Summary
- Meta raised the lower end of its 2026 capital expenditure forecast from $125bn to $130bn.
- Second-quarter capital expenditure reached $31.08bn, close to the company’s operating cash flow for the period.
- The programme shows how AI infrastructure is absorbing cash well before all associated services produce mature returns.
Meta expects to spend between $130bn and $145bn on capital expenditure during 2026 as the company continues to expand the data centres, servers, networks, and supporting systems behind its AI strategy.
The company raised the lower end of its forecast from $125bn while leaving the upper limit unchanged. The revised range includes principal payments on finance leases and follows $31.08bn of capital expenditure in the second quarter alone.
Meta generated $31.86bn of operating cash during the quarter, leaving free cash flow of $784m after capital spending and other adjustments. The comparison shows how heavily the current infrastructure programme is drawing on cash generated by the company’s advertising business.
Long-term debt stood at $83.66bn at the end of June, while cash, cash equivalents, and marketable securities totalled $90.26bn. The balance sheet gives Meta room to maintain the programme, but the scale of expenditure has sharpened investor scrutiny over the timing and durability of AI-related returns.
The capital budget is not a direct measure of data centre construction. It also includes servers, network equipment, finance leases, land, buildings, and other infrastructure supporting Meta’s AI efforts and established services.
Even so, the forecast provides a measure of the physical investment required by one of the world’s largest technology companies. AI systems depend on a supply chain extending from semiconductor fabrication and rack integration to substations, cooling plant, transmission capacity, and long-lead electrical equipment.
Cash is moving before workloads mature
Meta’s investment cycle illustrates the mismatch between infrastructure lead times and product revenue. Large data centres and grid connections can take years to plan and build, while AI services, model architectures, and customer demand can change far more quickly.
The company must therefore commit capital against forecasts of future compute requirements. Underbuilding risks leaving products constrained by unavailable capacity. Overbuilding creates underused assets, depreciation, financing costs, and facilities designed around hardware that may be superseded before the campus is fully occupied.
The narrowing of Meta’s guidance raises the minimum expected outlay without increasing the ceiling. It indicates greater certainty that spending will remain near the upper part of the previously announced range rather than a new maximum ambition.
Second-quarter expenditure also demonstrates how quickly physical deployment can affect cash generation. Meta reported $31.08bn of capital expenditure against $31.86bn of operating cash flow. That does not imply the company cannot finance the programme, but it leaves less internally generated cash available for acquisitions, debt reduction, shareholder returns, or unexpected costs.
The infrastructure is being developed across a period of tight equipment supply and increased utility scrutiny. High-voltage transformers, switchgear, generators, cooling components, and skilled construction labour are required across multiple hyperscale programmes at the same time.
Grid access remains a separate constraint from corporate spending power. A technology company may be able to fund a campus, but it cannot purchase an immediate transmission connection where the network lacks capacity or reinforcement work has not been completed.
That distinction is especially relevant in Europe, where planning requirements, energy reporting, grid queues, and regional restrictions can impose longer development paths than capital budgets alone suggest. Meta’s spending provides a demand benchmark, not a guarantee that capacity can be delivered in every target market.
The programme also shifts risk through the supply chain. Contractors, developers, equipment manufacturers, utilities, and colocation providers are being asked to reserve manufacturing slots and construction resources against demand forecasts originating with a small number of hyperscale buyers.
Meta’s 2026 range is therefore both a corporate finance figure and an infrastructure signal. A substantial portion of the company’s cash generation is being converted into long-lived physical systems before the full commercial shape of the AI market is settled.
Execution will be measured in more than model performance. The investment must produce powered, cooled, networked, and reliably operated capacity at a cost that future AI revenue can support.
