Hyperscaler Capex Q3 2026: What Microsoft, Alphabet and Meta Spending Tells Buyers of AI Services

Alphabet: The pack reports $195 billion–$205 billion of 2026 guidance and about $45 billion of Q2 spending. Its reported split—roughly 60% servers and 40% data centers and networking—suggests substantial funding for compute, but “servers” does not mean GPUs alone. Some expenditure supports existing and non-AI services, while buildings and networks are prerequisites rather than immediately rentable accelerator instances. Confirm the guidance and mix against Alphabet’s Q2 2026 earnings materials and the SEC exhibit before using them in a negotiation.

Meta: The pack reports $130 billion–$145 billion of 2026 guidance and about $31 billion of Q2 capex. Its principal route to a return is improving advertising, engagement and its own AI products—not selling buyers a general-purpose public-cloud instance. Meta’s spending can affect industry demand for equipment, but it should not be added wholesale to a forecast of rentable Azure or Google Cloud capacity. Confirm amounts against Meta’s Q2 2026 results release before contractual use.

Across all three, total capex is not an “AI-only” number. The evidence pack supplies no verified dollar allocation exclusively to AI hardware.

What the H2 step-up implies

The sharpest near-term signal in the pack is Microsoft’s reported guidance for more than $50 billion in calendar Q3 capex against approximately $41 billion in Q2: a sequential increase of more than 20%. It indicates an accelerated investment schedule, not that every purchased GPU is installed, networked and available to a customer in Q3. Buyers should distinguish committed spend from serviceable capacity in their required regions. (Microsoft FY2026 fourth-quarter investor materials)

Alphabet’s higher annual range and reported server-heavy Q2 mix likewise point to continued capacity investment. Neither observation establishes the date or price at which a particular GCP accelerator will be offered. The proposed blanket claim that H2 requires 25%–40% or greater quarter-on-quarter jumps is stronger than these figures support: Microsoft’s stated Q3 comparison establishes only more than 20%, while the pack lacks the quarterly guidance needed to calculate comparable jumps for Alphabet and Meta.

For Meta, the distinction is more fundamental. An H2 acceleration would principally signal more internal AI capacity; it would not, without a separate product announcement, expand the pool a cloud buyer can reserve. Nor does rising spend alone prove either durable scarcity or an eventual overbuild. Those outcomes depend on delivery, utilization and demand.

Decision rules for AI-services buyers

• Lock a multi-year commitment when scarcity is specific and priced. Require the vendor to name the accelerator class, region, quantity, availability date and remedies for missed delivery. Commit against a measured workload and minimum utilization case, not a companywide capex headline. Seek terms for upgrading or reducing capacity if the intended hardware or model economics change.
• Keep a second cloud viable when portability is worth more than the commitment discount. This is especially relevant if the application can move inference traffic, but its training or data-residency requirements concentrate risk in one region. Price the cost of portability—including data movement and operational duplication—before paying for it.
• Treat “short on GPUs” as credible only at workload level. A vendor should be able to identify which instance type and region are constrained, the reservation lead time and the alternative configuration. Large server budgets make supply constraints plausible; they do not prove a shortage for every service or justify an unspecified premium.
• Do not use Meta’s capex as a cloud-supply offset. Its infrastructure spending may matter to the broader equipment market, but it does not answer whether Azure or GCP can deliver your reserved capacity.

Evidence pack checklist for a negotiation

Pull the latest earnings release, call transcript and 10-Q or 10-K for each relevant provider. Record:

1. Definitions and periods: fiscal versus calendar quarter; reported cash capex versus equipment acquired under leases; any changed useful-life assumptions. Reconcile Microsoft’s approximately $175 billion and $190 billion framings before comparing providers.
2. Guidance and actuals: the annual range, year-to-date spending, Q2 amount and any explicit next-quarter guide. Calculate implied remaining spend, but do not manufacture a quarterly delivery schedule from an annual total.
3. Asset mix: disclosed servers or short-lived equipment versus buildings, power and networking. Mark any AI-only estimate as an estimate, not a reported figure.
4. Contract evidence: region-specific quotas, reservation start dates, capacity guarantees, substitution rights, service credits and price-adjustment terms. These determine whether industry investment becomes usable capacity for the buyer.

Sources

• Microsoft, FY2026 fourth-quarter investor materials — primary IR reference for Microsoft reporting.
• Alphabet, Q2 2026 Earnings Call materials — primary IR event page for Alphabet guidance and CapEx mix.
• Alphabet, Q2 2026 results exhibit (SEC) — primary filing exhibit.
• Meta, Reports Second Quarter 2026 Results — primary IR release for Meta CapEx and guidance.
• CFO Dive, Microsoft spending and lease-accounting report — secondary discussion of Microsoft guidance framing.
• MLQ, four-company 2026 capex range — secondary aggregate; reconcile accounting bases before comparison.