of the cost on Meta’s own books, with the rest sitting in a separate financing vehicle. Man Group has flagged a $2 trillion-plus gap between current funding sources and planned buildout, and researchers at the Bank for International Settlements estimate competitive pressure could be pushing AI investment to 1.5 to 3 times the economically efficient level.
Concentration risk is real. Nearly every leading AI chip in the world is fabricated by a single company, TSMC, on one island. That’s not a US-versus-China issue — it’s a shared structural vulnerability for the entire global AI supply chain.
The counterarguments
Demand may genuinely justify the spend. Morgan Stanley Research forecasts U.S. data center power demand could reach 74 gigawatts by 2028, against a projected shortfall of roughly 49 gigawatts in available power access — a gap, not a glut. From that vantage point, today’s investment looks less like speculation and more like catching up to real, underserved demand.
Data centers haven’t actually raised electricity costs yet. A working paper from the Electric Power Research Institute found that through at least 2024, data center operations were associated with lower, not higher, retail electricity costs. That complicates the simple “data centers are driving up your bill” narrative, even as forward-looking projections point the other way.
Adoption keeps climbing regardless of the debate. McKinsey’s State of AI research finds 88% of organizations now use AI in at least one business function, up from just 20% in 2017, and 78% of Fortune 500 companies run active generative AI initiatives, per the Stanford AI Index. Usage this broad doesn’t guarantee the capex pays off, but it does undercut the argument that this is hype with no underlying adoption.
What this means for client conversations
Advisors don’t need to resolve the bubble debate to be useful here. What clients actually need is help understanding concentration risk in portfolios that have become quietly AI-heavy through index funds and mega-cap tech exposure, and a clear-eyed view that “the technology is real” and “the financing structure is sustainable” are two separate questions. The buildout may well justify itself over a decade. Whether the way it’s being financed survives the next earnings cycle is a different, and much nearer-term, question worth raising before clients ask it first.