Executive summary

Amazon, Microsoft, Alphabet and Meta plan to spend $720 to 745 billion of capital expenditure in 2026, call it $733 billion at the midpoints of their guidance. The figure is nearly double the roughly $380 billion the four spent in 2025, which was itself nearly double 2024's $225 billion, and it is larger relative to the American economy than the peak year of the 1990s telecom boom, the closest historical parallel and the standard cautionary tale. Divided mechanically by the IMF's projected $32.38 trillion US economy, the 2026 budget comes to approximately 2.3%, an illustrative scale comparison, since the companies' spending is global while the denominator is American. The US telecom sector spent an estimated $120 billion in 2000, its peak year, when that was roughly 1.2% of GDP. On those two ratios the AI build-out is running at close to twice the telecom boom's peak intensity, and it is still accelerating.

The record behind that picture comes from the companies' own filings, assembled year by year and cross-checked in September 2026. Cumulative capex since January 2023 crossed $1.05 trillion in mid-2026, a reconstruction from filings rather than a company-reported total, with its definitional caveats set out in the methodology. Cash generation is straining under the load: market tallies compiled from the filings of the five largest hyperscalers (the four above plus Oracle) show combined quarterly free cash flow falling from roughly $61 billion in 2024 to under $5 billion by mid-2026, a decline FactSet's July 2026 review documents. Meta's individual figures make the squeeze concrete: $784 million of free cash flow on $31.86 billion of operating cash flow in the second quarter of 2026, though the first half still produced about $13.2 billion. External capital is arriving at the margin: Meta's October 2025 $27 billion Hyperion joint-venture financing with Blue Owl and its subsequent ~$30 billion private-credit raise, Oracle's borrowing against its OpenAI contracts, and AI-linked corporate debt issuance of roughly $141 billion in 2025 by credit-market tallies.

Concentration shows on both sides of the market. Nvidia's data-center revenue of $89.0 billion in the quarter ended July 2026 annualizes to roughly $356 billion, equivalent to about 49% of big-four capex, a scale comparison rather than a procurement share, since Nvidia also sells to AI labs, enterprises and sovereign customers worldwide. In its most recent filing Nvidia reports three direct customers at 16%, 15% and 13% of first-half revenue; a daloopa reconstruction of its Q4 FY2026 disclosures put four customers at roughly 61%. Electricity follows the same curve: US data-center consumption was 4.4% of national use in 2023, and Lawrence Berkeley National Laboratory's 2025 update projects 11.8% by 2030 in its reference case, with scenarios from 9.5% to 15.3%. Power interconnection has become the constraint on many projects in several major US markets.

If demand materializes, this is the fastest productive-capacity expansion for which modern markets have records. If it does not, the write-down would be measured in hundreds of billions of dollars, concentrated in a handful of balance sheets that also anchor American equity indices, pension savings and the corporate bond market. Either way, the boom is already large enough to move national statistics: its spending is visible in US GDP growth, its electricity demand in regional grids, and its borrowing in credit markets.

The figures here follow each company's headline definition, which differ in lease treatment (methodology below). No company discloses "AI-only" capex. Cumulative and GDP-relative figures are reconstructions with stated error bars. Forward-looking statements are company guidance, not forecasts.

Bar chart of combined big-four capital expenditure rising from 144 billion dollars in 2023 to a planned 733 billion in 2026

Key findings

  1. $720 to 745 billion is the 2026 plan, nearly double 2025. The big four spent roughly $380 billion in 2025 and plan $733 billion at guidance midpoints for 2026, a 93% one-year increase after 68% growth in 2025 and 56% growth in 2024.

  2. On the mechanical comparison, near twice the telecom peak's intensity. Big-four capex over US GDP equals about 2.3% in 2026; the US telecom sector's 2000 capex peak was about 1.2%. The numerators differ in kind, global company spending versus US-sector spending, so the pair travels as a scale illustration throughout.

  3. Capex is consuming cash generation. Meta's 2026 guidance equals roughly 69% of its 2025 revenue, Microsoft's about 62%, Alphabet's about 50%, Amazon's about 31%. Combined quarterly free cash flow across the five largest hyperscalers fell from roughly $61 billion in 2024 to under $5 billion by mid-2026, on tallies FactSet's July 2026 review corroborates.

  4. External capital matters more at the margin. Meta's $27bn Blue Owl JV and ~$30bn private-credit raise, Oracle's data-center borrowing and ~$141bn of AI-linked debt issuance in 2025 document the drift toward credit, without yet constituting a wholesale shift of the big four to debt financing. Moody's counts $662bn of uncommenced data-center leases across the five largest hyperscalers, 113% of their adjusted debt.

  5. One vendor sits at the narrowest point of the supply chain. Nvidia's annualized data-center revenue of roughly $356 billion is equivalent to about 49% of big-four capex, a scale comparison that includes its sales to buyers beyond the four companies. Its latest filing reports three direct customers at 16%, 15% and 13% of first-half fiscal-2027 revenue; a reconstruction of its Q4 FY2026 disclosures put four customers at roughly 61%.

  6. Commitments now extend far beyond spending. Meta alone discloses $237.7 billion of non-cancelable contractual commitments, $182.9 billion of signed-but-not-commenced leases and up to $14.7 billion of contingent capacity purchases as of 31 March 2026.

  7. OpenAI has said it is looking at roughly $1.4 trillion of infrastructure commitments over eight years, across Oracle (up to $300bn), Stargate and chip suppliers, an aggregate of separately announced deals, far above its current revenue. The deals transfer construction risk to suppliers and lenders today in exchange for revenue that must materialize tomorrow. One flagship Oracle-OpenAI data-center expansion was scrapped in March 2026 after terms could not be reached, an early sign of friction in that financing chain.

  8. Power has become the constraint in several major US markets. LBNL's 2025 update projects US data-center load at 11.8% of national electricity by 2030 in its reference case (9.5 to 15.3% across scenarios). Grid interconnection queues, transformer shortages and local opposition now set the pace of the build-out in those markets, and Microsoft has cited roughly $80 billion of unfulfilled Azure orders, which reporting has tied in part to power and capacity constraints.

  9. Backlogs show large contracted demand, as future obligations, not cash. Google Cloud's backlog reached $514 billion by mid-2026, contracted future revenue Alphabet expects to recognize over years, just over half within 24 months. Azure growth remained above 30%. The debate is not whether demand exists in 2026 but whether contracted growth rates persist through the multi-year economic lives of the assets being built.

  10. The telecom comparison is instructive but bounded. The 1996 to 2001 boom raised roughly $1 trillion (about $500bn of it capex) and left a fiber glut so deep that prices collapsed; that fiber later carried the streaming and mobile internet. Its most aggressive builders were unprofitable and debt-financed while incumbents also spent heavily; today's builders are exceptionally profitable, and their spending remains majority-funded from operations. The debt-financed-overbuild risk rhymes; the balance-sheet starting point does not.

Bar chart showing big-four capex as a mechanical share of US GDP reaching 2.3 percent in 2026 versus 1.2 percent at the 2000 telecom peak

The spending record, company by company

Three of the four report on calendar years, which makes their lines easy to draw. Amazon's cash capital expenditures ran $48.1 billion in 2023, $77.7 billion in 2024 and $128.3 billion in 2025, with the company attributing the increase primarily to technology infrastructure supporting AWS; in February 2026 it guided to $200 billion for 2026, and by mid-year the plan had risen to approximately $220 billion, attributed to higher component costs and AWS/AI capacity. Amazon's first-quarter free cash flow swung to an outflow of $7.6 billion as server purchases accelerated. Alphabet's purchases of property and equipment climbed from $32.3 billion (2023) to $52.5 billion (2024) and $91.4 billion (2025), and its 2026 guidance was raised twice, to $180 to 190 billion in April, then to $195 to 205 billion after a second quarter in which the company spent $44.9 billion and disclosed a Google Cloud backlog of $514 billion.

Meta's capex on its preferred measure, purchases of property and equipment plus finance-lease principal, ran $28.1 billion (2023), $39.2 billion (2024) and $72.2 billion (2025), with guidance for 2026 raised repeatedly to $130 to 145 billion and the first half of 2026 alone recording $50.9 billion. The board approved no buybacks in recent quarters as spending absorbed cash generation. Microsoft reports on a June fiscal year, and its line carries a lesson about definitions: cash payments for property and equipment rose from about $28.1 billion (FY2023) to $44.5 billion (FY2024) and $64.6 billion (FY2025), with finance leases adding substantially more capacity in kind than in cash. Guided to roughly $190 billion of calendar-2026 capex in April, the company reported about $175 billion in July after extending the useful lives of data-center buildings from 15 to 25 years, because more future leases now classify as operating rather than finance leases; it stated its underlying investment expectations were unchanged, an accounting illustration of why cross-company capex comparisons require care.

Summed at guidance midpoints, 2026 planned spending is approximately $733 billion: Amazon ~$220bn, Alphabet ~$200bn, Microsoft ~$175bn, Meta ~$137.5bn. The chart below shows how unevenly that total divides among four companies of very different sizes.

Pie chart of 2026 planned capital expenditure by company: Amazon 220 billion dollars, Alphabet 200 billion, Microsoft 175 billion, Meta 137.5 billion

The cumulative number: $1.05 trillion in three and a half years

Adding calendar-aligned spending across the four companies produces roughly $750 billion from January 2023 through December 2025, and about $300 billion more in the first half of 2026, a cumulative reconstruction of approximately $1.05 trillion since January 2023. Three qualifications attach to the figure. Definitions differ: Amazon reports cash capex, Meta adds finance-lease principal, Microsoft's headline moves with lease classification, so the reconstruction aligns quarters but cannot erase the definitional joins. "Capex" is not "AI": the disclosures include conventional cloud equipment, fulfillment centers, offices and networking, and no company reports an AI-only figure, so estimates that attribute everything to AI overstate it and estimates that exclude supporting infrastructure understate it. And leases blur the line: when a leased data center opens, years of future payments move from footnotes onto balance sheets in one quarter, which is how Microsoft's useful-life change moved roughly $15 billion of planned spending between categories with no change in physical investment.

A chronology of the escalation

The speed of the build-out is easiest to see as a sequence. Each step below is dated and sourced; together they show how quickly the spending base rate changed.

2024. Big-four capex runs about $225 billion for the year, up 56% on 2023, itself up sharply on 2022, as the ChatGPT-era demand shock reaches procurement budgets. Microsoft's finance-lease commencements begin visibly outrunning its cash spending, an early signal that capacity demand exceeds what conventional funding cycles anticipated.

September 2025. Oracle and OpenAI announce a cloud contract of up to $300 billion over five years, beginning in 2027, a commitment larger than Oracle's entire annual revenue, to be delivered by capacity not yet built. OpenAI's Stargate venture turns to debt-financed chip leasing. In the same month, Meta closes the $27 billion Blue Owl financing for its Hyperion campus.

November 2025. The Financial Times tallies OpenAI's infrastructure commitments at roughly $1.4 trillion over eight years and asks how they will be funded. AI-linked corporate debt issuance reaches roughly $141 billion for 2025. Oracle shares sell off hard as investors confront the borrowing behind its OpenAI contracts.

February 2026. The year's guidance round lands as a shock: Amazon guides to $200 billion, Alphabet to $175 to 185 billion, Meta to roughly $115 to 135 billion, Microsoft signals capex growth above its fiscal-2025 total. Financial press comparisons to national economies (Fortune: the spend 'rivals Sweden's GDP') enter the mainstream.

March to April 2026. The chain of announced commitments shows its first public friction: Oracle and OpenAI scrap plans to expand a flagship Stargate data center after terms cannot be reached. Alphabet raises guidance again (April). Meta signs roughly $24 billion of additional infrastructure contracts in April alone. The Wall Street Journal reports that syndicating Oracle's data-center debt is straining bank exposure limits.

July to August 2026. Microsoft's useful-life reclassification adjusts its reported capex to ~$175 billion with no change in physical plans. Reuters and FactSet quantify the free-cash-flow squeeze. Alphabet raises guidance a third time to $195 to 205 billion on a $514 billion cloud backlog; Meta narrows to $130 to 145 billion. The four-company total settles at $720 to 745 billion, roughly $85 billion above where the year's guidance began in February.

September 2026. S&P Global Market Intelligence projects more than $5 trillion of hyperscaler capex from 2026 through 2030; Goldman Sachs' June estimate for the big four alone was $5.3 trillion. By September 2026 the build-out was in its fourth calendar year, with no company guidance yet issued for 2027.

Each guidance round has been raised above the last, and each raise has been justified by orders taken in the preceding quarter. The repeated upward revisions indicate that spending expectations have so far behaved more like an escalating cycle than a one-time adjustment, a pattern that can, like any pattern in guidance behavior, also reverse.

Bar chart of big-four capex by year with the escalation from 2023 through planned 2026 spending

Scale against the economy: what the 2.3% comparison can and cannot say

Dividing $733 billion of planned spending by the IMF's projected $32.38 trillion US economy for 2026 produces approximately 2.3%. The division is mechanical and its reading requires care: the four companies' capital expenditures are global, spent across the United States, Europe, Asia and elsewhere, while the denominator is American GDP. The ratio is therefore an illustration of corporate scale against national accounting, not a measure of US-located investment. It is used that way here, and the telecom comparison below inherits the same caveat.

The historical context gives the number weight:

On the mechanical measure, the four hyperscalers are building, in a single year, at close to twice the telecom boom's peak intensity relative to the economy, and published projections anticipate more in 2027. Two things follow. First, the boom is macroeconomically material: at this scale, capex swings show up in national investment statistics and GDP growth arithmetic. Second, the honest comparison set is not only "other companies" but "other national infrastructure programs", because that is the intensity class the spending now occupies.

The deeper historical baseline

Placing the boom only against telecom understates the comparison set. Economic historians generally place peak railroad investment in the 1870s, 1880s and again around 1900 at several percent of GDP in the countries most affected, financed through a mix of equity, bonds and land grants, booms that delivered transformative networks and repeated investor wipeouts, sometimes in the same decade. American electric-utility construction in the late 1920s ran in the low single digits as a share of GDP before the Depression forced consolidation and regulation. Interstate highway spending, by contrast, was spread over 35 years and funded publicly, which is why its annual intensity looks modest even though its cumulative scale was vast.

Two features of that longer record apply directly here. The first is that infrastructure booms at this intensity are rarely completed on their original financial terms. Railroads were reorganized in receivership; utilities were consolidated under regulation; fiber carriers went through bankruptcy while the fiber itself kept carrying traffic. The asset typically survives and eventually earns its keep; the capital structure that built it at peak prices frequently does not. Applied to AI, the campuses being poured in 2026 will very likely be running workloads in 2035; the open question is which class of investor holds the paper, and at what price, when the returns are finally written down.

The second is that the financing cadence has mattered as much as the demand estimate. The canal and railroad manias did not end because demand for transport disappeared; they ended because incremental capital was raised against extrapolated demand after the easily-served market was saturated. The AI analogue is a chain of contracted commitments: OpenAI's $1.4 trillion of announced commitments, $662 billion of uncommenced leases, Meta's $237.7 billion of non-cancelable contracts, payment schedules written against demand that is contracted but not yet observed, maturing 2027 to 2032.

What a 2.3%-of-GDP-scale year does to the arithmetic of growth

The increment alone, roughly $350 billion of additional capex in 2026 versus 2025, would, if every dollar were spent domestically, add more than a percentage point to the level of US GDP in a $32.4 trillion economy. In practice the impulse is smaller for two reasons the national accounts make visible: a substantial share of the equipment content is manufactured abroad (advanced logic and memory wafers from Taiwan and South Korea, assembled into servers domestically), and some spending lands as inventory and intermediate goods rather than final demand. The domestic footprint remains large enough that economists now explicitly list AI data-center investment among the contributors to US investment growth, and its pause among the risks to it.

Horizontal bar chart of Meta's disclosed future obligations: 237.7 billion in contractual commitments, 182.9 billion in uncommenced leases, 14.7 billion contingent

Who funds it: operating cash still leads, outside capital grows at the margin

The boom's first phase was financed from operations. The evidence on the second phase is directionally consistent but needs careful statement: external capital has become materially more important, particularly across the wider ecosystem, without the four hyperscalers having abandoned cash financing.

The cash-flow squeeze. Reuters reported in July 2026 that hyperscaler capex plans exceed their operating cash flow growth: operating cash flow is projected to rise by roughly $340 billion between 2025 and 2027, while capex over the same horizon rises faster. Market tallies compiled from the five largest hyperscalers' filings (Amazon, Microsoft, Alphabet, Meta, Oracle) show combined quarterly free cash flow falling from about $61 billion in 2024 to under $5 billion by mid-2026. FactSet's July 2026 review documents the same trend and projects further decline; S&P Global Ratings expects the major US hyperscalers to run negative free operating cash flow into 2027. Microsoft has said it expects to remain free-cash-flow positive; Alphabet has said it plans to fund investment through a combination of operating cash flow, debt and equity.

The mechanisms that follow, gathered from disclosures and named-source reporting through mid-2026, are set out in a table because their variety is the finding:

Mechanism Example Scale What it does to the buyer's balance sheet
Private-credit joint venture Meta-Blue Owl Hyperion financing, Oct 2025 $27bn Debt held by the vehicle; capacity secured off-balance-sheet, ratings-adjusted metrics treat it as debt-like
Private-credit follow-on Meta package led by PIMCO and JPMorgan, late 2025 ~$29 to 30bn Same structure applied to additional campuses
Supplier (customer) borrowing Oracle bond deals against OpenAI contracts successive raises Debt builds capacity for delivery from 2027; syndication desks strained against exposure limits
Sector issuance AI-linked corporate debt, 2025 ~$141bn Record year by credit-market tallies, before a heavier 2026 calendar
Leases, not yet commenced Moody's count, five largest builders, Feb 2026 $662bn (113% of adjusted debt) Years of future payments move onto balance sheets only as centers open

The obligations already signed. Meta's filings disclose $237.7 billion of non-cancelable contractual commitments, $182.9 billion of leases signed but not yet commenced, and up to $14.7 billion of contingent cloud-capacity purchases as of 31 March 2026, plus ~$24 billion of additional infrastructure contracts signed in April. The widely quoted "$1.65 trillion of hidden debt" (Nikkei) mixes these categories and should be read as contractual exposure, not funded debt; the narrower, auditable numbers are large enough.

The direction of travel is clear: a boom that began on operating cash is adding credit at the margin, one financing at a time, and the marginal structure, not the average, is where risk concentrates first.

Why the buyers accept these structures

If external financing is more expensive and more fragile than operating cash, why do companies with the strongest balance sheets of their era accept it? Four factors, each verifiable in the disclosures, make the answer structural rather than imprudent.

The first factor is speed. A gigawatt-scale campus costs $10 to 25 billion and takes 18 to 36 months to build. Waiting two years to accumulate the cash from operations means entering service two years late against competitors who financed instead, and in a market where cloud backlogs measure years of committed demand, lateness is a revenue loss, not a saving. Meta's stated reasons for raising its 2026 guidance cite higher component pricing, additional data-center costs and future-year capacity; the company's disclosures also describe demand running ahead of available capacity, as its competitors' do.

The second factor is the accounting itself, and the Blue Owl structure matters less for its interest rate than for its accounting position: capacity secured off-balance-sheet, with Meta holding an equity interest and the special-purpose vehicle carrying the debt. Moody's and S&P both treat such arrangements as debt-like in adjusted metrics, 113% of adjusted debt in uncommenced leases is Moody's arithmetic, but ratings impacts and covenant tests differ from funded bonds. The structures let the builders get capacity today without the market pricing them as leveraged tomorrow. That is rational for the buyer; it concentrates the same risk in non-bank lenders instead.

The third factor is that the suppliers get pulled in. Oracle cannot deliver $300 billion of OpenAI capacity from its existing cash generation; its bonds are the mirror image of its contracts. Nvidia's financing programs, chip-lease structures at Stargate, and vendor pre-payment arrangements are the same mechanism one step upstream. When the largest buyer's largest commitments exceed the buyer's revenue, the financing burden shifts toward whichever party has the balance sheet to carry it until revenue arrives, one observed mechanism among several, not a law.

The last factor is depreciation arithmetic. Accelerated server depreciation (roughly 5 to 6-year schedules at several builders) means the server fleet consumes its accounting life quickly. Depreciation is an accounting estimate of useful life, not a physical replacement clock, actual refresh decisions depend on utilization, hardware generations and obsolescence, and companies have revised server lives before. What the schedules do establish is the pace at which reported earnings absorb the capex: as the 2023 to 2026 cohort ages, depreciation charges rise whether or not the equipment is physically replaced, and that income-statement weight is examined in the revenue section below.

What "$733 billion" actually buys

Distinguishing capacity types sharpens the picture, because the boom's constraint differs by type.

Contracted capacity, ordered and in the supply chain: large and growing, but its exact share of 2026 spending is not disclosed by any company. What is documented: tight allocation across parts of the HBM and advanced-packaging supply chain, component lead times stretching well forward, and procurement practices (orders placed 6 to 18 months ahead) that point to a heavily committed pipeline.

Built capacity, racked and powered: materially less than contracted, and the gap is the power story. Microsoft's cited ~$80 billion of unfulfilled Azure orders, reported in January 2026 and attributed by reporting in part to power and capacity constraints, is the clearest published measure of the gap between what customers have committed to buy and what can physically run.

Available (grid-ready) capacity, the constraint in several major US markets: interconnection queues running years in some regions, transformer lead times of 2 to 4 years, and gas-turbine backlogs measured in years, Siemens Energy reported a firm gas-turbine backlog near 69 GW in mid-2026 with lead times of three years or more, and GE Vernova's backlog reached roughly 100 GW with slots sold well into 2028 to 2029. Electricity-demand projections describe how much power data centers may draw, not how much contracted capacity will be energized on schedule; the two are related but distinct quantities.

Economically viable capacity, the open question: capacity that earns its cost of capital at plausible 2028 to 2030 prices for AI services. Nobody's filing discloses this number, and it is where overbuild would show up first: not in construction statistics but in utilization and pricing, one to two years after the buildings open.

Through 2026, then, the boom is supply-constrained, with demand exceeding capacity at the margin. The risk case is not present in current utilization, it lives in the 2027 to 2029 cohorts, where capacity decisions are being made today against demand that is contracted but not yet observed.

Bar chart comparing 2026 capex to 2025 revenue: Amazon 31 percent, Microsoft 62 percent, Alphabet 50 percent, Meta 69 percent

Who supplies it: the Nvidia funnel and the concentration question

Money spent on AI compute passes through a narrow supply chain, and its narrowest point is one vendor. The company reported $96.2 billion of revenue in the quarter ended 26 July 2026, of which $89.0 billion was data center, annualizing to roughly $356 billion. Set against $733 billion of big-four capex, that annualized figure is equivalent to about 49% of the four companies' planned spending. The equivalence has a precise meaning and a common misreading. What it establishes: the scale of Nvidia's data-center business relative to the buyers' budgets, and since Nvidia also sells to AI labs, enterprises, sovereign programs, server integrators and other cloud providers, its revenue is not a measure of what the big four alone buy. What it does not establish: a procurement share. No company discloses what fraction of its equipment spending reaches Nvidia. What can be said from disclosures: the big four are repeatedly identified among Nvidia's largest customers, Alphabet's 60%-of-capex server line is the biggest single component of at least one major buyer's budget, and server integrators (Supermicro, Dell) add Nvidia content beyond the direct sales. The defensible statement is the scale comparison, roughly $356 billion of annualized data-center revenue against $733 billion of planned capex, with the procurement share unknown and plausibly large.

Concentration on the customer side has a precise, dated record. Nvidia's most recent filing (10-Q for the quarter ended July 26, 2026) reports one direct customer at 16% of quarterly revenue, and three direct customers at 16%, 15% and 13% of first-half fiscal-2027 revenue, roughly 44% combined, with the filing cautioning that indirect-channel attribution is estimated. Earlier, a daloopa reconstruction of Nvidia's Q4 FY2026 disclosures identified four customers at roughly 61%. The two vintages are not comparable measurements, different periods, different estimation bases, but both describe a customer base in which a handful of buyers accounts for a large share of revenue. Nvidia's market value, which crossed $5 trillion during 2026, rests on that base continuing to spend; conversely, the hyperscalers' AI product roadmaps depend on one supplier's roadmap. Both sides have spent 2025 to 2026 trying to diversify, custom silicon (Google TPU, Amazon Trainium, Microsoft Maia), AMD accelerators, in-house networking, but each year's spending has so far concentrated further.

One step downstream, the funnel narrows again: SK Hynix's HBM lines are heavily allocated well into 2027; TSMC's CoWoS advanced-packaging capacity is booked years forward; high-bandwidth memory and copper supply tightness has pushed component costs up, a driver Amazon explicitly cited for raising its 2026 budget. That concentration, with allocation layered on top, is what makes the chain fragile: a demand pause at two or three buyers would propagate in a way diversified demand would not.

Above all of it sits the circularity question. The ecosystem contains loops that complicate demand signals: Nvidia has invested in OpenAI and other customers; Microsoft invests in OpenAI, which rents Microsoft and Oracle capacity; Oracle borrows to build for OpenAI, whose revenue depends partly on products run on that capacity. Each loop is rational individually; collectively they make contracted demand harder to read, because some buyers' ability to pay is linked to the sellers financing the purchase. Circularity is not fraud, and most such deals are disclosed, but it is a real analytical problem when judging whether end-demand supports the build-out.

Bar chart comparing big-four 2026 capex of 733 billion dollars with Nvidia annualized data-center revenue of 356 billion dollars

The physical system: what a $733 billion year actually builds

Each stage of the chain, from budget to electrons, carries a different kind of constraint.

The chain starts with the budget and ends in components. Alphabet has said roughly 60% of its capex goes to servers, with about 40% to data centers and networking equipment. A server is an assembled system, processors, memory, storage, chassis, power components inside it, so this split divides computing hardware from the facilities and network that house it, not silicon from everything else; other companies' mixes are similar in kind. Components are ordered 6 to 18 months ahead; servers for 2026 were largely contracted in 2025.

From components the money moves to the data center, where an AI-ready hall takes 18 to 36 months from groundbreaking to racks, limited less by construction than by power: a single 1-gigawatt campus draws the electricity of a mid-sized city and needs transformers that currently carry multi-year lead times, high-voltage interconnections that queue behind regional grid studies, and in some markets water for cooling that local authorities must approve.

Only then does the chain reach a product: racked accelerators train and serve models, revenue arrives only when enterprises or consumers pay for tokens, software or cloud. The demand evidence is real, Google Cloud backlog $514bn, Azure growth 30%+, record API adoption, but backlog is contracted future revenue, not cash received and not a guarantee of demand persistence.

Underneath all of it runs the electricity arithmetic. US data centers used about 4.4% of US electricity in 2023 (DOE/LBNL). The Department of Energy's December 2024 report projected 6.7 to 12% by 2028. Lawrence Berkeley National Laboratory's 2025 Update, the current edition, projects 11.8% by 2030 in its reference case (about 649 TWh), with scenarios spanning 9.5% to 15.3%. Globally, the IEA's base case reaches ~945 TWh by 2030, roughly double today's share of about 1.5% of world electricity. Those percentages describe projected electricity demand from data centers, an input to, not a measurement of, how much contracted capacity gets energized on schedule. The supply side, new gas turbines (Siemens Energy's backlog near 69 GW with multi-year lead times; GE Vernova's near 100 GW), nuclear restarts and PPAs, grid batteries, transmission, is being procured in parallel, and in several markets the power procurement, not the capital budget, is the critical path. Microsoft's publicly discussed constraint of roughly $80 billion in unfulfilled Azure orders, reported in January 2026 and tied in part to power availability, is the clearest single illustration.

Horizontal bar chart of data center electricity share: 1.5 percent global 2025, 4.4 percent US 2023, 11.8 percent US 2030 reference case

The telecom baseline: what the comparison does and does not establish

The 1996 to 2001 telecom boom remains the closest historical parallel, and it deserves precise treatment.

What rhymes.

What does not rhyme.

The comparison supports a narrow claim: the AI boom's intensity and financing direction rhyme with telecom; its starting balance sheets and current demand do not. The historical analogy argues for watching leverage and end-demand, not for predicting a 2001-style collapse.

How the telecom boom actually ended: the case in detail

The ending is worth setting out in detail, because it is the part most often remembered loosely.

The build itself ran from 1996 to 2001, in which telecom operators, incumbents, competitive local exchange carriers, and a generation of newly licensed long-haul builders such as Global Crossing, Qwest and Level 3, laid tens of millions of miles of fiber. Sector capex ran at roughly $90 to 120 billion a year at the peak, funded by high-yield debt and equity raised against the expectation that internet traffic would double every 100 days (a figure that held only briefly and was extrapolated for years).

Then came the turn. Traffic kept growing, the demand forecast was eventually right in direction and wrong in cadence, but the capacity arrived faster than revenue, and wholesale bandwidth prices collapsed as lit fiber competed for the same routes. Debt service did not fall with prices. WorldCom's accounting fraud (capitalizing operating costs to show profit) and bankruptcy in July 2002, then the largest in US history, marked the finale; Global Crossing had failed in January 2002. Roughly half a trillion dollars of market value and a large share of the sector's debt were destroyed. Equipment vendors fell harder than carriers: Nortel went from Canada's most valuable company to a fraction of itself; Lucent survived only in merger.

The fiber's afterlife is the third act. It kept working. Buyers acquired the assets in bankruptcy at cents on the dollar; the glut the boom created became the cheap substrate on which video streaming, cloud and mobile broadband were built a decade later. Society captured the value; the original capital largely did not.

The record suggests three indicators observable in AI today: whether token demand compounds as fast as capacity is contracted, what happens to generic inference pricing as capacity scales, and whether the uncommenced leases and JV obligations flex when revenue disappoints. The first currently passes; the second is unresolved and worth watching as capacity scales; the third is the one the new financing structures were designed, deliberately, to blur.

The demand side: what is contracted, and what is promised

Distinguishing contracted from promised demand is where the assessment has to be most careful.

Contracted or observable today:

Against that stands the demand that exists only on paper. OpenAI's roughly $1.4 trillion of announced infrastructure commitments over eight years sit against annualized revenue reported in the tens of billions, with the gap financed by successive equity and debt raises whose terms depend on continued market confidence. Vendor-financed purchases and capacity pre-payments embed demand assumptions in their own financing. Sovereign AI programs announced across dozens of countries mostly lack identified budgets. Each of these is real as an obligation or an announcement; none of them is yet revenue arriving on a meter.

The mismatch between promised and contracted demand is not necessarily fatal; venture-scale bets are how transformative infrastructure gets overbuilt and then used. But it defines the risk precisely: the boom is safe while contracted demand grows; it becomes fragile when the promised layer (OpenAI-scale commitments, vendor financing, sovereign announcements) becomes the marginal driver of capacity decisions.

What the overbuild reading has to explain

The strongest evidence against the overbuild thesis:

The demand itself is contracted, not imagined. Google Cloud's $514 billion backlog is contracted future revenue, sovereign AI programs, enterprises, AI labs, recognized over years. Nvidia's order book is tight across a customer base whose quarterly filings independently confirm the purchases. Overbuild episodes historically feature unsold capacity; today's constraint is the opposite.

Then there is the behaviour of prices, which are not collapsing but rising. In the telecom record, the tell was falling wholesale bandwidth prices while capacity grew. In AI, GPU rental prices, HBM contract prices and data-center power prices have all been rising through 2025 to 2026, and component cost inflation is a stated reason for Amazon's guidance raise. A market where the commodity output gets more expensive as capacity expands is not behaving like a glutted market.

The character of the builders separates the two eras as well. The most aggressive telecom builders were sub-scale companies whose failure would erase them. The four hyperscalers generated roughly $1.6 trillion of revenue in 2025 from search, retail, cloud, advertising and enterprise software that continue growing, the AI bet rides on cash flows that exist independent of its outcome. Even a total AI write-off, implausible as it is, would leave the builders profitable; that asymmetry has no telecom analog.

The technology, unlike fiber, improves under the market's feet. Inference costs per token have fallen roughly 10-fold per year across model generations; algorithmic efficiency gains (distillation, sparsity, better architectures) keep squeezing more capability per FLOP. The fiber glut was static glass; AI capacity gets more productive with each model generation. Falling unit costs historically expand, not shrink, total addressable demand (Jevons dynamics), which is the bull case's strongest technical footing.

And the power constraint, for all its cost, caps the downside. The constraint on many projects also functions as a market stabilizer: if demand disappoints, interconnection delays and transformer lead times naturally throttle how fast marginal capacity arrives, blunting the classic overshoot. The telecom glut was possible because fiber could be laid faster than demand grew. AI campuses cannot be energized faster than grids allow. The overbuild ceiling is, structurally, lower.

Taken together, the risks emphasised in the bear case, financing drift, layered commitments, concentration, are real but unfold over years; the strengths of the build-out, contracted demand, rising prices, funded builders, are present-tense facts. That combination describes a boom early in its risk cycle rather than late. It is also, on the historical record, how earlier booms looked in the year before their turn, which is why the indicator list that follows, not either narrative, is the better guide.

The geopolitical dimension: statecraft arrives at a capex story

A private construction program of this scale cannot stay private for long, and three forces of state now bear on it.

The supply side is shaped by export controls. US semiconductor export controls (October 2022, tightened 2023 to 2025, with further rounds under discussion in 2026) restrict China's access to frontier accelerators, while Chinese firms develop domestic alternatives under the constraint. The effect on the boom is twofold: it locks the four US hyperscalers in as the demand base for the highest-end supply, and it adds a geopolitical premium to every allocation decision, chips flow where policy allows as well as where revenue beckons. Malaysia, Thailand and Mexico have become the assembly and data-center staging grounds of the build-out partly for tariff and control-routing reasons.

Competing for the same inputs are the sovereign AI programs. Dozens of governments, the UK, France, Germany, Japan, Saudi Arabia, the UAE, India among them, announced national AI-infrastructure programs between 2024 and 2026, most pairing public capital with the same hyperscalers and the same chip vendors. In several European and Gulf projects, sovereign wealth funds have taken direct equity stakes in data-center joint ventures alongside Blue Owl-style private credit. The competition for a supply chain that is already tightly allocated means sovereign demand is additive pressure on prices and lead times, and it embeds the build-out in national security planning, with the US government treating AI compute as strategic infrastructure (presidential directives, federal land expedite programs, grid coordination initiatives through 2026).

And the power question is becoming a national planning matter. When data centers approach double-digit shares of a national grid's load, interconnection approval stops being a utility matter and becomes an industrial-policy one. The US response has included federal coordination on grid interconnection queues, expedited permitting on federal land, and explicit utility-rate debates about who pays for the transmission upgrades data centers require, ratepayers or the buyers. The outcome of those rate-design fights, now active in Virginia, Georgia and Texas, will shape where the boom can physically expand through 2030.

The security dimension has a financial side: a boom this concentrated, a handful of dominant buyers, one dominant chip vendor, a few grid regions, is a systemic surface. US policymakers now discuss AI data centers the way previous generations discussed ports and refineries: too big to fail individually, too interconnected to let fail collectively. No rescue framework exists; the boom's financing record, set out above, assumes none is needed. That assumption is itself part of the risk case.

Where the losses would sit if the demand case failed

The boom's downside, if it comes, will not land where the boom is visible. The transfer chain runs from the most visible balance sheets to the least, and each link carries a different kind of exposure:

Position in the chain Who holds it The exposure What buffers it, and what does not
First loss Equity holders Hyperscalers and Nvidia together represent roughly a quarter of the S&P 500, so a meaningful AI-linked correction comes out of the index's largest weights Diversified across retirement accounts and index-fund savers, but concentrated in the same few names
Construction risk Private credit and JV lenders (Blue Owl, Apollo, Ares and peers) Campuses financed against long-term leases from investment-grade buyers Lease cash flows senior to equity; a demand shortfall would still hit renewal and re-leasing economics at the margin
Supplier risk Bondholders and banks Oracle's debt stack, syndicated data-center loans, vendor-financing books Bank single-borrower concentration caps are the pressure point Wall Street has already flagged
End of the chain Pension and insurance capital Allocations to private-credit funds, opaque by design, marked quarterly by appraisal rather than by market Slow to signal stress; the telecom-era analog was high-yield mutual funds, the AI-era version is larger and less transparent
Contingent Taxpayers Enter only if failure meets systemically-large exposure, the scenario in which the 'too interconnected' logic overrides the no-bailout assumption Nothing in the current structure contemplates it; the same was said of the banks in 2006

None of this describes inevitable losses; it describes where losses would sit if the demand case fails. Risk transfer was the part of the telecom record nobody priced in time: the fiber ended up used, the capital ended up elsewhere, and the difference was measured in who had bought at the peak.

The revenue side: what the spending must earn

Whatever the financing structure, the spending ultimately has to be justified by the revenue it enables, and the question is what has to be true for $733 billion a year to earn its cost of capital.

Start with what the spending has already accumulated. The measurable AI-linked revenue flowing to the builders, cloud consumption, API tokens, AI subscriptions, AI-adjacent advertising improvements, sits in the low hundreds of billions annually across the industry. Against $1.05 trillion spent since 2023 and roughly $2 trillion more planned through 2028, the revenue base must multiply several-fold within the assets' economic lives. For scale: if the entire big-four capex since 2023 were required to earn a 10% return on capital, it would need to generate on the order of $105 billion of annual profit, roughly the total annual profit of a Fortune-10 company, attributable to AI capacity alone.

Where that revenue comes from matters as much as how much of it there is, because the streams differ in reliability. Contracted cloud consumption, the backlogs, is the most bankable of the streams and also the most cyclical, since enterprise cloud budgets flex with the economy. New product revenue, Copilot-style seats, API platforms, agentic software, is real but young, and its pricing is still deflating as competition and efficiency gains pass through. Advertising uplift, better targeting and generation inside existing ad systems, is the stream most tallies miss in scale: even a mid-single-digit improvement in ad efficiency across the hyperscalers' ad businesses is worth tens of billions annually, and it is already partially banked. A fourth stream, AI embodied in Amazon’s logistics, warehouses and retail operations, shows up as margin rather than revenue and is therefore invisible in the industry-wide counts.

Then there is the weight the spending itself places on the income statement. Each builder's server fleet depreciates on a 5 to 6-year schedule; buildings over 15 to 25 years (Microsoft's new schedule is 25). As the 2023 to 2026 cohort ages, annual depreciation across the four companies will rise substantially, a precise figure requires an asset-level cohort model (asset mix, in-service dates, useful lives, lease treatment) that no public disclosure supports, so any dollar estimate here would be spurious precision. What can be said without one: revenue growth and depreciation growth interact through margins, utilization and pricing, not one-for-one, matching growth rates do not mechanically produce an earnings squeeze, but rising depreciation against decelerating revenue would compress reported earnings, and that is the accounting channel through which overbuild historically surfaced. The useful-life extension Microsoft adopted in July 2026 (15 to 25 years for buildings) flatters reported earnings by slowing the charge; investors reading the boom's P&L need to normalize for such choices.

Finally, the efficiency lever cuts both ways. Inference cost per token has fallen roughly an order of magnitude per model generation. Falling costs expand demand, more use cases clear the economic bar, the Jevons dynamics that anchor the bull case. But cost curves are not price curves: what providers actually charge depends on competition, margins, model differentiation and enterprise contracts. A data center financed on 2026 prices may find 2028 prices materially lower, by how much is a market outcome, not a deduction from the cost curve. The build-out is therefore a race between volume growth and price movement, and the indicators that follow are designed to observe exactly that race.

The loops that make booms accelerate, and turn

Three feedback loops are currently running in the same direction, which is why guidance keeps rising; a fourth runs the other way and has not yet engaged.

The deepest loop runs through capability itself: more compute enables better models; better models expand use cases; expanded use cases justify more compute. That loop has run unbroken since 2023, and it is the fundamental one. Its natural governor is the pace of model improvement: enterprise procurement responds to capability, price, reliability, integration and security together, so a stretch of model-quality stagnation would slow the loop's demand leg, how much stagnation, and for how long, is not established by any dataset, which is what makes it a scenario variable rather than a measurement.

Alongside it runs the supply-chain loop. Every guidance raise commits multi-year component orders (HBM, CoWoS, transformers, turbines), which tighten the supply chain, which raises component prices and lead times, which raises the dollar cost of the next guidance round even at constant unit demand. Amazon's mid-year guidance raise cited exactly this. The loop converts scarcity into headline capex: a $733 billion year partly reflects dollar inflation of the same physical build.

The newest loop is financial. Rising asset values and revenue growth relax credit terms, which funds more capacity, which supports the asset values. Private credit's entry into data centers at scale is this loop engaging. In earlier infrastructure cycles, credit-financed expansion has tended to extend the upswing and make the eventual turn more abrupt.

Running against all three is the counterweight that has not yet engaged: every dollar spent depreciates on a 5 to 25-year schedule, and from 2027 onward the cumulative $1 trillion will generate rising annual depreciation charges across the four companies. When depreciation expense starts visibly compressing reported earnings without matching revenue growth, boards get their first hard accounting signal that capacity has outrun demand. This is the mechanism most likely to surface first from 2027; in earlier infrastructure cycles, the turn showed up in income statements before it showed up in order books.

Where the boom physically lives: the regional map

The build-out has a geography, and the geography has frictions.

Northern Virginia remains the world's largest data-center market, tens of gigawatts of concentrated load on a single grid region (PJM), where interconnection delays and local moratoria have already throttled expansion and pushed buyers south and west. Texas leads the new build on power availability and permitting speed, with ERCOT connection volumes dwarfing other regions and gas-turbine orders priced accordingly. Georgia, the Carolinas and Ohio absorb spillover with utility-scale campus announcements from every major builder. The Pacific Northwest competes on hydro and cooling; Arizona and Nevada on land and solar-adjacent sites, with water politics as the counterweight.

Internationally, Ireland hit its grid ceiling, national-level grid connection freezes pushed new projects to continental Europe; Frankfurt, London (West), Paris and Amsterdam (the 'FLAP' markets) carry European load under land and power constraints; Malaysia's Johor corridor and Thailand have become the assembly-and-cloud belt serving Southeast Asia; Japan, Singapore and Australia anchor APAC enterprise demand. Gulf sovereign projects (Saudi Arabia's HUMAIN, UAE's Stargate-linked campuses) pair sovereign capital with US technology in a bid to import the capability entire.

Two conclusions follow from the map. The boom's binding constraint differs by region (power in Virginia, water in Arizona, land in Ireland), so national aggregates hide it. And the growth of new regions (Texas, Johor, the Gulf) is the physical expression of the boom's second phase: the marginal campus now carries higher cost and longer lead times than the average already built.

The macro footprint: what the boom does to prices, wages and grids

At 2.3% of GDP on the mechanical comparison, the boom stops being a sector story and starts leaving visible traces across the macroeconomy.

The most visible trace is in goods prices. The demand wave has repriced its inputs: copper for grounding and busbars, aluminum for racking, transformers and switchgear, gas turbines (Siemens Energy's backlog near 69 GW, GE Vernova's near 100 GW, both with multi-year delivery horizons), and HBM memory, where contract prices rose sharply through 2026 and Samsung, SK Hynix and Micron allocated capacity years forward. General electrical-equipment inflation is now partly an AI-data-center phenomenon, a cost passed through to ordinary construction.

Labor carries the next one. Skilled electrical trades are the bottleneck profession of the build-out: data-center electricians and linemen command premiums, union training pipelines have expanded to multi-year backlogs, and the build-out's labor demand is large enough to appear in regional wage statistics.

Grid planning and rates carry a third. In PJM, the grid region covering Northern Virginia, capacity-market prices cleared at records in the 2025 and 2026 auctions, with data-center load growth cited as a driver, a cost that flows to all regional ratepayers, feeding the political debate over who should pay for AI's electrons. Utilities across the American South have introduced or proposed large-load tariff classes specifically for data centers.

And capital markets carry the last: Data-center private credit, data-center REITs and AI-linked issuance have become a distinct asset class. When a category grows this fast, its marginal pricing influences broader credit spreads, another channel through which a demand disappointment would propagate beyond technology.

These traces are why the GDP-relative framing matters: the boom's costs are already socialized through prices and grids even while its returns remain privatized and uncertain. That asymmetry, diffuse costs, concentrated bets, is the classic signature of infrastructure at scale, and the reason the financing question is the central unresolved variable.

Three paths the spending could take

Scenario A: absorption. The condition is enterprise and consumer AI revenue continues compounding at 30%+ annually, backlogs convert on schedule and power arrives at the pace the LBNL midrange assumes, capex growth decelerates from ~90% to 20 to 30% annually by 2028 as the base effect dominates, the build-out matures the way cloud did after 2016, and the hyperscalers' free cash flow recovers as revenue catches up. Observable markers: backlog conversion rates, token revenue disclosures, data-center utilization, FCF trajectory through 2027.

Scenario B: pause and digest. The trigger is a demand disappointment, enterprise budgets flattening, a model-quality plateau, or a macro downturn, coinciding with rising component costs and power delays; orders then slow before capacity does: 2027 capex flattens or falls 20 to 30%, lease commencement slows, second-tier suppliers (server OEMs, memory, power equipment) feel it first, and hyperscaler FCF recovers as depreciation catches up. Markets reprice AI-linked equities and credit; the fiber-glut analog appears in generic inference pricing. No systemic crisis, the builders remain profitable, but a lost year for suppliers and a visible write-down cycle in speculative projects. Observable markers: order-book revisions at Nvidia and server OEMs, data-center lease commencement rates, hyperscaler guidance language, credit spreads on AI-linked debt.

Scenario C: overbuild and work-down. In this path contracted demand keeps capacity expanding to 2028 on assumptions that then prove optimistic, generic token prices sag toward commodity economics, specialized assets (training campuses) are repriced or repurposed; financing structures (JVs, private credit, vendor financing) shift losses to non-bank lenders and, via them, pension and insurance capital. The physical infrastructure finds downstream uses, agents, robotics, scientific computing, media generation, the way dark fiber eventually carried streaming, but on a multi-year horizon and at investor cost. Observable markers: token price deflation, asset impairments, JV repricing events, secondary-market discounts on data-center paper.

The evidence in mid-2026, backlogs, utilization, tight supply chains, sits closest to Scenario A's conditions; the financing drift and the layered commitments are the leading indicators of Scenario B's and C's preconditions. Both can be true at once, a productive boom whose marginal projects are speculatively financed.

Indicators that would confirm or weaken each scenario

The thresholds below are judgment calls, set to make direction visible, not derived from a statistical model.

  1. Hyperscaler free cash flow. Does combined quarterly free cash flow of the five largest builders recover through 2027 (absorption) or stay pinned near zero (financing stress)? FactSet and company filings publish this quarterly.
  2. Backlog conversion. Does Google Cloud's $514bn backlog convert on schedule, and do AWS/Azure backlog disclosures grow? Slowing conversion with growing backlog is the classic overbuild signature.
  3. Nvidia's concentration. Does the direct-customer share reported in Nvidia's filings keep rising? Rising concentration would mean the supply chain is betting harder on fewer buyers.
  4. Power. Do LBNL/DOE electricity-share estimates track toward the reference case on schedule? Interconnection delays appearing in ISO queues and transformer lead times would shift the binding constraint earlier.
  5. Credit spreads. Do spreads on data-center JVs, private credit and AI-linked corporate paper widen ahead of equity volatility? Credit markets have historically priced financing stress earlier than equity markets in some episodes, a hypothesis worth monitoring here rather than an established regularity.

Limitations

Conclusion

Measured against the historical programs examined above, the telecom build-out of the late 1990s, the railway and canal manias, the four companies' own spending history, no private investment program in those comparisons approaches the AI build-out's scale or pace. $1.05 trillion spent in three and a half years; $733 billion more planned in one year; close to twice the telecom bubble's peak intensity relative to GDP on the mechanical comparison; a funding model adding credit at the margin.

None of that arithmetic settles whether it ends in absorption, pause or overbuild. The demand evidence is genuinely strong today, the builders are genuinely strong balance sheets, and both facts coexist with layered commitments ($1.4 trillion in announced OpenAI commitments, $662 billion of uncommenced leases, $237.7 billion of Meta contracts) whose payment schedules now extend a decade into an uncertain future.

What the record establishes is which indicators separate the scenarios while the distinction can still matter. Free cash flow, backlog conversion, supplier concentration, power interconnection and credit spreads each bear on a different link in the argument: whether demand holds, whether contracted revenue converts to cash, whether the supplier base strains, whether physical capacity arrives, and whether lenders price the leverage before equity markets do. None of them calls a top on its own; together they keep the deciding evidence visible while the outcome is still open.

Appendix A. Data tables

Table A1. Big-four capex, headline definitions, $ billions

Company 2023 2024 2025 2026 (plan) Source basis
Amazon (cash capex) 48.1 77.7 128.3 ~220 10-K filings; Feb & mid-2026 guidance
Alphabet (purchases of P&E) 32.3 52.5 91.4 195 to 205 10-K filings; Apr & Aug 2026 guidance
Meta (P&E + finance-lease principal) 28.1 39.2 72.2 130 to 145 Filings; Q2 2026 guidance
Microsoft (cash P&E, fiscal years) 28.1 (FY23) 44.5 (FY24) 64.6 (FY25) ~175 (CY26, adj.) Filings; Jul 2026 guidance
Combined (calendar-aligned) ~144 ~225 ~380 720 to 745 Reconstruction from filings

Table A2. Derived measures (calculations from the filings above)

Measure Value Basis
2026 capex midpoint $733bn Sum of guidance midpoints
2026 capex ÷ US GDP (mechanical) ≈2.3% $733bn ÷ $32.38tn (IMF WEO Apr 2026); global capex vs US GDP, illustrative, not US-located investment
Telecom 2000 capex as % of GDP ≈1.2% ~$120bn ÷ ~$10.25tn (2000 US GDP)
Cumulative big-four capex, Jan 2023, Jun 2026 ≈$1.05tn Filing reconstruction (±$50bn)
Meta 2026 capex ÷ 2025 revenue ≈69% $137.5bn ÷ ~$200bn
Microsoft 2026 capex ÷ 2025 revenue ≈62% $175bn ÷ ~$281.7bn
Alphabet 2026 capex ÷ 2025 revenue ≈50% $200bn ÷ $402.8bn
Amazon 2026 capex ÷ 2025 revenue ≈31% $220bn ÷ $716.9bn
Nvidia DC revenue, annualized ≈$356bn $89.0bn (Q2 FY27) × 4
Nvidia DC revenue ÷ big-four capex ≈49% $356bn ÷ $733bn, scale equivalence, not a procurement share

Table A3. Financing and commitments

Item Value Date Source
Meta-Blue Owl Hyperion JV financing $27bn Oct 2025 Global Data Center Hub
Meta subsequent data-center financing ~$29 to 30bn Late 2025 Bloomberg/Octus tallies
AI-sector corporate debt issuance, 2025 ~$141bn 2025 Octus via Bruce Richards tally
Uncommenced data-center leases, 5 companies $662bn Feb 2026 Moody's
Meta non-cancelable contractual commitments $237.7bn 31 Mar 2026 Meta filings
Meta uncommenced leases $182.9bn 31 Mar 2026 Meta filings
OpenAI announced infrastructure commitments ~$1.4tn / 8 yrs Nov 2025 FT (aggregate of separately announced deals)
Oracle-OpenAI cloud contract up to $300bn Sep 2025 WSJ/Reuters
Google Cloud backlog $514bn Mid-2026 Alphabet Q2 2026
Hyperscaler cumulative capex 2026 to 2030 >$5tn Proj. S&P Global MI; Goldman ~$5.3tn (big four)
Siemens Energy gas-turbine backlog ~69 GW Q3 FY2026 Utility Dive (lead times 3+ years)

Table A4. Electricity

Measure Value Source
Global data-center electricity share, 2025 ~1.5% Our World in Data (Jul 2026)
US data-center share, 2023 4.4% DOE/LBNL (Dec 2024)
US data-center share, 2028 (DOE 2024 report scenarios) 6.7 to 12% DOE (Dec 2024)
US data-center share, 2030 reference case 11.8% (649 TWh) LBNL 2025 Update (Jun 2026)
US data-center share, 2030 (scenario range) 9.5 to 15.3% LBNL 2025 Update (Jun 2026)
Global demand, 2030 (IEA base case) ~945 TWh IEA Energy & AI

Appendix B. Methodology notes

Capex reconstruction. Calendar-year alignment: Alphabet, Amazon and Meta report on calendar years and were summed directly under headline definitions. Microsoft reports on a June fiscal year; its quarters were mapped to calendar quarters using its quarterly disclosures, and its calendar-2026 estimate uses the company's July 2026 guidance (~$175bn) following the useful-life reclassification, with the prior ~$190bn estimate noted where relevant. "Headline definitions" means: Amazon = cash purchases of property and equipment; Meta = purchases of property and equipment plus principal payments on finance leases; Microsoft = cash P&E with finance-lease commencements noted; Alphabet = purchases of property and equipment. Figures in billions of nominal US dollars; no inflation adjustment applied within the 2023 to 2026 window (adjusting 2023 dollars upward would make the growth rates marginally smaller).

GDP-relative comparison. Numerator: 2026 guidance midpoint, $733bn, the companies' global capex. Denominator: IMF WEO April 2026 US nominal GDP, $32.38tn. The result (2.26%) is a mechanical scale comparison between global corporate spending and US national accounts, not a measure of US-located investment; the US-only capex share is not disclosed and would lower the ratio. Telecom numerator: ~$120bn US telecom capex in 2000 (Fabricated Knowledge's compilation of carrier disclosures and industry tallies; Princeton's Starr and FRB studies corroborate the order of magnitude), a US-sector figure, closer in kind to its denominator. Denominator: US nominal GDP 2000, ~$10.25tn (BEA). Result: 2.26% vs 1.17%, with the numerator-mismatch caveat stated wherever the pair appears. Sensitivity: if the telecom numerator is taken at $150bn (upper industry estimates), the ratio is 1.46%, and the AI boom remains roughly 1.5 to 2× the telecom peak under any pairing the estimates support.

Ratios to revenue. 2026 capex midpoints divided by fiscal-2025 revenues as reported in Q4 2025 releases (Amazon $716.9bn; Alphabet $402.8bn; Meta ~$200bn; Microsoft ~$281.7bn, Microsoft's figure is its fiscal year ended June 2025, the other three calendar 2025). This pairs a forward spend with trailing revenue by design; pairing against 2026 revenue estimates would lower every ratio by 10 to 20%.

Nvidia annualization and concentration. Q2 FY2027 data-center revenue ($89.0bn, quarter ended 26 July 2026) × 4. Nvidia's fiscal calendar runs ahead of the calendar year; annualization is indicative, not a forecast, and quarterly revenue is rising, so the trailing-four-quarter figure is somewhat lower. The resulting $356bn is compared with big-four capex as a scale equivalence only, Nvidia's data-center revenue includes customers beyond the four companies, so it is not their procurement share of equipment spending. Customer-concentration figures are reported by vintage: H1 FY2027 from Nvidia's 10-Q (16/15/13%); Q4 FY2026 from daloopa's reconstruction (~61% across four customers); the two are not directly comparable.

Commitment figures. Taken directly from filings or named-source reporting as cited; the $1.65 trillion aggregate is reported but not reproduced here as a sum, because its components mix categories (see the financing section). OpenAI's $1.4tn is treated as an aggregate of separately announced commitments. Scenario conditions above are qualitative descriptions tied to observable indicators; no probability is assigned.