Big technology companies are on a scale-up tear. Estimates for 2026 put combined AI-related spending on compute, data centres and supporting infrastructure at roughly $650–660 billion — a sum that feels both breathtaking and inevitable. That figure, splashed across headlines, has revived talk of an AI bubble. But the reality on the ground looks messier than a single label.
Where the money is going
A huge slice of that spending is tangible: racks of GPUs, sprawling data‑centre campuses, substations and miles of fibre. Hyperscalers are writing big cheques to lock in capacity and keep pace with ever-larger models. Those projects are producing concrete work: girders, switchgear, high-voltage connections and long-term service contracts. In some corners the buildout takes on imaginative forms — from suburban mega‑campuses to experimental ideas about off‑planet capacity like Google’s Project Suncatcher.
The construction and power needs are not hypothetical. Analysts point to an emerging “electron gap” — a shortfall in electrical generation and distribution capacity measured in the tens of gigawatts — and to many billions earmarked for grid upgrades. Utilities and contractors are already retooling: M&A is accelerating among electrical firms, and owners are chasing recurring maintenance revenue as much as one‑off build fees.
Bubble talk — and why some experts push back
It’s easy to evoke the dot‑com bust. Back then, speculative companies sold visions without supply chains or customers. Today’s AI spending is backed by operating businesses and physical assets. That matters. A data centre built for AI still generates decades of revenue streams: colocation fees, cloud services, and ongoing upkeep.
Still, real assets don’t immunize the market against mispricing. Investors are asking a simple question: when will these investments pay back? The answer is uneven. Some companies are starting to show productivity gains and clearer monetization paths; others are still burning cash while investors demand evidence of durable revenue lifts.
Market reactions: haves and have‑nots
Wall Street’s response has been distinctly bifurcated. Firms that can point to near‑term AI monetization — targeted ad improvements, higher ARPU, or concrete enterprise wins — have seen share rebounds. Others, notably some cloud and enterprise software players, have been punished after earnings that failed to show faster cloud growth or convincing returns on huge AI bills.
That dynamic is creating winners and losers in stock performance, and pushing analysts to separate real adoption from hype. Some argue we’re still in “phase zero” of enterprise AI, where the plumbing (models, data platforms, observability) is the place to look for durable upside. Meanwhile, companies rushing to build proprietary multimodal models or specialised image-generation engines — such as Microsoft’s in-house work on MAI image models — are also inflaming demand for more compute and custom infrastructure (Microsoft MAI‑Image‑1).
Who stands to gain (and how)
Contractors and infrastructure providers are a clear near‑term beneficiary. Data centres require complex electrical systems, continuous maintenance and long lifecycle services — a steady revenue stream once a site is live. That has sent private‑equity and strategic buyers hunting for established contractors with data‑centre pedigrees. Grid modernization, pegged in many forecasts at over a trillion in spending through the decade, broadens the opportunity beyond hyperscalers and into transmission, distribution and control systems.
Software vendors and platform players also have opportunities, but they must solve hard problems around data governance, security and integration. Tools that let enterprises surface insights from their files and mail — the sort of features that Google is rolling into Workspace with deeper Gemini integrations — can convert AI spending into measurable productivity gains (Gemini Deep Research). If customers see efficiency and measurable ROI, the buildouts begin to make clearer financial sense.
The downside scenarios
Oversupply is one risk. If multiple providers overbuild capacity in the same regions, pricing pressure could follow. Another risk is misplaced investment — money spent on specialised hardware or campuses that become obsolete as model architectures and compute needs evolve. Finally, macro shocks — higher interest rates or a sharp slowdown in enterprise IT budgets — could extend the payback horizon and test investor patience.
The conversation doesn’t end with “bubble” or “no bubble.” It’s about time horizons and optionality. Some bets are on infrastructure that will pay over decades; others are on speculative growth that must prove itself in quarters.
If there’s a throughline, it’s simple: AI is forcing a capital‑intensive reconfiguration of computing and power. That reshuffle will create winners — contractors, utilities, cloud builders and a subset of software companies — and leave others to trade on narratives rather than cash flows. For now, money is still moving fast. The question for investors, contractors and policy makers is whether they’ll be building for a decade of demand or for a shorter, fevered sprint.