Alphabet Inc. — Bull & Bear Memo

Bull & bear thinking anchor — not a thesis, valuation, or recommendation

1. Business in one line

Alphabet sells access to commercial intent: an auction prices advertiser access to a search query, a video view or an AdMob impression at close to zero incremental cost per unit sold — and Alphabet now spends that cash flow to build, run and sell AI compute. In FY2025 Google Services produced $342.7B of revenue and $139.4B of operating income, a 40.7% margin; Cloud $58.7B and $13.9B; Other Bets $1.5B against a $7.5B loss; and $16.8B of shared AI R&D sat in an unallocated line no segment margin reflects [FY2025 10-K].

Type: a stalwart carrying a fast grower through a capital cycle. Services is the stalwart, mid-teens growth at roughly 41% margins; Cloud is the fast grower, now capital-intensive enough to make the whole company behave cyclically — 2026 capex guidance of $195–205B against FY2025 operating cash flow of $164.7B [Q2 2026 call; FY2025 10-K]. That decides the memo: the bear case does not live in Search losing users, it lives in a capital cycle priced as a growth story.

2. Bull case — Peter Lynch pitch

The simple reason this stock could work is that AI is making people use Google more, not less — and Alphabet owns the silicon, the models, the surfaces and the auction, so the extra usage lands on its own P&L.

A1. AI expands Search volume faster than it dilutes the price of a click

What must happen — Paid-click growth stays well ahead of any decline in cost-per-click, so Search & other keeps compounding double digits past the Q3 2026 lapping period.

Why this company — An AI answer on a Google surface still clears through Google's own auction; a rival answering the same query must build a demand side from nothing.

Evidence — Search & other grew 17% to $63.3B in Q2 2026, on paid clicks +13% against cost-per-click +3% [Q2 2026 10-Q]. Management says AI Mode passed one billion MAU and is "driving an incremental increase in Search queries overall" [Q2 2026 call].

What to monitor — The clicks/CPC split each quarter. Revenue is the output; the split is the mechanism.

A2. The Cloud backlog is real demand that converts without giving back the margin

What must happen — Just over half of the $513.9B Cloud backlog converts within 24 months, and the 35.6% operating margin survives the end of scarcity.

Why this company — Alphabet is the only hyperscaler serving its own frontier models on chips it designed, pricing capacity against an internal cost curve rather than a merchant vendor's margin.

Evidence — Cloud revenue +82% to $24.8B, operating income $8.8B, margin from 20.7$50B sequentially to $513.9B of $519.5B total RPO [Q2 2026 10-Q; call].

What to monitor — Cloud operating margin in Q3 and Q4 2026, after the bridging capacity management has flagged.

A3. Owning the silicon makes AI usage margin-accretive rather than dilutive

What must happen — Cost per AI response keeps falling faster than usage grows, so serving more AI queries adds gross profit rather than consuming it.

Why this company — Ten years of TPU development makes the inference cost curve Alphabet's own roadmap rather than a supplier's price list — and the same chips are now a merchant product.

Evidence — Management reports Q2 brought AI Mode response cost to its lowest level since launch while adding capability; Alphabet booked its first TPU system revenue, the vast majority expected in 2027 [Q2 2026 call; 10-Q].

What to monitor — Cloud margin as TPU hardware revenue scales through 2027 — hardware is structurally lower-margin than consumption services.

A4. Capex converts into revenue faster than it converts into depreciation

What must happen — Gross profit growth stays ahead of depreciation growth across a six-year server life [FY2025 10-K].

Why this company — Unlike a merchant capacity builder, Alphabet has captive internal demand: Search, YouTube and Gemini consume the compute whether or not a customer signs.

Evidence — So far it is winning: H1 2026 depreciation rose 43% to $13.6B while revenue rose 23% to $229.7B and operating cash flow rose to $84.9B from $63.9B [Q2 2026 10-Q].

What to monitor — Depreciation growth versus Cloud revenue growth. The quarter those cross is the quarter this fails.

Why the market might be missing it: reported earnings currently describe something other than the business. Q2 2026 net income of $112.2B sits on $40.8B of operating income — $99.0B of the gap is unrealised gains on equity securities, mostly SpaceX and one private company [Q2 2026 10-Q]. With that on one side and $5.8B a quarter of unallocated AI R&D on the other, both sides argue from headline numbers that do not describe the operating business, while the driver metrics that do sit disclosed and ignored.

3. Bear case — Munger invert

The most likely way I lose money is that Alphabet commits $400B+ over two years to assets with a six-year life, while the price of a click quietly deflates and the Cloud margin turns out to have been scarcity rent.

Ordered by permanence and value destruction, not by the sequence of the bull section.

B1. The spending is contractual; the demand is not

Attacks A4 — and the funding behind A2

How it could fail — 2026 capex was guided up to $195–205B from $180–190B, rising "significantly" again in 2027. Behind it sit $811.0B of purchase commitments and other contractual obligations ($200.7B short-term), plus $43.8B notional of credit derivatives and $7.6B of guarantees backstopping third-party data centres [Q2 2026 10-Q; call]. None of it cancels if AI demand normalises or inference gets cheap faster than the build-out assumes.

What would confirm failure — Cloud revenue decelerating while depreciation accelerates; backlog conversion slipping past 24 months; capex guided higher again without a step-up in backlog.

What it does to the economics — Fixed cost lands on the P&L for six years regardless of revenue. H1 2026 operating cash flow of $84.9B against $80.6B of capex left $4.3B, versus $24.3B in H1 2025; Q2 free cash flow was negative $5.9B [Q2 2026 10-Q; call].

Permanent or fixable — Permanent. Spent capital does not come back, and a six-year life means six years of it.

B2. The price of a click is deflating while the volume is celebrated

Attacks A1 directly

How it could fail — The mix of Search growth has already flipped, and the disclosure shows it. Cost-per-click contributed +7% in FY2025, +5% in Q1 2026 and +3% in Q2 2026, while paid clicks went +6%, +13%, +13% [FY2025 10-K; Q1 2026 10-Q; Q2 2026 10-Q]. AI surfaces generate more clicks of lower average commercial value — growth is now volume-led, and volume is the component with a ceiling.

What would confirm failure — Cost-per-click turning negative for two consecutive quarters while click growth decelerates toward query growth — Search & other falling to single digits without Google losing a user, while management still truthfully reports record query volume.

What it does to the economics — This is the segment funding capex, Other Bets and the AI R&D. Services earned 41.8% in Q2 2026, below Q1's 45.3% [Q2 2026 10-Q].

Permanent or fixable — Medium. If it is mix, ad formats catch up. If AI answers structurally satisfy high-value commercial queries without a click, no product fixes it.

B3. Cloud's 35.6% margin may be scarcity rent plus a hardware optical effect

Attacks A2 and A3

How it could fail — The margin nearly doubled year over year in a quarter management calls supply-constrained, and a constrained seller earns rent. Cloud also now books hardware: inventory jumped from $2.4B to $10.0B in six months as Alphabet built TPU systems for customers' data centres, the vast majority of that revenue expected in 2027 — so the Q2 margin is not yet measured on a meaningful hardware base. Management has separately guided to "modest margin pressure in the near-term" from third-party bridging capacity in Q3 [Q2 2026 10-Q; call].

What would confirm failure — Cloud margin falling into the low 30s or below once bridging capacity and TPU hardware revenue are both in the run rate; or sequential backlog growth stalling.

What it does to the economics — It is also a revenue-quality question: Alphabet holds $131.5B of non-marketable securities and backstops or funds counterparties in the ecosystem it sells to, without sizing that overlap or disclosing Cloud growth ex-TPU.

Permanent or fixable — Medium. A margin can recover; a de-rating on revenue quality is slower to undo.

B4. Remedies that hand rivals the input Google's advantage compounds on

Attacks A1 at the distribution layer

How it could fail — The December 2025 final judgment in the DOJ Search case restricts how Google distributes its services and requires it to share certain search data with, and offer syndication to, competitors; both sides appealed in early 2026. Separately, the Eastern District of Virginia ruled in April 2025 that Google's publisher ad tools unlawfully excluded rivals, and the DOJ's remedy proposal is structural [FY2025 10-K].

What would confirm failure — A divestiture order that survives appeal; or the data-sharing and syndication remedies taking effect on schedule.

What it does to the economics — Syndication is the more dangerous, because search quality is a feedback loop: a rival handed query and click data starts the loop rather than spending a decade building it, and it lands while AI is already resetting user habits.

Permanent or fixable — Permanent if structural. A divestiture removes a profit stream; a data remedy removes part of the moat.

B5. The funding regime flipped from returning capital to raising it

Attacks no single assumption — it attacks the financing of the whole plan

How it could fail — Alphabet repurchased $45.4B of stock in FY2025 and $0 in H1 2026. In its place: $30.5B of common stock issued, $19.1B of 6.25% mandatory convertible preferred and $56.2B of gross debt, taking long-term debt from $46.5B to $98.2B in six months [Q2 2026 10-Q; FY2025 10-K].

What would confirm failure — Further equity raised beyond the announced ATM programme, or debt raised at a widening spread.

What it does to the economics — Share count went from 12,088m to 12,230m and preferred dividends now appear in the EPS bridge, so per-share results absorb both dilution and a coupon. The signal matters more than the arithmetic: the company that used to shrink its share count now grows it.

Permanent or fixable — Fixable. It reverses the quarter capex stops growing.

4. Signals to monitor

Section 5 (external challenge) is not included — it runs only on request.

6. Bottom line

1. The stock could work because AI is expanding rather than cannibalising the core: paid clicks are growing double digits while Alphabet serves them on silicon it designed, on surfaces it already owns.

2. For that to pay, cost-per-click must stop decaying faster than clicks grow, the $513.9B backlog must convert near a 35% margin once supply is no longer scarce, and gross profit must outrun the depreciation from $195–205B of capex.

3. It most likely breaks through arithmetic rather than a competitor: $811.0B of irreversible commitments meeting a Search line whose price component fell from +7% to +3% in four quarters, with a structural ad-tech remedy as the tail risk.

4. What would change my mind, either way, is the clicks/CPC split and the Cloud margin after bridging capacity — two disclosed numbers, reported quarterly, that between them settle whether this capex is deferred earnings or destroyed capital.