The capital spending outlook.
Company expectations. Our forward view. A broader map of the industries turning investment into capacity and earnings.
Where the next investment dollar goes.
Compare company disclosures with our scenarios, then explore the businesses exposed to each spending layer.
New house scenarios · 2026-09-13. Calendar-year USD billions in issuer-specific investment definitions. The sum can include leases, accruals and overlapping assets; it is not reconciled cash CapEx, AI-only spending or a deduplicated market size.
Who contributes?
Company bars show the selected annual house scenario, USD bn.
Scenario paths are assumptions, not probability intervals or street consensus. The base case has spending growth decelerating as deployments mature.
Budgets become systems. Systems become earnings.
Choose a layer. Each company card distinguishes a sourced buyer relationship from broader industry exposure.
Supplier dollars are not allocated: public disclosures do not reconcile purchases by recipient. Foundry, equipment and component companies occupy different tiers and cannot be summed as incremental spending.
Explore the animated software-to-hardware system ↗Company guidance, new house forecasts and accounting definitions
summary. Reported actuals and management guidance retain each issuer's accounting definition and fiscal period. The ten-company future series is a modeled calendar-year investment envelope using explicit company scenarios, not a reconciled cash PP&E total or an AI-only market-size estimate.
guidance. Guidance is the company's expectation as of its source date, not consensus. Where first-party call text was inaccessible, a clearly identified reproduced issuer transcript, syndicated issuer presentation or newswire account is linked. No 2027–29 street consensus was available and none is fabricated.
houseView. These numeric forecasts were newly drafted for this site on September 13, 2026. They extend the original Ideas' demand-to-memory-to-equipment and power-constraint reasoning; they are not historical published Alex Chen forecasts. Base, downside and upside have no assigned probabilities.
calendarAlignment. Microsoft's $175bn guide is explicitly CY2026. Oracle's $90–95bn guide is FY2027 ending May 31; its CY2026 house estimate uses a separately explained quarterly bridge. Alibaba's March-end fiscal actuals remain separate from its calendar house estimates. All other modeled years run January–December.
fx. CNY house projections use a constant 6.7851 CNY per USD, the June 30, 2026 convenience translation rate cited in Baidu's release. This is a scenario assumption for future years, not an FX forecast. Historical actuals stay in original currency.
aggregation. Only houseForecast values are summed for the annual envelope. Differences in gross/net, cash/accrual, finance leases and construction-in-progress definitions remain. Intercompany leases or shared projects may represent overlapping underlying assets, so this is a budget monitoring sum, not an audited or deduplicated economic-investment total. Changes reflect the stated model assumptions.
doubleCounting. Cloud procurement commitments, customer prepayments, financing announcements, supplier revenues, foundry capital budgets and landlords' spend are not added to the buyer total. A supplier's sale may already include memory or components bought from another supplier.
relationships. Paths show operational transmission and relevant companies. Only sourced relationships are labeled disclosed; no inferred supplier allocation dollars are published. Supplier membership is not a buy recommendation.
excluded. Private-company spending without comparable published financials—including OpenAI, Anthropic, xAI and ByteDance—is not assigned a numeric contribution. Their workload demand is visible in the software layer. Multi-year pledges are not treated as one-year expenditure.
Private workloads from OpenAI, Anthropic, xAI and ByteDance matter to demand, but comparable public CapEx contributions are not assigned.
Original same-quarter cash PP&E comparison ↓