Phase A — Understand the business
Lens 1 · Company Overview
Google DeepMind is Alphabet's unified AI research and model-development organization, formed by the April 2023 merger of Google Brain and DeepMind into one lab under CEO Demis Hassabis ``. It does not sell to customers directly. Its product is frontier AI capability — the Gemini model family, the research that feeds it (and spinouts like Isomorphic Labs), and the software layer that makes Alphabet's custom TPU silicon useful. That capability is monetized through three parent channels:
- Google Services ($342.7B FY25 revenue, +12% YoY) — Gemini embedded in Search (AI Overviews, AI Mode), YouTube, Android, Workspace, and the consumer Gemini app. All 15 of Alphabet's half-billion-user products (7 of them at 2B+ users) now run Gemini models ``.
- Google Cloud ($58.7B FY25 revenue, +36% YoY) — Gemini sold to enterprises via Vertex AI (200+ foundation models), Gemini Enterprise, and Gemini for Workspace, plus raw AI-infra (TPU/GPU capacity) ``.
- Other Bets / moonshots — Isomorphic Labs (AI drug discovery, the AlphaFold lineage), Waymo, quantum ``.
Business model in plain terms: Alphabet is still, at its core, an advertising machine — Google advertising was $294.7B of $402.8B total FY25 revenue (73%) . DeepMind's job is to (a) defend that ad franchise against AI-native search substitutes by making Google's own surfaces AI-native first, and (b) build the second growth engine (Cloud) into a credible #3 hyperscaler. **Contract structure:** ads are auction/CPC-based (no recurring lock-in at the user level, high switching ease for advertisers); Cloud is consumption-based + subscriptions, increasingly **backlogged** ($462B remaining performance obligations — see Lens 5) .
Customers / suppliers / competitors: Customers = billions of consumers (free, ad-monetized) + millions of advertisers + enterprise Cloud buyers (incl. Anthropic, now a TPU customer — see Lens 2). Suppliers = Broadcom (TPU co-design), TSMC (fab), Nvidia (GPUs it still buys + resells), energy/datacenter providers. Competitors = OpenAI/Microsoft, Anthropic, Meta (Llama), xAI on models; AWS + Azure on cloud; Perplexity + ChatGPT on the search-substitute frontier.
Lens 2 · Supply Chain
DeepMind's "supply chain" is the AI compute stack — uniquely, Alphabet is vertically integrated across most of it, which is the core structural advantage.
Upstream → DeepMind → end customer:
- Silicon design: Google designs its own TPUs in-house (Ironwood = 7th-gen, GA March 2026)
. Co-design + manufacturing partner is **Broadcom** — a multi-year deal estimated ~**$21B through 2029** covering TPU 8t/8i . This is the single most important chokepoint relationship: it lets Alphabet train Gemini and serve Cloud AI without paying the Nvidia tax on the marginal unit.
- Fabrication: TSMC (advanced nodes) — shared single-source chokepoint with every AI peer.
- GPUs (parallel path): Alphabet still buys Nvidia GPUs and offers them in Cloud alongside TPUs `` — a hedge, not a dependency.
- Datacenters + power: capex $91.4B FY25, $35.7B in Q1'26 alone; data-center leases not yet commenced ballooned to $75.6B (from $58.5B at YE25); a $9.9B power purchase agreement signed Jan 2026 ``. Power is the emerging chokepoint, as it is for the whole sector.
- DeepMind (the lab): consumes the compute → produces Gemini + research → ships into products.
- End customers / buyers: consumers via Search/YouTube/Gemini app; advertisers; enterprises via Cloud; and notably Anthropic, whose Claude models are trained predominantly on Google TPU pods under a Google commitment of up to $40B ``. Anthropic is the only frontier lab whose primary training stack is not Nvidia GPUs — a validation of the TPU as a merchant product, and a self-funding loop (Google invests in Anthropic; Anthropic spends it on Google TPUs/Cloud).
Chokepoints / single-source: TSMC (industry-wide), Broadcom (TPU manufacturing — concentrated but contractual), power/grid interconnects (the new binding constraint — Pichai said Cloud is "compute constrained in the near term," i.e. revenue is supply-capped not demand-capped) ``.
Lens 3 · Competitive Advantages (moats)
The moat is the full stack, owned end-to-end — Alphabet is the only player that controls silicon (TPU) + model (Gemini) + distribution (15 products with 500M+ users each) + a hyperscale cloud to rent it all out. Specifically:
- Distribution moat (the deepest). Gemini doesn't need to win a standalone app war — it ships into Search, Gmail, Android, Chrome, Workspace, YouTube by default. AI Overviews reached all-time-high search query volumes in Q1'26 while monetizing "at a rate similar to traditional Search" ``. That is the moat working: the AI transition is happening inside Google's owned surfaces, not around them.
- TPU cost moat. Training/serving on owned silicon structurally lowers the cost-per-token vs. competitors renting Nvidia. The Anthropic TPU win proves it's good enough to be a merchant product, not just captive ``. This compounds: cheaper inference → can give Gemini away in more free products → more data → better models.
- Data + scale. Search, YouTube, Maps, Android telemetry — a proprietary multimodal data estate no lab can replicate.
- Talent / IP. DeepMind under Hassabis is a top-2 research lab globally (AlphaFold Nobel-adjacent science, Gemini 3 first model to cross 1500 Elo on LMArena) ``.
Bargaining power. Over advertisers: very high (must-have demand channel). Over Cloud customers: rising (a $462B backlog = customers pre-committing years of spend). Over suppliers: high vs. Nvidia (TPU gives a credible alternative — Alphabet is one of the few buyers that can walk); moderate vs. TSMC/Broadcom (shared with the field). Where the moat is thinnest: the frontier-model lead is contested, not owned — Claude Opus 4.6 retook the LMArena top spot from Gemini 3.1 in early 2026 ``. Capability parity is a rotating crown; the durable edge is distribution + cost, not "best model this month."
Lens 4 · Segments
Alphabet reports three segments. The AI franchise shows up most cleanly as Cloud acceleration + margin expansion and the Alphabet-level (shared AI R&D) cost line.
FY2025 (year ended Dec 31, 2025) — revenue by type ``:
| Line | FY2024 ($M) | FY2025 ($M) | YoY |
|---|
| Google Search & other | 198,084 | 224,532 | +13% |
| YouTube ads | 36,147 | 40,367 | +12% |
| Google Network | 30,359 | 29,792 | −2% |
| Google subs, platforms, devices | 40,340 | 48,030 | +19% |
| Google Services total | 304,930 | 342,721 | +12% |
| Google Cloud | 43,229 | 58,705 | +36% |
| Other Bets | 1,648 | 1,537 | −7% |
| Total revenues | 350,018 | 402,836 | +15% |
Segment operating income ``:
| Segment | FY2024 OI ($M) | FY2025 OI ($M) | FY25 margin |
|---|
| Google Services | 121,263 | 139,404 | 40.7% |
| Google Cloud | 6,112 | 13,910 | 23.7% (vs 14.1% FY24) |
| Other Bets | (4,444) | (7,515) | n/m |
| Alphabet-level (shared AI R&D) | (10,541) | (16,760) | — |
| Total operating income | 112,390 | 129,039 | 32.0% |
Q1 2026 (three months ended Mar 31, 2026) ``:
| Segment | Q1'25 rev ($M) | Q1'26 rev ($M) | Q1'25 OI ($M) | Q1'26 OI ($M) | Q1'26 margin |
|---|
| Google Services | 77,264 | 89,637 | 32,682 | 40,589 | 45.3% |
| Google Cloud | 12,260 | 20,028 (+63%) | 2,177 | 6,598 | 32.9% |
| Other Bets | 450 | 411 | (1,226) | (2,100) | n/m |
| Alphabet-level (shared AI R&D) | — | — | (3,027) | (5,391) | — |
| Total | 90,234 | 109,896 | 30,606 | 39,696 | 36.1% |
The trend that matters: Google Cloud is accelerating (FY25 +36% → Q1'26 +63%) and expanding margin violently (14.1% → 23.7% → 32.9% in five quarters). That is operating leverage on AI infra demand. Meanwhile the Alphabet-level AI R&D cost line is growing fast (−$10.5B → −$16.8B FY25; −$3.0B → −$5.4B Q1'26, annualizing toward ~$22B) — this is DeepMind's cost, and the bet is that Cloud + Search monetization more than pays for it. Geography: US 48%, EMEA 29%, APAC 17%, Other Americas 6% (stable) . The one soft spot is **Google Network −2% FY25 / −4% Q1'26** — third-party publisher disintermediation as AI Overviews keeps clicks on Google's own surfaces .
Phase B — Measure performance
Lens 5 · Earnings Result (Q1 2026, reported Apr 29, 2026)
The print ``:
- Revenue $109.9B, +21.8% YoY (vs ~$107.2B consensus → ~2.5% beat) ``.
- Operating income $39.7B, op margin 36.1% (up from 33.9% Q1'25) — the clean read on the business.
- Net income $62.6B; diluted EPS $5.11 vs ~$2.62 consensus ``. Caveat (forensic): the headline EPS beat is flattered by $37.7B of OI&E, almost entirely unrealized gains on non-marketable equity securities (private-company mark-ups). Strip that and the operating story is strong but not a 95% beat. Use operating income, not net income, to judge the franchise.
- R&D $17.0B (+26% YoY) — the AI investment ramp.
What drove it: Google Cloud +63% to $20.0B with operating income tripling to $6.6B was the headline. Search & other +19% to $60.4B — acceleration, the single most important data point against the "AI kills search" bear case . Consumer Gemini app had its strongest-ever quarter for paid AI plans .
Guidance / capital intensity: Management raised 2026 capex guidance to ~$180–190B (one source $175–185B), with 2027 expected "significantly" higher . FY25 capex was $91.4B; Q1'26 was $35.7B . Balance-sheet flags: cash + STM $126.8B at YE25; operating cash flow $164.7B FY25; $48.5B senior notes; $15.6B short-term accrued legal/regulatory (mostly EC fines) ``. Receivables/inventory not flashing relative to revenue.
Market reaction: stock jumped ~10% to ~$382 post-print `` — the market re-rated on Cloud acceleration + the AI-monetization proof. (Note the price has since drifted; see Lens 7/8.)
Lens 6 · Earnings Calls (sentiment trend)
No transcripts on disk (transcripts/ empty) — sourced ``. The tonal arc across the last several calls:
- Mid-2025: defensive — answering "are you still relevant / will AI eat Search?" The framing was protect the core.
- Q4 2025 → Q1 2026: the tone flipped to offense and constraint. The recurring phrases are now "full-stack," "compute constrained," "backlog," and "AI Overviews monetizing at parity." Pichai explicitly said "our cloud revenue would have been higher if we were able to meet the demand" `` — a demand-not-the-problem signal that management leans into.
- What they stopped saying: the apologetic "we're being thoughtful about the AI transition in Search" hedge of 2024-25. Now it's "queries at an all-time high."
- What they added: AGI ambition is now on-record from the lab side — Hassabis (May 2026) saying we're "close to AGI" and forecasting a "golden era of discovery within 15 years" ``. Read as long-dated optionality, not a guidable line.
Net: management sentiment has shifted from defending the franchise to rationing supply of a thing everyone wants. That is the single biggest narrative change in the name.
Lens 7 · Comps
Mega-cap AI/compute peers. Multiples are `` with date; where I cannot source cleanly I mark n/a. Do not treat these as precise — they're snapshot reads from screeners around 2026-06-12 to 06-17.
| Company | Ticker | Fwd P/E | EV/EBITDA | Notes |
|---|
| Alphabet | GOOGL | ~26 (some screeners ~29.6) | ~27.8 | Cheapest mega-cap on fwd P/E `` |
| Microsoft | MSFT | ~21.3 | ~16.1 | Lowest EV/EBITDA of the group `` |
| Nvidia | NVDA | ~33.1 | ~29.5 | Most expensive on fwd P/E `` |
| Meta | META | n/a | n/a | — |
| Amazon | AMZN | n/a | n/a | — |
| Anthropic (private) | — | n/a — private | n/a — private | TPU customer / $40B Google commitment `` |
| OpenAI (private) | — | n/a — private | n/a — private | Azure-aligned rival `` |
Read: Alphabet trades at a discount to Nvidia and roughly in line with / slightly above Microsoft despite owning the only vertically-integrated TPU-to-product stack and the fastest-accelerating cloud. The discount is explicitly the antitrust + AI-disruption overhang . Dividend yield is small (raised quarterly dividend to $0.21/share, +5% in Apr 2025) . 5-yr avg ROE: n/a (would need to compute from history; FY25 net income $132.2B on a very large equity base implies high-20s%/low-30s% ROE, but I will not fabricate the 5-yr average).
Lens 8 · Stock-Price Catalysts (last ~5 years, what moves GOOGL >5%)
Pattern, all ``:
- Antitrust rulings = binary, large moves. Aug 2024 DOJ Search liability ruling (against Google); Dec 2025 final Search remedy (behavioral, no Chrome/Android divestiture — relief rally because the feared structural breakup didn't land) ``. The Apr 2025 ad-tech "mixed" ruling. The market reacts violently to the gap between feared and actual remedy.
- AI-disruption narrative = the dominant swing factor 2024-25. Every "ChatGPT is eating Search" data point pressured the stock; every "AI Overviews monetizes fine / queries at ATH" data point relieved it. The Q1'26 print (+~10%) was the clearest "the bear thesis is wrong" repricing ``.
- Cloud growth rate is now a primary print-day mover (the +63% / $462B backlog was the Q1'26 story).
- Capex guidance cuts both ways: the $180–190B 2026 guide is read as either "demand is real" (bull) or "FCF is getting torched" (bear) — consensus 2026 FCF estimates collapsed ~70% YoY on the capex ramp ``.
What the market actually reacts to for this name: (1) antitrust headlines, (2) proof-points on AI monetization of Search, (3) Cloud growth/margin, (4) the capex/FCF trade-off. Macro and rates matter less than for unprofitable AI names — this is a cash machine.
Phase C — Judge people & books
Lens 9 · Management
- Sundar Pichai (CEO, Alphabet & Google) — CODM for segment reporting ``. Long tenure (CEO since 2015/2019); steered the company through the AI transition from a defensive 2023 start to the offensive 2026 posture. Track record: turned Cloud from a money-loser (−$ billions) into a $13.9B-OI / 23.7%-margin segment and orchestrated the Brain+DeepMind merger.
- Demis Hassabis (CEO, Google DeepMind) — the scientific founder archetype; co-founded DeepMind (acquired by Google 2014), Nobel-adjacent science (AlphaFold), now runs the merged lab and chairs Isomorphic Labs ``. Skin in the game is scientific reputation + Alphabet equity. This is a rare founder-scientist running a captive lab at scale — implies durable research edge but also key-person risk.
- Capital-allocation history: reinvest-heavy and shareholder-friendly simultaneously — FY25: $45.4B buybacks (240M shares retired), $10B+ dividends, new $70B repurchase authorization (Apr 2025, $69.5B remaining), while funding $91.4B capex
. M&A: **$32B Wiz** (cloud security) + **$4.8B Intersect** (datacenter/energy) agreed 2025, both closing 2026 subject to regulatory approval; **$16B Waymo round** (mostly Alphabet-funded, Feb 2026) . This is aggressive-but-disciplined: buying the security gap (Wiz) and the power/datacenter gap (Intersect) — both bottlenecks identified in Lens 2.
- Red flags: (1) Waymo valuation-based SBC charge — a $2.1B comp charge in Q4'25 tied to estimated stock valuation, inside the Other Bets loss
; non-cash but a reminder of Other Bets' opacity. (2) **Dual-class control** — Page + Brin beneficially own ~89.3% of Class B (10 votes/share); public Class C (GOOG) has *no* votes . Governance is founder-controlled; minority holders cannot influence.
- Archetype: professional manager (Pichai) executing on top of founder-scientist research (Hassabis) under founder financial control (Page/Brin). For this stage — defending a cash cow while building the next platform — that's a strong configuration.
Lens 10 · Forensic Red Flags
Accounting risks (every figure `` unless noted):
- OI&E / equity-securities gains = the biggest "quality of earnings" flag. FY25 OI&E was $29.8B, of which $24.1B was net gains on equity securities (mostly unrealized fair-value adjustments on non-marketable — i.e. private, level-3, mark-to-model — holdings). Q1'26 OI&E was $37.7B. These flatter net income and EPS but are non-cash, non-operating, and reversible. The Q1'26 "95% EPS beat" is largely this. Judge the franchise on operating income, not net income.
- Capex → depreciation lag. Capex exploded ($52.5B → $91.4B FY25 → ~$180–190B guided 2026) but depreciation lags ($21.1B FY25). As "assets not yet in service" (multi-year datacenter builds) come online, depreciation will surge and compress margins — a future earnings headwind hidden in today's numbers. Watch the gap.
- Off-balance-sheet / commitment build-up: data-center leases not yet commenced = $75.6B at Q1'26 (up from $58.5B); a $9.9B PPA; $7.7B content-licensing commitments; $5.7B backstop guarantees for power equipment ``. Real future cash obligations not yet on the balance sheet.
- SBC: R&D comp rose materially incl. a $4.2B SBC increase FY25 (some Waymo valuation-based). Non-GAAP isn't reported by Alphabet the way it is for smaller names, so SBC flattering is less of an issue here than usual.
- Legal accruals: $15.6B short-term accrued legal/regulatory; G&A rose $7.3B FY25 primarily on the $3.5B EC ad-tech fine + a $1.4B legal accrual ``. These are real and recurring for this company.
- Cash vs earnings: operating cash flow $164.7B vs net income $132.2B — cash flow exceeds earnings (depreciation + SBC add-backs), the healthy direction. No receivables/inventory blowout.
Regulatory findings (read from regulatory/regulatory-findings.md + 10-K Note 10):
- SEC: No Litigation Releases or AAERs naming the company, 2021-2026 ``. Clean on accounting enforcement.
- Antitrust (10-K Note 10, "Antitrust Matters" — the material legal exposure, all ``):
- US Search (DOJ + states): Aug 2024 liability ruling against Google; final judgment Dec 2025 imposing remedies — restrictions on how Google distributes its services + requirement to share certain search data with, and offer syndication services to, competitors. Google appealed + moved to pause remedies (Jan 2026); DOJ + state AGs also appealed (Feb 2026, seeking structural breakup) ``. No Chrome/Android divestiture ordered — the feared structural remedy did not land.
- US Ad Tech (DOJ + states, E.D. Va.): Apr 2025 mixed ruling — advertiser tools + DoubleClick/AdMeld acquisitions cleared, but publisher tools "unfairly excluded rivals." DOJ seeking structural remedies "that could have a material adverse effect"; closing args Nov 2025; final judgment awaited. Texas AG case to follow.
- EC fines: Shopping €2.4B (paid $3.0B in 2024); Android €4.3B→€4.1B (ECJ appeal pending); AdSense €1.5B (annulled by General Court 2024, EC appealing); Ad-tech €3.0B (Sept 2025) for self-preferencing — Google took a $3.5B charge Q3'25, posted bank guarantees, appealed Nov 2025.
- Epic / Google Play: Dec 2023 jury verdict against Google; remedies implemented Oct 2025; Supreme Court appeal pending; Oct 2025 Epic settlement to modify the injunction (court approval pending).
- EU DMA: preliminary non-compliance findings (Play + Search), March 2025.
- Privacy: $1.4B settlement finalized Oct 2025; ACCC + JFTC Search-distribution settlements/orders (JFTC no monetary penalty).
Forensic verdict: Accounting is clean (no SEC actions, cash > earnings). The two real flags are (1) earnings quality inflated by unrealized private-equity marks and (2) the largest antitrust exposure of any company on earth — material, multi-front, and with structural remedies live in the ad-tech case and on appeal in Search.
Phase D — Project & stress-test
Lens 11 · Forward Projection
**Approach: bottom-up from FY25 actuals; output .** I deliberately do **not** anchor on screener "consensus EPS" — the figures sourced were inconsistent ($14.22 for 2026 vs $11.51 for 2027 implies 2026 includes one-time equity gains; not a clean GAAP base) . Goldman's revenue path is the cleaner external anchor: $407B (2026) → $472B (2027) → $541B (2028) ``.
FY25 actuals base ``: revenue $402.8B, operating income $129.0B (32.0% margin), diluted EPS $10.81 (GAAP, includes equity gains).
Because GAAP EPS is distorted by equity-securities marks, I project operating income (the franchise) and treat below-the-line as noise.
| Driver | Base | Bull | Bear |
|---|
| Revenue CAGR FY25→FY28 | +14%/yr → ~$540B FY28 `` | +17%/yr (Cloud >50%, Search holds) → ~$600B | +9%/yr (Search ad pressure, macro) → ~$480B |
| Operating margin FY28 | ~32% (capex/depreciation drag offsets Cloud leverage) | ~35% (Cloud margin >35%, AI Overviews accretive) | ~27% (depreciation surge + remedy costs + ad mix) |
| FY28 operating income `` | ~$173B | ~$210B | ~$130B |
- Bull math: Cloud compounding 40%+ at expanding margin + Search holding on AI-Overviews parity + operating leverage. Op income ~$210B by FY28.
- Base math: revenue ≈ Goldman ($540B FY28), margin flat at 32% as the depreciation wave from $180B+ capex eats the Cloud leverage → ~$173B op income (
= $540B × 0.32).
- Bear math: Search ad growth decelerates as AI Mode cannibalizes ad slots + a structural ad-tech remedy forces a divestiture + depreciation surge → margin to 27% on ~$480B → ~$130B op income (roughly flat vs FY25).
The swing variable is not the model — it's the antitrust remedy + the capex/depreciation curve. Per the --watchlist rule, no Brier forecast logged in breadth mode.
Lens 12 · Bull vs Bear
Bull case. Alphabet is the only company that owns the entire AI value chain — TPU silicon, the frontier model (Gemini), the world's best distribution (15 products × 500M+ users), and the fastest-growing hyperscale cloud — and it just proved the AI transition is accretive, not dilutive: Search +19% with AI Overviews monetizing at parity, Cloud +63% with a $462B backlog (≈6 years of revenue, >50% converting within 24 months) ``. The TPU is now a merchant weapon (Anthropic trains Claude on it). And you buy all of this at ~26× forward earnings — the cheapest mega-cap — because the market is pricing antitrust fear that, in the base case, resolves as behavioral remedies, not a breakup. The AGI/Isomorphic optionality is a free call.
Bear case (permanent-impairment risks).
- Structural antitrust remedy. The ad-tech case (publisher tools ruled exclusionary) has the DOJ seeking structural remedies; the Search case is on appeal with DOJ + 35 states pushing for a breakup. A forced divestiture of the ad exchange or a real distribution remedy could permanently impair the highest-margin part of the business.
- Search disintermediation, the slow version. Even if AI Overviews monetizes at parity today, a world where users get one AI answer instead of a page of ads structurally shrinks ad inventory over years. Google Network is already −4% — the canary.
- Capex without returns. $180–190B 2026 capex (and "significantly higher" 2027) on a depreciation lag — if AI demand growth slows before the assets are monetized, margins compress hard and FCF (already guided down ~70%) stays depressed.
Pre-mortem (18 months out, thesis broke): The D.C. Circuit ordered a structural Search remedy and the E.D. Va. ad-tech judgment forced an ad-exchange divestiture; simultaneously, capex hit $200B+ while Cloud growth decelerated from 60%+ to 30s as the compute shortage eased and pricing normalized — so margins compressed just as depreciation surged. The stock de-rated from 26× to ~16× on "structurally impaired ad monopoly + capex black hole."
Are multiples too high? No — 26× forward for mid-teens revenue growth, a cash machine, and embedded AI optionality is not demanding. The multiple already embeds the antitrust discount.
Contrarian view (what the market refuses to see): The market is still trading GOOGL as "the AI loser whose Search monopoly is under attack." The Q1'26 print said the opposite — Alphabet is winning the AI transition on its own surfaces and renting the picks-and-shovels (TPU) to its rivals. The antitrust overhang is the reason it's cheap, and the most-feared outcome (Chrome/Android breakup) already didn't happen in the Dec 2025 Search judgment.
Lens 13 · Devil's Advocate (short-seller)
Dismantling the bull case:
- Revenue concentration is brutal. ~73% of revenue is Google advertising; the entire company's value rests on the durability of the search-ad model ``. Everything else (Cloud, DeepMind, Other Bets) is a rounding error on profit ($13.9B Cloud OI vs $139.4B Services OI). If AI permanently compresses search-ad economics, nothing else is big enough to matter.
- The moat the bulls love is exactly what the DOJ is dismantling. "Default distribution" — Gemini-in-everything, Search-as-default — is the precise conduct two courts have now ruled illegal. The distribution moat and the legal liability are the same asset.
- Most dangerous competitor bulls underestimate: Microsoft + OpenAI in the enterprise. Azure + Copilot + OpenAI is bundling AI into the enterprise software stack Google doesn't own (no Office, no Windows, weaker enterprise sales motion). Cloud's +63% is impressive but off a base where it's still #3.
- Worst capital-allocation tells: $32B for Wiz at a rich multiple; $16B into Waymo (still loss-making, with opaque valuation-based comp charges); $180B+ capex with depreciation deferred — the bill comes later.
- Assumptions that must hold for today's price: (a) AI Overviews keeps monetizing at parity as usage shifts, (b) antitrust resolves behaviorally not structurally, (c) Cloud growth stays >30% long enough to absorb depreciation, (d) the TPU edge persists vs Nvidia's roadmap.
- If growth disappoints 20–30%: revenue to ~$470B FY28 instead of $540B, margin to 27% on depreciation drag → op income ~$130B (flat for 3 years) → the multiple compresses to mid-teens → 30–40% downside.
- Single scenario that permanently impairs: a forced structural separation of the ad exchange (sell-side) from Search/buy-side, breaking the closed-loop ad-tech economics that drive the 40%+ Services margin. Plausibility: moderate and rising — it's a live remedy ask in the E.D. Va. case, not hypothetical.
Lens 14 · Management Questions (ordered by information value)
- In the E.D. Va. ad-tech case, if the court orders divestiture of the publisher ad server or exchange, what is the revenue and operating-margin impact, and what's your contingency?
- AI Overviews "monetizes at parity" today — but as queries shift from 10 blue links to single AI answers, what happens to ad inventory volume (impressions), not just rate, over 3 years?
- 2026 capex is $180–190B with 2027 "significantly higher." At what point does depreciation peak as a % of revenue, and what's the trough operating margin during that wave?
- Cloud backlog is $462B but you're "compute constrained." What's the realistic conversion curve, and how much is TPU hardware resale (lower margin) vs. higher-margin platform/AI services?
- How much of the $40B Anthropic commitment flows back as TPU/Cloud revenue, and how do you account for the circularity (you fund a customer who buys your compute)?
- What is the durable cost-per-token advantage of Ironwood/TPU 8 vs. Nvidia's current generation, quantified, and how does the Broadcom dependency affect it?
- Other Bets lost $7.5B in FY25 with opaque Waymo valuation-based comp charges. What's the path to Waymo profitability, and will you ever break it out?
- With Page/Brin controlling ~89% of votes via Class B, how do you ensure minority (Class C, no-vote) holders' interests are represented in a forced-restructuring scenario?
- OI&E added $24B of unrealized equity gains in FY25. How should investors think about the recurring vs. one-time split of below-the-line income?
- Where is Gemini's frontier-capability lead most durable, given Claude retook the LMArena top spot — is the bet on "best model" or on "good-enough model + best distribution + lowest cost"?
- Isomorphic Labs put its first AI-designed drug into FDA clearance (ISM8969) — at what point does drug discovery become a reportable, material business vs. a moonshot?
- If the D.C. Circuit orders structural Search remedies on appeal (decision expected late 2026), what's the operational and financial plan?
- Wiz ($32B) and Intersect ($4.8B) close in 2026 — what's the integration thesis, and how do they change the Cloud margin trajectory?
- How do you think about returning the ~$127B cash + $164B annual operating cash flow to shareholders while funding a $180B+ capex year — what's the buyback/dividend floor?
- What single external development would most change your AI infrastructure investment plan?