Skip to content
Rush Commerce
Field Notes3 min read

Alphabet's $205B capex: your cloud bill isn't coming down

Alphabet raised 2026 capex to as much as $205B with a $514B cloud backlog. Here's what the AI infrastructure spend means for the cloud and tokens you rent.

Alphabet reported Q2 2026 on July 22, and the number that moved the stock wasn't revenue — it was the spending plan behind it. The company raised its full-year 2026 capital expenditure guidance to as high as $205 billion, up from the $180–190 billion it forecast a quarter earlier. If you rent cloud or buy tokens, that guidance is a forward look at your cost floor.

What actually happened

Per Alphabet's earnings release and CNBC's coverage:

  • Consolidated revenue rose 24% to $119.8 billion, beating estimates, with operating margin at 34%.
  • Google Cloud grew 82% to $24.8 billion — an acceleration, not a plateau.
  • The Cloud backlog hit $514 billion, up from $460 billion the prior quarter. That's committed demand the company hasn't filled yet.
  • Q2 capital expenditures hit a record ~$44.9 billion, pushing free cash flow to roughly -$5.9 billion.
  • Management flagged modest Cloud margin pressure ahead as it leans on third-party capacity "as a bridge" to meet demand.

Read those together. The largest infrastructure buyer on earth is spending a record amount, running cash flow negative to do it, and still has a half-trillion-dollar backlog it can't serve without renting someone else's capacity. Supply is not catching demand.

Why this matters for your business

When a hyperscaler is capacity-constrained and warning about its own margins, the discount you were hoping for isn't scheduled. The AI and cloud you rent are priced off this supercycle, and the trend is up-and-to-the-right on spend, not down on price. Betting your unit economics on "compute gets cheap soon" is betting against the people building the compute.

What we do about it instead of hoping:

  1. Price the work, not the tokens. Track cost per completed task, not per-1M-tokens. A cheaper model that needs three tries is more expensive.
  2. Keep the stack portable. If a provider's price or capacity changes, you want a config swap, not a rebuild — the whole argument for keeping your software portable.
  3. Right-size the model to the job. Most workflows don't need the frontier tier. Route the boring 80% to the cheapest model that clears your quality bar.

The capex is a signal, not a threat — if you build so the cost floor can move without breaking you.

Key takeaways

  • Alphabet raised 2026 capex guidance to as much as $205B (from $180–190B) on July 22
  • Google Cloud grew 82% to $24.8B, but backlog hit $514B and the company will use third-party capacity as a bridge
  • Record ~$44.9B quarterly capex pushed free cash flow negative — supply isn't catching demand
  • Your operator move: measure cost-per-task, keep the stack swappable, and route work to the cheapest model that passes

Worried your AI costs are locked to one vendor's price? We build vendor-agnostic systems where the model is a config value, not a foundation. See how we keep your stack portable — or bring us the workflow you're worried about.

Sources: Alphabet Q2 2026 results, CNBC.

  • #cloud-costs
  • #ai-infrastructure
  • #capex
  • #vendor-pricing
  • #portability
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.