ALPHABET Q2
CLOUD_+82%_YoY
CAPEX_$205B_GUIDE.

Google Cloud Q2 2026 revenue $24.8B +82% YoY AI datacenter CapEx TPU GPU

TL;DR для инженеров: Alphabet Q2 2026 — textbook case «growth beat vs cash flow repricing». Revenue $119.8B (+24% YoY, beat), Google Cloud $24.8B (+82% YoY, massive beat), operating margin Cloud 35.6%. Но: FY CapEx guide $195–205B (3rd upward revision), Q2 CapEx $44.9B, FCF −$5.9B (flip from +$5.3B YoY). GOOGL AH −4% → ~$327. Дополнительно: Gemini 3.5 Pro slip (code SLO miss), Frozen v2 ASIC leak, Cloud backlog $514B. Ниже — полный stack trace: P&L, CapEx waterfall, inference throughput metrics, Q3 watchlist, FAQ.

Hard metrics · 30 sec

Total revenue$119.8B · YoY +24% · beat ($116.9B cons.)
Google Cloud$24.8B · YoY +82% · op margin 35.6% · RPO/backlog $514B
After-hoursGOOGL −4% → ~$327 · EPS/revenue beat
CapEx guide FY26$195–205B (Q1 guide $180–190B) · Q2 actual $44.9B
FCF / risk−$5.9B · Gemini 3.5 Pro delay · FY27 CapEx «materially higher»

1. Signal vs noise: beat и −4% AH не противоречат

  1. Market prices NPV, not quarterly beat. CapEx guide trajectory: $175–185B → $180–190B → $195–205B. Street discounting ROI payback period extension before celebrating Cloud acceleration.
  2. GAAP EPS $9.11 — non-recurring noise. ~$99B unrealized gain on equity stakes (Anthropic, SpaceX). Adjusted EPS ~$2.88 inline. Для DCF моделей использовать adjusted, не headline GAAP.
  3. Cloud +82% YoY — infra demand real. Но flagship model gap: GPT-5.6, Claude Fable 5, Grok 4.5 shipped July 2026. Gemini 3.5 Pro — partner eval only, no GA date. Enterprise RFP overhang на inference SLA/code benchmarks.

2. Q2 2026 financials (full table)

MetricQ2 2026ConsensusQ2 2025YoY Δ
Total revenue$119.8B$116.9B$96.4B+24% ✅ beat
Diluted EPS (GAAP)$9.11~$2.88*$2.31+294%
Google Cloud$24.8B$22.4B$13.6B+82% ✅ major beat
Google Search & other$63.3B$63.4B$54.2B+17% ≈ inline
YouTube ads$11.1B$9.8B+13%
Total ads$81.6B
Operating income$40.8B$31.3B+30%
Cloud op. margin35.6%20.7%+1490 bps
Q2 CapEx$44.9B$22.5B+100%
Free cash flow−$5.9B+$5.3B⚠️ negative flip
Cloud backlog (RPO)$514B~$464BQoQ +$50B

*Adjusted EPS ~$2.88 excludes ~$99B unrealized equity gains (Anthropic, SpaceX stakes).

Primary sources: Alphabet Q2 2026 Earnings Release (PDF) · SEC Exhibit 99.1

3. Google Cloud: +82% YoY acceleration

Strongest segment metric: revenue $24.8B, YoY +82% — acceleration from Q1 +63%, Q4'25 +48%. Cloud operating income $8.8B (prior year $2.8B), margin 35.6%. Backlog $514B, QoQ +$50B — >5 years revenue visibility at current quarterly run-rate.

Demand drivers (infra stack):

  1. Enterprise AI compute (TPU/GPU cluster rental, H100/B200 adjacency via GCP)
  2. Gemini Enterprise — ~90% Fortune 100 deployed
  3. Core GCP workload growth
  4. First disclosed TPU system hardware revenue (sell-through, not just internal)
  5. Wiz acquisition — cloud security attach
CEO Sundar Pichai (earnings call transcript): "Q2 was a remarkable quarter — Alphabet revenue grew 24% YoY, Google Cloud accelerated to 82% growth, driven by AI infrastructure and AI solutions demand."

4. Search & YouTube: AI Overviews не cannibalize ads (Q2 data)

  • Google Search & other: $63.3B (+17% YoY)
  • YouTube ads: $11.1B (+13% YoY)
  • Total ads: $81.6B

AI product telemetry (disclosed):

  • AI Mode global MAU >1B (post Oct 2025 global rollout)
  • Gemini App MAU 950M
  • Gemini API throughput 22B tokens/min (Q1: 16B tokens/min — +37.5% QoQ)
  • AI search features drive billions of outbound clicks/week to publishers
  • Antigravity coding agent WAU 2.4M

Query volume still growing. AI Mode/AIO = search experience extension, not replacement — at least Q2 cohort. Pichai directly addressed zero-click publisher FUD with click-through data.

5. Beat → −4% AH: three repricing drivers

5.1 CapEx guide — third upward revision

TimestampFY2026 CapEx guide
Jan 2026$175–185B
Q1 earnings (Apr)$180–190B
Q2 earnings (Jul)$195–205B

Q2 CapEx actual $44.9B (+100% YoY). Split: ~60% servers/compute, ~40% datacenter + network. Management: FY2027 CapEx «materially higher» than FY2026.

5.2 FCF negative flip

Q2 FCF −$5.9B vs +$5.3B prior year. Record revenue, but burn rate > OCF generation — classic hyperscaler AI infra cycle.

5.3 Gemini 3.5 Pro slip

Next-gen flagship not GA while competitors shipped July. CFO Anat Ashkenazi: demand signals stronger YoY, investments under strict ROI framework. Market thesis: growth realized, invoice arrived.

6. Gemini 3.5 Pro: technical delay post-mortem

Gemini 3.5 Pro — target GA June 2026, currently delayed.

  • Bloomberg (Jul 16): code generation / agent benchmarks below internal SLO, slip measured in months
  • Late June training data refresh for code capability — insufficient lift
  • Model exists, in partner eval — Google confirmed 3.5 Pro + upgraded Flash in testing
  • No GA date disclosed
  • Jul 16 rumor: ~$200B market cap evaporation (−4.4% single session)
  • Raymond James Josh Beck: delay shifted perception «leading edge → trailing edge»

Competitive context (Jul 2026): GPT-5.6, Claude Fable 5, Grok 4.5 GA. Code/agent evals = current frontier battleground. Antigravity 2.4M WAU shows product traction, but base model gap matters for enterprise inference contracts.

Google response:

  • 3.5 Pro + upgraded Flash in partner eval
  • Gemini 4 R&D initiated
  • Near-monthly model release cadence planned
  • Gemini 3.5 Flash Cyber shipped (security-focused, Pichai: «frontier-class cost-efficiency»)

7. Frozen v2 ASIC: silicon-level Gemini bet

While waiting on 3.5 Pro GA, Google betting on hardware efficiency path. The Information leak: Frozen v2 custom inference ASIC in development:

SpecDetail
ArchitecturePartial Gemini transformer graph hardwired — not general-purpose TPU
EfficiencyEst. 6–10× tokens/watt vs current TPU generation
ScopeGemini inference only (not training, not generic LLM)
PositioningComplements TPU fleet; low initial volume
P&L impact↓ inference $/token → Cloud margin expansion (2027+)

Power efficiency = $/token at hyperscale. If Frozen v2 hits spec, Gemini inference COGS potentially below competitors renting Nvidia H100/B200 clusters. TPU system sell-through revenue recognition weighted to 2027, not 2026.

8. Capital deployment stack

Line itemEvidence
AI datacenters40% CapEx → facilities + network
AI compute hardware60% CapEx → servers/TPU/accelerators
Equity raise (Jun 2026)$49.6B common + preferred issuance
Debt (Q2)$20.3B senior unsecured notes
Cash + securities$242.5B (post-raise)
Third-party compute bridgeQ3 external GPU/TPU capacity — near-term Cloud margin headwind

Not Google-specific — all hyperscalers in AI infra arms race. Core question: can $200B+/yr CapEx generate equivalent incremental AI revenue, and on what timeline.

9. QoQ growth comparison

SegmentQ1'26 YoYQ2'26 YoYTrend
Total revenue+22%+24%⬆️ accelerating
Google Cloud+63%+82%⬆️ major accel
Google Search+19%+17%⬇️ slight decel
YouTube ads+11%+13%⬆️ recovery
Operating margin36%34%⬇️ investment drag

10. Q3 2026 watchlist (infra + model)

  1. Gemini 3.5 Pro GA date — any announcement = vol event
  2. Search revenue comp — CFO flagged tougher YoY base Q3
  3. Cloud margin — third-party compute bridge impact
  4. FY2027 CapEx guide — «materially higher» already telegraphed
  5. Competitive releases — Kimi K3, GPT-5.6, Claude Fable 5 (Jul GA/OSS)
  6. Frozen v2 production timeline — any tape-out/ramp signal
  7. TPU external sell-through revenue — 2027 acceleration

11. Five actionable items (dev + infra leads)

  1. Recalc AI infra ROI: $514B backlog = demand-side validation, but CapEx 3× guide raise → cost pass-through to enterprise pricing. Re-eval TCO: Gemini Enterprise vs OSS/API (DeepSeek V4 GA, Kimi K3).
  2. Monitor 3.5 Pro GA window: if Q3 still no date, enterprise contracts need multi-model fallback (OpenRouter / LiteLLM routing layer).
  3. Antigravity + API first: 2.4M WAU = coding agent traction. During flagship delay, validate workflows on Gemini API + Antigravity; don't single-model bet frontier.
  4. Track Frozen v2 / TPU sell-through: 2027 TPU hardware revenue may shift «rent Nvidia vs rent GCP TPU» cost equation — run POC early.
  5. Strip one-time items: GAAP EPS $9.11 includes $99B unrealized gains — valuation models use adjusted ~$2.88.

12. Deep dive: 2026 AI commercialization ROI inflection

Alphabet Q2 2026 = microcosm of 2026 AI monetization inflection. Bull case quantified: Cloud +82%, Search +17%, Gemini ~1B MAU scale, 90% Fortune 100 on Gemini Enterprise — full-stack AI (model + ASIC + cloud + consumer) producing measurable financial returns.

Bear case equally hard: $200B CapEx run-rate, FCF negative, flagship slip, FY27 CapEx «materially higher». Market pricing invoice before growth — closes loop with 2026 AI funding supercycle $830B compute capex narrative. Growth story done; now ROI story must land.

For Mac/MLX devs: Gemini API 22B tokens/min throughput, Antigravity 2.4M WAU = Google retains scale advantage in coding/agent API layer. But 3.5 Pro slip makes local MLX inference + remote API hybrid the pragmatic architecture. When hyperscalers pass CapEx invoice to cloud pricing, Apple Silicon on-device + on-demand API split wins on TCO.

Sources: Alphabet Q2 2026 Earnings Release · CNBC Live · Bloomberg (Gemini 3.5 Pro) · The Information (Frozen v2)

13. Summary

Bull: AI monetization landing — Cloud +82%, Search growing, Gemini ~1B scale.

Bear: Frontier maintenance cost exponential. $200B CapEx, FCF flip, flagship delay — market prices invoice first.

For AI infra observers: growth story closed; ROI story is the open ticket.

14. FAQ

Q: Alphabet Q2 2026 core metrics?
A: Revenue $119.8B (+24%), Cloud $24.8B (+82%), EPS $9.11 GAAP (adj. ~$2.88).

Q: Beat but stock down?
A: CapEx guide $195–205B, FCF −$5.9B, Gemini 3.5 Pro slip — growth beat, future CF pressure.

Q: Gemini 3.5 Pro GA when?
A: No public date; partner eval. Bloomberg: code SLO miss, months slip.

Q: Frozen v2?
A: Gemini-specific inference ASIC, arch embedded in silicon, est. 6–10× TPU efficiency, 2027 impact.

Q: Cloud backlog $514B meaning?
A: >5yr run-rate visibility — demand-side validates CapEx narrative.

Q: FY2027 CapEx?
A: Management: «materially higher» than 2026 — price longer ROI cycle.

15. Closing: Cloud API anywhere, Agent stress-tests on isolated Mac node

Read earnings, tweak Gemini API routing — works on Windows/Linux. But for Antigravity + Gemini API long-session coding, multi-repo agent stress-tests, MLX + Cloud API hybrid validation: Apple Silicon unified memory + Metal toolchain = lowest dev friction.

Pragmatic split: primary machine → Gemini/multi-model API; agent load, long-context batch, quantization validation → MACGPU remote Mac mini M4 node — on-demand, SSH-isolated. In $200B CapEx era, local+remote hybrid beats pure cloud dependency on controllability.