Global active AI compute reaches $1 trillion in ownership cost

What AI 2027 Predicted

The late-2026 dashboard specifies $1 trillion as the ownership cost of global active AI compute. The supplement includes servers and networking and excludes datacenter buildings, cooling and power infrastructure. It assumes roughly four years of chip use. Its separate 2026 table implies approximately $0.8T from 40 million H100-equivalents at $20,000 each. We retain the dashboard’s $1T target while recognizing that these scenario quantities are approximate. AI 2027 Compute Forecast

How We Track This

We estimate the acquisition value of computing hardware that is commissioned and still active. Annual spending, future commitments and depreciated book values remain separate. Our working service-life and commissioning assumptions are explicit; the original source does not supply a fully reconciled valuation formula for its rounded dashboard.

We compare a supplier-shipment reconstruction with an owner-investment reconstruction. These approaches overlap in underlying market evidence and are never added together. Calculation inputs and results

Current Evidence

The buildout is substantial and close to the scenario’s scale. Working central estimates are $0.67-0.71T now and $0.83-0.86T at year-end. They are model estimates, not measured global inventories.

Approach	September 7 estimate	December 31 estimate	Selected year-end scenarios
Supplier shipments and hardware costs	$0.67T	$0.83T	$0.68-1.01T
Owner expenditure converted to active hardware	$0.71T	$0.86T	$0.52-1.18T

These scenario ranges are neither confidence intervals nor exhaustive uncertainty bounds. The owner calculation is a corroborating sensitivity exercise; its common hardware/AI allocations and incomplete cash-to-acquisition reconciliation do not provide an independent measured total.

The supplier calculation starts with $331.166B of estimated accelerator value in Epoch’s 2023-2025 shipment cohorts. It adds modeled commissioned 2026 values of $227B NVIDIA, $10B AMD, $24B Google and a $25B other-supplier allowance, then applies a 1.35 residual server/network multiplier: $617.166B × 1.35 = $833.174B. The central commissioning lag is approximately two months. Every 2026 extension and the residual multiplier is an analyst assumption informed by the cited data. Epoch chip-sales data · Shipment dataset · Methodology

Current supplier results substantiate the scale of deliveries. NVIDIA reported $89.023B Data Center revenue for the quarter ending July 26 and guided total next-quarter revenue to $108B. Broadcom reported $16.7B AI semiconductor revenue for its quarter ending August 2 and guided $21.7B for the next quarter. Their networking and accelerator components must be reconciled before constructing hardware totals. NVIDIA results · Broadcom results

Owner disclosures also show realized investment. Amazon reported $96.3B first-half cash capex, plus $20.620B additional acquired-but-unpaid equipment and $2.128B finance-lease equipment additions. Microsoft’s latest $175B calendar-year capex outlook reflects a lease-classification change; management said underlying investment expectations were unchanged. This updates the earlier $190B headline without implying a physical investment cut. Amazon filing · Microsoft July call

Counterevidence & Limitations

The central year-end cases remain 14-17% below the literal $1T target. The hardware multiplier needed to cross $1T in the supplier central case is approximately 1.62. Raising the custom-silicon multiplier to 2.0 while retaining 1.35 for NVIDIA/AMD produces approximately $0.90T. Treating 10% of the 2023-2025 cohort as inactive lowers the central estimate to approximately $0.79T.

Epoch values are reconstructions rather than invoices. Some server content is already captured, while historical Trainium1 and Siyuan 590 records have no assigned dollar value. Custom-chip overhead, untracked owners, retirement and payment timing create uncertainty in both directions. We do not count unfinished campuses or cloud customers’ rental payments as additional global hardware.

The assessment remains emerging for the exact cutoff, with strong evidence of investment progress. Earlier annual-guidance evidence remains useful context; the new calculation explains why it cannot alone settle the stock target.

What Would Change Our Assessment

• Confirm: A reconciled inventory supports at least $1T active computing hardware in the specified period.
• Strengthen: Owner-specific asset additions, hardware costs and commissioning records narrow the estimates above the threshold.
• Weaken: Comparable deadline-period inventories show a material shortfall. Missing disclosure alone does not establish failure.

Update History

Date	Update
2026-09-07	NVIDIA and Broadcom reported large realized AI infrastructure revenues, while Microsoft explained a lease-classification change in its guidance. Shipment and owner-investment reconstructions place central end-year active-hardware estimates at $0.83-0.86T; higher scenarios cross $1T. The exact outcome remains unresolved.
2026-09-06	Assessment revised from on-track (0.95) to emerging (0.65). Investment guidance supports a large pipeline, but active-compute ownership value is a stock measure and remains unresolved.
2026-08-03	Amazon reported that property and equipment purchases increased by $66.1B year over year, primarily reflecting AI investment. This adds realized-spending evidence for the cumulative capex trajectory, while the figure remains broader than AI-only capex.
2026-07-27	Alphabet raised 2026 capex guidance to $195B-$205B. NVIDIA also announced a $500B-plus SK Group initiative under letters of intent and proposed or conditional NAVER-related financing. The cumulative trajectory remains on track, while announced amounts remain separate from realized AI-only spending.
2026-06-01	Added first-party Meta and Microsoft capex guidance. Meta raised 2026 capex guidance to $125B-$145B, and Microsoft said it expects roughly $190B of calendar-year 2026 capex. These figures support the cumulative $1T trajectory, while still requiring care around AI-specific versus broad cloud capex.
2026-05-18	TrendForce forecast combined 2026 capex for the world’s top nine CSPs at about $830B, up 79% year over year. This supports continued very large AI infrastructure spending while preserving caution about cumulative versus annual capex definitions and broad CSP capex versus AI-specific spend.
2026-05-04	Post-Q1 earnings compilations put 2026 capex guidance for Google, Amazon, Microsoft, and Meta near $725B, up 77% from 2025. This further strengthens the cumulative $1T trajectory, though final confirmation should still distinguish realized AI-specific spend from broad capex guidance. Confidence adjusted 0.90 → 0.95.
2026-03-23	Dell’Oro Group reports datacenter capex surged 57% in 2025, with top 4 US cloud providers up 76% (Dell’Oro). Multiple sources now project 2026 Big Four spending at $650-700B: Amazon ~$200B, Google ~$175-185B, Microsoft and Meta rounding out (TechCrunch). DataCenter Knowledge estimates total hyperscaler+sovereign capex exceeding $600B in 2026 alone. Cumulative $1T is now a certainty. No status change.
2026-03	2026 commitments from major tech companies suggest cumulative $1T AI capex is effectively locked in by late 2026.
2026-01	Meta guided $116–118B infrastructure spend for 2026 alone. Amazon guided “even more” than 2025’s $125B total. Combined with Stargate commitments and sovereign AI spending globally, cumulative 2024–2026 global AI capex is on track to reach $1T by mid-to-late 2026.
2025-10	Q3 Big Tech capex comes in at $114B in a single quarter — 76% YoY growth. Anthropic’s 1M TPU deal represents tens of billions in additional compute commitment. CEOs describe power availability, not capital, as the primary constraint.
2025-08	Combined Big Tech full-year 2025 capex guidance reaches approximately $364B. Nvidia’s Q2 earnings confirm hyperscaler demand is absorbing all available Blackwell supply ($27B in a single quarter). At current pace, Big Four alone will approach $400-450B for 2025. Trajectory toward $1T cumulative (2024-2026) increasingly plausible.
2025-07	Early 2025 capex numbers from hyperscalers (Microsoft $80B, Google $75B, Meta $65B) align with trajectory toward $1T cumulative.
2025-05	Stargate UAE 1 GW announcement joins US Stargate $500B commitment. Big Tech combined 2025 capex guidance approaching $364B. The cumulative trajectory toward $1T by mid-2026 is becoming visible in disclosed numbers.