NextFin News - The artificial intelligence revolution is about to become the largest construction project in human history. Global spending on data centers is set to reach $31.6 trillion through 2050 to satisfy the world's appetite for AI, an investment boom with no precedent, according to PwC. If adoption accelerates beyond the firm's central forecast, the total could climb to $50 trillion over the next 25 years. For scale: the figure is larger than the entire annual economic output of the United States, which runs at roughly $30 trillion.
The projection, released this week, reframes the AI buildout not as a technology story but as an infrastructure story — one that will reshape capital markets, power grids, and the balance sheets of every company that depends on compute. It also raises the question investors are only beginning to ask: how much of this spending is a durable structural shift, and how much is a cycle that could turn?
The Scale of the Commitment
PwC's forecast arrives on top of its Global Infrastructure Outlook, published in April, which modeled annual global infrastructure spending rising from $4.4 trillion in 2024 to $6.9 trillion in 2050, with cumulative investment across all sectors reaching $151.1 trillion over the period. That analysis — covering nine sectors, 20 subsectors, and 45 countries and territories representing 88% of global economic output — found that in real terms, infrastructure spending over the next 25 years will be roughly double that of the past 20 years.
Within that tidal wave, data centers stand out. PwC's April outlook showed annual investment in data center buildings alone rising 2.2 times between 2024 and 2027, from $113.8 billion to $251.8 billion, with total building investment from 2024 to 2032 topping $1.5 trillion. The firm was explicit that this building spend comes in addition to investment in ICT equipment, such as chips and servers — meaning the $31.6 trillion total through 2050 captures a much broader chain than construction alone.
The data center buildout sits within a broader convergence of power, transport, and digital infrastructure. Transport and power will together account for about half of global infrastructure spending through 2050 — roughly $50 trillion and $25 trillion respectively. Defence is the fastest-growing sector, with annual spending on physical installations expected to reach $168 billion in 2050, 2.3 times the 2024 level of $73 billion, as governments respond to intensifying geopolitical risks.
This is not a traditional construction cycle. This next generation of infrastructure will be intelligent, connected and adaptable. Systems will need to anticipate demand, allocate resources dynamically and optimise performance—delivering structural productivity gains across every sector. — PwC
Why the Estimates Keep Rising
The $31.6 trillion figure is not an isolated forecast. It sits at the top of a ladder of estimates that have been ratcheting upward for two years. Dell'Oro Group projected in August 2026 that worldwide data center capital spending would surpass $3 trillion by 2030 — an outlook the firm said had nearly doubled since January 2026, driven by higher hyperscaler spending guidance, increased estimates for global data center power capacity, and rising commodity costs. Goldman Sachs Research has estimated roughly $7.6 trillion of cumulative data center capex between 2026 and 2031. Allianz expects annual data center investment to rise from around $500 billion in 2024 to more than $1 trillion as early as 2027.
What explains the escalation? The answer lies in the physics of AI workloads. Unlike the cloud migration wave of the 2010s, which largely shifted existing workloads to more efficient facilities, generative AI creates entirely new demand for compute — and it does so at a density that strains the grid. The International Energy Agency estimates that global data centers consumed about 415 terawatt-hours of electricity in 2024, roughly 1.5% of global electricity consumption, and projects that figure will more than double to around 945 TWh by 2030 — slightly more than Japan's total electricity consumption today. In the United States, data centers account for nearly half of electricity demand growth between now and 2030.
This is the transmission mechanism that makes the data center boom structurally different from previous technology capex cycles: AI compute demand is inseparable from physical infrastructure. A chip order is not just a semiconductor sale; it is a commitment to buildings, cooling systems, transformers, transmission lines, and often dedicated power generation. Dell'Oro estimates AI accelerators will account for about one-third of the $3 trillion in data center capex projected for 2030 — but the accelerators themselves require servers, specialized networking for AI clusters, and storage to support training and inference. The multiplier from a dollar of chip spending to a dollar of infrastructure spending is the engine behind the $31.6 trillion.
That multiplier is also why the boom will not self-correct quickly. Data centers have development lead times measured in years, not quarters. Once a project is permitted, financed, and under construction, it continues absorbing capital even if the demand outlook softens. This creates a form of capital inertia: spending that is locked in by physics and construction timelines rather than by the quarterly earnings cycle.
The Financing Layer: Cash Flow Gave Way to Leverage
The shift in how the buildout is funded is as important as the shift in how much is spent. During the early part of the current cycle, hyperscalers funded most capex from their own cash flow, which created a natural rate limit on growth and on system risk. That changed in 2025, as data center spending soared and absorbed an ever-larger share of earnings. The pattern is visible in the quarterly capex disclosures of the largest AI-focused companies: spending grew faster than operating cash flow, and the gap widened.
That gap has been filled by a widening set of financing structures. Sale-leaseback transactions, through which operators sell facilities to investors and lease them back, have moved billions of dollars of data center assets off the balance sheets of technology companies and onto the books of real estate investors and private funds. Special-purpose vehicles have been created to finance individual campuses, often with revenue tied to long-term leases from a single anchor tenant. Private credit has stepped in where traditional bank lending reaches its limits, offering flexibility at a higher cost.
The risk embedded in these structures is duration mismatch. A data center lease may run 15 years, but the debt financing it, the power purchase agreements underpinning it, and the technology inside it all have different lifecycles. If utilization falls or refinancing costs rise, the first losses appear not at the hyperscaler level — where balance sheets remain relatively robust — but at the level of the leveraged vehicles built to monetize the buildout. This is why the counter-thesis does not require a collapse in AI demand to be painful; it only requires demand to grow more slowly than the debt schedule assumed.
The Counter-Thesis: Is This the Next Fibre-Optic Bust?
The strongest argument against the bullish case is simple: capacity can be built faster than revenue. Man Group, in a 2026 analysis of AI bubble risks, laid out a data-centre overbuild scenario in which low utilization becomes visible as new facilities come online. In that world, compute special-purpose vehicles struggle to refinance at higher rates, data center operators face rent compression, and private-credit vehicles take net asset value write-downs. The analogy is the telecom fibre bust of the early 2000s, when physical capacity far outstripped monetizable demand.
There is already evidence of strain. Omdia, in a 2026 report, flagged a downside scenario reflecting a failure to realize productivity gains through AI use quickly enough, and after five years of accelerated investment, investors get spooked. The firm identified 2027 as a key year, because major AI developers have made revenue commitments for that date. If those commitments are missed, the funding model for the buildout comes under pressure.
But the counter-thesis has a weakness: it assumes AI demand is a single wave rather than a platform shift. The fibre bust happened because long-distance bandwidth was built once and then sat idle. Data center demand, by contrast, is recurring and compounding — every new AI application, every autonomous agent, every enterprise workflow that moves on-model adds incremental inference load. The question is not whether some projects will be stranded; some will be. The question is whether the aggregate utilization rate falls far enough, for long enough, to trigger a systemic financing event.
The Real Constraint Is Not Money — It Is Power
The most binding constraint on the $31.6 trillion forecast is not capital availability but electricity. Dell'Oro's $3 trillion by 2030 forecast assumes global data center power availability will increase to more than 200 gigawatts. Getting there requires not just generation but transmission, substations, and grid interconnection — the parts of the energy system with the longest permitting timelines and the most local opposition.
Allianz reported that in the United States alone, local opposition delayed or blocked at least 75 data center projects worth almost $130 billion during the first quarter of 2026. The Netherlands has also seen projects challenged. These are not marginal delays; they are evidence that the social license for data center construction is becoming a priced risk, not a given.
This constraint cuts both ways. On one hand, it means the $50 trillion upside scenario is far from guaranteed — if power and permitting bottlenecks persist, actual spending will fall short of PwC's central forecast. On the other hand, it means that projects which do secure power have durable competitive advantages, because new entrants cannot quickly replicate their position. Scarcity of power, in other words, creates scarcity value for the operators who already have it.
The geographic distribution of spending reflects this constraint. The United States accounts for the largest share of projected data center electricity demand growth, followed by China, according to the International Energy Agency. But within the United States, the centers of gravity are shifting away from traditional Northern Virginia hubs toward markets with available power — the Ohio Valley, the Upper Midwest, and the Southwest — where hyperscalers are building campuses anchored to dedicated generation rather than grid supply.
Who Benefits, Who Is Exposed
The beneficiaries of the buildout extend well beyond the chip designers that have captured most of the investor attention so far. The spending breaks across a chain: land and site development, construction, power generation and cooling equipment, grid infrastructure, semiconductors, servers, networking, and storage. PwC's Global Infrastructure Outlook identifies power, transport, and digital infrastructure as the three sectors that will converge into intelligent networks — meaning the companies that sit at the intersections, such as electrical equipment manufacturers and grid-technology providers, stand to capture recurring revenue across multiple infrastructure categories.
The exposed parties are the financiers of marginal projects. Special-purpose vehicles built on optimistic utilization assumptions, private-credit funds with long-duration exposure to single-tenant data center leases, and operators that locked in power at peak prices face the highest downside if the overbuild scenario materializes. The asymmetry is clear: the owners of scarce, permitted, powered sites have limited downside; the providers of capital to greenfield projects in unconstrained locations have significant downside.
What to Watch
The forecast will be tested by three signals over the next 18 months. First, hyperscaler capital expenditure guidance for 2027 — a cut in guidance would be the earliest warning that the financing model is straining. Second, data center power availability versus the 200 GW benchmark Dell'Oro assumes for 2030; a persistent shortfall would cap the upside scenario. Third, utilization rates at newly opened facilities, which will reveal whether demand is compounding or plateauing.
The falsifying signal for the structural-shift thesis is specific: if AI-related data center utilization at newly commissioned facilities falls below 50% for four consecutive quarters while capex guidance remains elevated, the boom is being funded by expectation rather than demand, and a cyclical correction becomes likely. Conversely, if utilization holds above 70% and power availability tracks the 200 GW path, the $50 trillion upside scenario moves into play.
Short term, the market will react to quarterly capex prints and any signs of guidance cuts. Medium term, the bottleneck shifts from chips to power and permitting — the companies that solve those constraints become the bottleneck owners. Long term, the question is whether AI proves to be a general-purpose technology that compounds demand indefinitely, or a wave of applications that saturates. PwC's $31.6 trillion central forecast, and its $50 trillion upside, price the former.
The $31.6 trillion figure is not a prediction that every dollar will be spent wisely. It is a statement about the scale of the commitment already in motion — and a reminder that the AI revolution will be built in concrete and copper long before it shows up in productivity statistics.
Explore more exclusive insights at nextfin.ai.
