🏛️

GOOG

💡
Thesis. Google Cloud grew 82% with backlog at $514B and it raised capex to $205B — the largest single buyer of almost everything on this watchlist, now spending so hard that free cash flow went negative and the stock fell on a beat.
Reviewed 19 Sep 2026, after Q2 2026 results on 22 Jul. Next review after Q3 2026 earnings on 28 Oct 2026.

What it does

Alphabet is unusual in this file: it is mostly a buyer of AI infrastructure rather than a seller. Two engines. Google Services — Search, YouTube, subscriptions, devices — is the cash machine. Google Cloud rents compute, increasingly to AI labs, on both GPUs and Google's own TPUs (custom accelerators co-designed with Broadcom).
The most important thing to know about the Q2 print is that the reported EPS is not operating profit. Other income went from $2.7B to $98.0B on unrealised mark-to-market gains on the Anthropic and SpaceX stakes. That is real accounting and not real cash, and it reverses as easily as it appeared. Read operating income instead.
Alphabet's capex line is the revenue line for a large part of the rest of this database. The number that frightens its shareholders is the number that funds everyone else.

Bull case

  • Q2 2026 revenue $119.8B, +24% YoY — a twelfth straight double-digit quarter, beating ~$116.9B consensus
  • Operating income $40.8B, +30%, at a 34% margin (+2pts) — the honest scoreboard, and it is improving
  • Google Cloud $24.8B, +82%, with operating income more than tripling to $8.8B
  • Cloud backlog $514B, up $50B in a single quarter from $462B. Just over 50% is expected to convert within 24 months. The most bullish datapoint on the page
  • Google Services $94.5B, +15% — Search +17%, YouTube ads +13%, subscriptions +15%. "AI kills Search" is not showing up in the numbers yet
  • TPU systems began delivering to customer datacentres in Q2, with the vast majority of revenue landing in 2027. A Blackstone-backed venture will rent TPU compute from 2027 — Alphabet entering the neocloud business
  • SpaceX is paying ~$920M per month for AI compute
  • Cash $242.5B against $98.2B of long-term debt

Bear case

  • Capex is the bear case. $44.9B in the quarter (+100% YoY), FY26 guidance raised to $195–205B from $180–190B, and 2027 guided higher still. The stock fell 7%, losing ~$294B of value, on a beat
  • Free cash flow was −$5.9B in the quarter (TTM $53.3B). Spending is outrunning cloud revenue and the payback period is unknown
  • ~$98–99B of the quarter's reported "earnings" was unrealised mark-to-market on Anthropic and SpaceX. Anyone quoting the "200% EPS beat" is quoting a paper gain
  • Still supply-constrained — CFO Ashkenazi, for multiple consecutive quarters. Alphabet will rent third-party cloud capacity in Q3 as a bridge, accepting margin pressure. Cloud revenue would have been higher with capacity
  • Operating margin faces rising depreciation as the capex lands on the balance sheet
  • Regulatory and antitrust overhang on the Services business that funds all of this

Major customers

  • Advertisers — the overwhelming majority of revenue, across Search and YouTube. Millions of accounts, no meaningful concentration
  • Google Cloud customers — enterprises and, increasingly, AI labs renting TPU and GPU capacity. SpaceX at ~$920M/month is the largest disclosed
  • Device and subscription consumers — Pixel, YouTube Premium, Google One
🔗
Read-through — Alphabet is a demand input, not a category constituent. It is a major customer of
🕸️
AVGO (the TPU is co-designed with Broadcom, so TPU volumes drive a meaningful slice of Broadcom's custom-silicon revenue — and any move to bring TPU design in-house would hit AVGO directly),
🏛️
TSM (which manufactures the TPUs) and
🔆
COHR (the Ironwood architecture is validating Coherent's optical circuit switch category). Its capex guidance is one of the four hyperscaler numbers driving the photonics, networking and power categories, and the $514B cloud backlog is the demand pool every supplier in this file is ultimately drawing from. Note the direction of travel: a $15B capex raise that cost the stock 7% means the market is now pricing the cost of the buildout at the hyperscalers while still rewarding the beneficiaries — that divergence is the thing to watch. Alphabet also stands behind
☁️
CIFR's largest lease via the Fluidstack backstop, and is about to compete with
☁️
CRWV and
☁️
NBIS by renting TPU compute from 2027.

What would change the view

  1. Cloud growth rate and backlog trajectory — $514B is the number to watch quarter to quarter
  2. Capex against free cash flow, and whether FCF returns to positive
  3. Search resilience, currently +17%
  4. Operating margin under rising depreciation from the buildout
  5. Whether the supply constraint eases, and how long the third-party capacity bridge runs
  6. TPU external sales converting in 2027 as guided — the point at which Alphabet becomes a seller in this file, not only a buyer
  7. Any reversal of the Anthropic and SpaceX marks, which would flatter-then-flatten reported earnings in the opposite direction

Update log

19 Sep 2026 — Converted to the standard template. Consolidated the two overlapping Q2 write-ups that were both on the page. Valuation and verdict note: the "Fair" valuation and an earlier "Watch" verdict were recorded as placeholders when the live quote was paywalled; the property now reads Constructive, and the valuation still warrants confirmation.
7 Sep 2026 — Verdict unchanged: Constructive. Framed explicitly as a capex-sentiment read rather than a category constituent.
22 Jul 2026 — Q2 2026 reported. Revenue $119.8B (+24%), operating income $40.8B (+30%) at 34% margin. Google Cloud $24.8B (+82%) with operating income tripling to $8.8B; cloud backlog $514B (+$50B QoQ). Capex $44.9B (+100%); FY26 guidance raised to $195–205B. Free cash flow −$5.9B. GAAP EPS $9.11 inflated by ~$98–99B of unrealised gains on SpaceX and Anthropic. Stock fell 7%.
Research and education only — not investment advice.