Alphabet

Refresh delta (since 2026-06-18)

What changed — the five load-bearing moves

1. Alphabet raised $84.75B of equity — the largest single capital-markets event in its history as a public company — explicitly to fund AI capex. Announced 2026-06-01 at $80B, upsized and priced 2026-06-02 at $84.75B: $18.0B of Class A + Class C common (25,459,689 shares of each class at $355.1982 and $351.8018 respectively), $16.75B of two series of 6.25% mandatory convertible preferred (167.5M Series A + 167.5M Series B depositary shares at $50, each a 1/20th interest in a $1,000-liquidation-preference share, mandatorily converting ~May 2029), a $40B at-the-market program for Class A/C expected to begin in Q3 2026, and a $10B private placement to an affiliate of Berkshire Hathaway (14,212,035 Class A at ~$351.81 + 14,359,656 Class C at ~$348.20). All over-allotment options — 3,818,953 additional shares of each common class and 50,000,000 additional depositary shares — were exercised in full on 2026-06-03, taking the gross figure to ~$89.95B. [primary: 8-KForm 8-KA filing for something that happened between reports and matters enough to tell shareholders now. filed 2026-06-04, Items 1.01/7.01/8.01 + Exhibits 99.1 (2026-06-01) and 99.2 (2026-06-02); 8-K filed 2026-06-05, Item 1.01] ·

2. 2026 capex guidance was RAISED to $180–190B on the Q1 call (2026-04-29), superseding the $175–185B February guide — and 2027 is guided to "significantly increase." Alphabet's own words, SEC-filed: "During its Q1 2026 earnings call, Alphabet announced that its 2026 capital expenditures are expected to be $180-$190 billion, and that it expects 2027 capital expenditures to significantly increase compared to 2026." [primary: 8-K filed 2026-06-04, Exhibit 99.1, dated 2026-06-01]. The June 3 investor presentation frames it as "six times larger than 2022 and double last year's" [primary: Alphabet investor presentation, 2026-06-03, abc.xyz/investor via blog.google].

3. Cloud backlog nearly doubled sequentially to $462.3B — and the driver is a merchant TPU-hardware business that did not exist as a disclosed line six months ago. At 2026-03-31: $467.6B total remaining performance obligations, of which $462.3B is Google Cloud [primary: 10-QForm 10-QThe quarterly version of the annual report. Lighter, and not audited. filed 2026-04-30, Note 2]. The 10-Q's own explanation: "Google Cloud has entered into a limited number of agreements to supply multiple gigawatts of TPU hardware to customers who require or provide on-premises infrastructure for specialized, high-scale workloads… We expect to begin recognizing revenues from these agreements later in 2026, with the significant majority to be recognized in 2027. In connection with certain of these agreements, we have agreed to provide credit backstops to support third-party data centers and power infrastructure." [primary: 10-QForm 10-QThe quarterly version of the annual report. Lighter, and not audited. 2026-Q1, MD&A "Other Information"]

4. Buybacks went to ZERO, and Alphabet flipped from net share repurchaser to net share issuer. "In the three months ended March 31, 2026, there were no repurchases of the company's Class A or Class C shares." $69.5B of the $70.0B April-2025 authorization remained unused at 2026-03-31 [primary: 10-Q 2026-Q1]. Against that: $84.75B of announced equity issuance. The dividend was raised 5% to $0.22/quarter on 2026-04-27 [primary: 8-K filed 2026-04-29, Ex. 99.1].

5. Total debt crossed $100B pro-forma, after >$85B raised across six currencies in twelve months. Q1-2026 alone: $20.0B USD notes (4.80% WAC, 15y WAM), £5.5B sterling (5.31%, 31y), CHF3.1B (1.06%, 10y) — net proceeds $31.1B. Q2-2026 added €9B + C$9.5B (closed 2026-05-11) and ¥576.9B (closed 2026-05-21). Senior unsecured notes carrying value was $79.1B at 2026-03-31 vs. $46.5B of long-term debt at FY2025 year-end. Alphabet's own framing: "over the last year, Alphabet has raised over $85 billion of debt across six major currencies and markets, bringing its total debt balance to over $100 billion." The registered-securities list now includes 6.125% Senior Notes due 2126 — a hundred-year bond. [primary: 10-Q 2026-Q1 Note 6 + MD&A "Financing"; 8-K 2026-05-11 Item 8.01; 8-K 2026-05-21 Item 8.01; 8-K 2026-06-04 Ex. 99.1; 8-K cover-page XBRL, 2026-06]

What held

The structural thesis from 2026-06-18 stands unchanged and is not re-litigated here: the four-layer full-stack position (TPU silicon → owned data-centre/network fabric → frontier Gemini models → six 2B+-user distribution surfaces), the supply-chain map (TSMC leading-edge + CoWoSCoWoSTSMC’s method of packing a processor and its memory onto one carrier so they sit close together. Supply of it has been a hard limit on how many AI chips can be built., HBMHigh-bandwidth memoryMemory stacked in layers beside a processor so data reaches it faster. The usual bottleneck in AI chips. from SK Hynix/Samsung/Micron, grid power as the binding constraint), and the moat ranking. What changed is not the strategy — it is the financing of it. The June raise is the same thesis, funded differently.

Live questions the KB surfaced — resolved

Q1 · The backlog conflict: $240B (Feb 2026) vs ~$462B (Jul 2026) — RESOLVED

Both figures are real, both are Alphabet's own, and they are two different balance-sheet dates fourteen weeks apart — not a contradiction, and the "Jul 2026" date on the second is wrong. The full primary series, all from the same disclosure line ("Revenue Backlog," Note 2 of the relevant filing):

Reconciliation, in order of magnitude:

• The "$240B (Feb 2026)" figure is $242.8B at 2025-12-31, disclosed in the 10-K filed 2026-02-05. The KB has dated it by filing date, not measurement date. Correct label: $242.8B as of 2025-12-31.
• The "~$462B (Jul 2026)" figure is $462.3B — the Google Cloud portion — as of 2026-03-31, disclosed in the 10-Q filed 2026-04-30 and repeated by Pichai in the Q1 release ("backlog nearly doubling quarter on quarter to over $460 billion") and in the June 3 investor presentation ("Cloud's backlog nearly doubled sequentially, reaching $462 billion at the end of the first quarter"). There is no July figure. The next update is tonight. Correct label: $462.3B (Google Cloud) as of 2026-03-31. The KB is mis-dated by one quarter.
• A definitional change occurred in Q1 2026, but it explains almost none of the jump. The 10-Q states: "In the first quarter of 2026, we elected to change our reporting of revenue backlog to now also include contracts with an original expected term of one year or less. As of March 31, 2026, the portion of our revenue backlog related to contracts with an original expected term of one year or less was approximately $7.3 billion." The exclusion language also shifted (Q1-2026 excludes "cancellable contracts and payments we make to our customers not expected to be in exchange for distinct goods and services"; prior periods excluded ≤1-year and cancellable contracts). $7.3B of a ~$225B sequential increase is ~3.2%. The definitional change is real, must be disclosed when quoting the series, and is not the story.
• The story is the multi-gigawatt TPU-hardware supply agreements described above, plus — per third-party reporting — Anthropic's reported five-year, ~$200B commitment to Google Cloud and TPU capacity, which one outlet computes at "more than 40%" of the disclosed backlog. Anthropic's own newsroom confirms the compute deal and the gigawatt framing but states no dollar figure: a partnership with Google and Broadcom for "multiple gigawatts of next-generation TPU capacity" expected to come online starting in 2027 [primary-for-counterparty: anthropic.com/news/google-broadcom-partnership-compute, 2026-04-06]. The $200B number is single-sourced press reporting, not a filed figure — label it as such and do not promote it to primary.

Definitional precision for anyone quoting this series: "revenue backlog" here is ASC 606 remaining performance obligations — contractual commitments not yet recognised as revenue. It is not an order book of shipped-and-unbilled goods, and it is not comparable to a Microsoft "commercial bookings" or an Oracle "RPO" without checking each company's cancellability and duration exclusions. Alphabet's stated duration profile is "just over 50% over the next 24 months, with the remainder thereafter" — so the ~$462B is roughly a $231B/2-year near book plus a $231B long tail. Backlog excludes cancellable contracts, which is the quality test that matters; Alphabet does not disclose what fraction of the counterparties are themselves pre-profit AI labs.

The circularity that this series makes visible — every leg filed or reported, none of it inferred:

1. 2026-03: Alphabet "committed to a $40.0 billion investment in a private company consisting of a $10.0 billion capital commitment and $30.0 billion of future capital funding contingent upon the achievement of specified operational and financial milestones" [primary: 10-Q 2026-Q1 — the filing does not name the counterparty]. Press reporting identifies the counterparty as Anthropic, at a $350B valuation.
2. 2026-04-06: Anthropic announces multiple gigawatts of TPU capacity with Google + Broadcom, online from 2027 [primary-for-counterparty: anthropic.com].
3. 2026-05-05: Anthropic reported to have committed ~$200B over five years to Google Cloud and TPUs.
4. That commitment sits inside the $462.3B backlog [primary: 10-Q].
5. Alphabet "agreed to provide credit backstops to support third-party data centers and power infrastructure" serving those agreements — accounted for as credit derivatives (data-centre VIEs) and financial guarantees (energy-infrastructure VIEs), both unconsolidated [primary: 10-Q 2026-Q1, Notes 3/5/10].
6. Alphabet raises $84.75B equity + >$85B debt to build the capacity those commitments require [primary: 8-K Ex. 99.1/99.2].

That is the demand side of AI compute, financed. It is the most editorially load-bearing fact in this refresh: Alphabet's largest disclosed source of contracted future cloud revenue is a counterparty it is also funding, and whose data-centre and power suppliers it is credit-backstopping. No judgment attached — the mechanism is simply now documented end-to-end from filings.

Q2 · The 2026 capex figures: $175–185B, $180–190B, and the $610–725B aggregate — RESOLVED

Alphabet's own number is $180–190B, not $175–185B. The KB's figure is stale by exactly one quarter:

• $175–185B was guided on the Q4-2025 call, 2026-02-04.
• $180–190B replaced it on the Q1-2026 call, 2026-04-29, and is confirmed verbatim in an SEC-filed Alphabet press release [primary: 8-K filed 2026-06-04, Ex. 99.1] and in the June 3 investor presentation [primary: Alphabet IR, 2026-06-03]. 2027 is guided to "significantly increase" against that.

What the prior quarter actually spent: $35.7B in Q1 2026, vs. $17.2B in Q1 2025 — +108% YoY [primary: 10-Q 2026-Q1]. FY2025 total was $91.4B.

The arithmetic that makes tonight's capex line the most watchable number in the print:

• Q1 annualised: $35.7B × 4 = $142.8B — $37–47B below the guide.
• Guide less Q1 actual: $180B − $35.7B = $144.3B over Q2–Q4 → $48.1B/quarter; at the high end $190B − $35.7B = $154.3B → $51.4B/quarter.
• The guide therefore requires a ~35–44% step-up from the Q1 run-rate, sustained for three quarters.

The $610–725B big-four aggregate contest — resolved, and it is not a fabrication. It is three separate definitional slips stacked on top of each other. Per-company guidance, by vintage:

• High-end sum, April vintage: 190 + 200 + 145 + 190 = $725B. — the headline reconciles exactly to the sum of the top ends of the four post-Q1 guidance ranges. The KB's claim that "~$725B exceeds the sum of its own per-company parts" is arithmetically correct against the parts the KB was holding (the February vintage, and/or Microsoft's fiscal-year number) and arithmetically wrong against the current parts.
• Midpoint sum, April vintage: 185 + 200 + 135 + 190 = $710B.
• Low-end sum, April vintage: 180 + 200 + 125 + 190 = $695B.
• The ~$610B figure is the February vintage with Microsoft on a fiscal-year basis: 175 + 200 + 115 + 120 = $610B.

Three axes explain the entire $610–725B spread, and any citation must fix all three:

1. Guidance vintage — Feb (Q4-25 calls) vs. Apr (Q1-26 calls). Three of four raised in April; the aggregate moved ~$100B in one quarter.
2. Microsoft fiscal vs. calendar — ~$120B FY2026 (Jul-2025→Jun-2026) vs. ~$190B calendar-2026. A ~$70B swing on one company, and the single largest source of the spread.
3. Basket membership — "big four" vs. "top five hyperscalers including Oracle." CreditSights' ~$750B is a five-name basket and is not comparable to any four-name number.

A fourth, softer axis: some published aggregates are derived from a growth rate, not summed from guidance (e.g. $410B × 1.77 = $725.7B), which coincidentally lands near the high-end sum and gets quoted interchangeably with it. Treat any aggregate without a stated basket, vintage, and fiscal convention as unusable.

Q3 · What would have to be true tonight for "hyperscaler capex is still accelerating" to break?

Stated as falsification conditions, not predictions. Each is a specific line in tonight's release; the implication is mechanical, not a view.

The compact version: the acceleration reading breaks if Q2 capex prints below ~$45B, or the FY2026 range/2027 language is walked back, or the Cloud backlog falls sequentially. Any one of those three is sufficient. Cloud growth, Cloud margin, and the depreciation/useful-life lines determine whether an intact capex number is still earning.

---

Phase A — carried forward

• Lens 1 · Company Overview — carried, see the previous dossier. Three material additions from the Q1-2026 10-Q that the prior dossier did not capture: (a) Wiz closed 2026-03-11 for $29.5B, after purchase-price adjustments and excluding post-combination compensation, and now reports inside the Google Cloud segment — Q1-2026 Cloud revenue and margin are therefore not clean organic comparisons; (b) Intersect (Intersect Power) closed 2026-03-10 for $5.9B, allocated across segments — an outright acquisition of a clean-energy developer, i.e. Alphabet buying power generation, not contracting for it; (c) GFiber is being contributed to a newly formed entity for $1.5B cash + a $2.0B note receivable + a 49.99% equity interest, expected to close late 2026. Headcount 194,668 at 2026-03-31 vs 185,719 a year earlier. [primary: 10-Q 2026-Q1]
• Lens 2 · Supply Chain — carried, see the previous dossier. One structural change: the chain now runs outward as well as inward. Alphabet is supplying TPU hardware into third-party and on-premises data centres, and is credit-backstopping the data-centre and power-infrastructure entities that host it [primary: 10-Q 2026-Q1]. It is simultaneously customer, supplier, investor, and credit support to the same buildout. Third-party reporting names TeraWulf and Cipher Mining as data-centre infrastructure providers and Fluidstack as on-site operator in the Anthropic TPU arrangements — not filed, do not treat as primary.
• Lens 3 · Moats — carried, see the previous dossier. The one moat item genuinely re-scored is TPU merchant supply (below, Lens 14 close-out).
• Lens 4 · Segments — carried for FY2025; the Q1-2026 quarterly split is restated below as the comparison base for tonight.

Q1-2026 segment actuals — the base tonight is measured against [primary: 8-K filed 2026-04-29, Ex. 99.1]:

Derived: Google Services margin 45.3%; Google Cloud margin 32.9%. The Alphabet-level activities line — where shared AI R&D sits — is growing faster than any revenue line and is now a $5.4B/quarter drag; it is the least-discussed cost line in the company.

Year-ago Q2 comparison bases, derived from the shelf:

Cross-check: consensus revenue of ~$116.84B against the derived $96,427M implies +21.2% YoY, which matches the growth rate quoted alongside that consensus — the derivation is sound.

---

Phase B — re-run

Earnings setup (PRE-PRINT — there is no Q2 result)

Reporting: 2026-07-22, after the US close.

Consensus expectations:

EPS-quality warning — this is the single most misreadable number tonight. Q1-2026 diluted EPS was $5.11, but OI&E was $37.7B including $36.9B of net gains on equity securities, "primarily related to unrealized gains on our non-marketable equity securities" [primary: 8-K 2026-04-29 Ex. 99.1; 10-Q 2026-Q1 MD&A]. The cash-flow statement backs out $36,804M as "Loss (gain) on debt and equity securities, net" [primary: 10-Q, Consolidated Statements of Cash Flows]. Consensus of ~$2.89 for Q2 evidently assumes approximately zero equity-mark contribution. Therefore:

• A Q2 EPS print far above $2.89 driven by marks is not an operating beat — check OI&E before the headline.
• A Q2 EPS print below $2.89 because marks reversed is not an operating miss — non-marketable equity marks are volatile in both directions and reverse violently.
• Read income from operations (+30% in Q1) and the segment operating-income lines. Ignore net income for judging the business. This is unchanged guidance from the 2026-06-18 dossier and is now more, not less, important: the mark contribution has grown, and Alphabet's non-marketable book has grown with the Anthropic commitment.

Balance-sheet / cash-flow items entering the print [primary: 10-Q 2026-Q1 unless noted]:

• Operating cash flow $45.8B in Q1-2026; $174B TTM to 2026-03-31 [primary: 8-K Ex. 99.1].
• Capex $35.7B in Q1-2026 → implied quarterly FCF ~$10.1B.
• Depreciation of property and equipment $6.5B (Q1-26) vs $4.5B (Q1-25).
• Stock-based compensation $6,751M (Q1-26) vs $5,516M.
• $75.6B of future lease payments on data-centre leases that have NOT YET COMMENCED, commencing between 2026 and 2031, non-cancelable terms primarily 1–25 years. Total commenced undiscounted lease payments: $18.8B operating + $2.6B finance. The not-yet-commenced figure is roughly 3.5× the entire commenced lease book and does not appear on the balance sheet today. This is a genuine off-balance-sheet capacity commitment the prior dossier did not surface.
• Zero buybacks; $69.5B of $70.0B authorisation unused.
• Commercial paper programme $25.0B, zero outstanding; credit facilities $11.7B with $1.2B drawn.

Market reaction context entering the print:

• Close 2026-07-21: $347.15; market capMarket capitalisationThe share price multiplied by the number of shares. What the market says the equity is worth. ~$4.29T; YTD +11.05% total return.
• All-time closing high $402.38 on 2026-05-13 → currently −13.7% from the high.
• Below the June 2 offering prices of $355.1982 (Class A) and $351.8018 (Class C) — −2.3% — and below Berkshire's private-placement prices of $351.81 / $348.20 — −1.3%. Every buyer of the June raise, Berkshire included, is currently underwater.
• Prior-print reaction function: the Q4-2025 print (2026-02-04) produced the year's worst single-day move, −8.7% on a Cloud revenue miss; the Q1-2026 print (2026-04-29) was received as the sector's best capex-justification story.

Management communication (no new earnings call since 2026-06-18 — but two new primary documents)

No earnings call has occurred since the prior dossier. However, the June capital raise produced two Alphabet-authored primary documents that function as a management-tone update, and the shift from the Q1 call is measurable:

• The capacity boast hardened into a stated shortage. Q1 framing was "capacity-constrained." The June 1 release says demand is "exceeding the company's available supply"; the June 3 investor presentation escalates to "meaningfully exceeding our available supply." [primary: 8-K Ex. 99.1; investor presentation 2026-06-03]
• A new register appeared: balance-sheet defensiveness. The June 1 release includes a section headed "Investing in a Balanced Way," which pre-emptively lists operating cash flow and debt capacity alongside the equity raise, and states the plan is "to fund its investments in a balanced way while retaining a healthy balance sheet." Companies that are not being asked about balance-sheet strain do not write that paragraph. [primary: 8-K Ex. 99.1]
• Token throughput became the headline demand metric. Q1 release (2026-04-29): "more than 16 billion tokens per minute via direct API use… up 60% from last quarter." June 1 release: "19 billion tokens per minute, a 6x increase year-over-year," plus "over 8.5 million developers." [primary: both 8-K exhibits]
• Other Q1 operating disclosures now serving as tonight's bars: 350M paid subscriptions; Gemini Enterprise paid MAU +40% QoQ; Waymo past 500,000 fully autonomous rides/week; 11th consecutive quarter of double-digit revenue growth [primary: 8-K 2026-04-29 Ex. 99.1].
• Signature detail: the May 11 and May 21 note-offering 8-Ks are both signed by CFO Anat Ashkenazi; the June 5 accounting-officer 8-K is signed by an Assistant Secretary. Alphabet appointed Marsida Saraci (VP & Controller, at Alphabet since 2011, ex-KPMG) as Principal Accounting Officer effective 2026-06-05 [primary: 8-K filed 2026-06-05]. A routine appointment, noted for completeness — the CFO seat is unchanged.
• transcripts/ remains empty on the shelf. The Q1-2026 call (2026-04-29) has not been ingested; this refresh reads the call's guidance only through Alphabet's own SEC-filed restatement of it. Shelf gap — flagged in Open Items.

Comps — carried, with a staleness flag

Carried from the previous dossier. Do not reuse the multiples in that table without re-sourcing: they were mid-June marks, and GOOGL has moved from $370.84 to $347.15 (−6.4%) since. The only comp-relevant fact re-sourced this session is Alphabet's own: ~$4.29T market cap at the 2026-07-21 close, YTD total return +11.05%. Peer multiples: n/a — not re-sourced this session. Fabricating a refreshed comp table off a 34-day-old base would be exactly the failure mode the provenance rules exist to prevent.

One structural note the comp table cannot express: the mandatory convertible preferred introduces a preferred dividend ahead of common for the first time — 6.25% on $19.25B of liquidation preference (including the fully-exercised greenshoe) ≈ ~$1.20B/year, accruing until mandatory conversion ~May 2029. Any EPS comparison across the conversion date, or against peers with no preferred layer, must handle this. Capped call transactions were entered on 2026-06-02/03 to limit the dilutionDilutionIssuing new shares, so each existing share owns a smaller slice of the same company. on conversion [primary: 8-K 2026-06-05, Item 1.01].

Stock-price catalysts (new events since 2026-06-18)

• 2026-06-01/02 — the $80B → $84.75B equity raise. The defining event of the period. Announced 2026-06-01 pre-market, priced 2026-06-02, common closed 2026-06-04, preferred closed 2026-06-05 [primary: 8-Ks]. The stock has traded below the offering price ever since.
• 2026-06-01 — Berkshire Hathaway's $10B private placement. A step-change from the Q3-2025 secondary-market stake the prior dossier logged as a "sentiment stamp": Berkshire is now a primary-issuance participant with registration rights, and its total position reportedly tops $31B.
• Mid-July 2026 — a Buffett endorsement moved the stock ~+3.65% to ~$370.36 intraday, since retraced to $347.15 by the 2026-07-21 close.
• 2026-06-05 — 2026 Annual Meeting. All ten directors re-elected (Page, Brin, Pichai, Hennessy, Arnold, Chávez, Doerr, Ferguson, Shriram, Washington); E&Y ratified; the 2021 Stock Plan reserve increased by 200,000,000 Class C shares; say-on-pay approved with a notable 2.33B votes against (~19% of votes cast). The equal-shareholder-voting proposal drew 3.85B for vs 8.50B against — ~31% support, by far the strongest of the fourteen proposals and a standing signal on the dual-class structure. All twelve shareholder proposals failed. [primary: 8-K filed 2026-06-11, Items 5.02/5.07]
• 2026-07 — Frozen v2. Reporting that Google is developing a Gemini-architecture-specific inference chip targeting 6–10× the inference efficiency of current TPUs, deployment targeted as early as 2028, and reportedly not intended for external Cloud customers. Unconfirmed by Alphabet; not filed.
• 2026-Q2 — power. Google reportedly leased part of a record 9.6 GW of power in Q2 for AI data centres, and struck an agreement to curtail data-centre power draw during peak grid demand. Directionally consistent with the filed Intersect acquisition and the energy-infrastructure guarantees; treat the gigawatt figure as unverified.
• Ad-tech remedy (EDVA, Judge Brinkema): still pending, no ruling located. Reporting from the closing-argument record continues to describe the judge as skeptical of structural relief — she reportedly noted no buyer for AdX had been identified and called the divestiture proposal abstract. No change from the 2026-06-18 dossier's read: this remains the live structural risk and is unresolved.

Updated reaction function. The prior dossier said the market had shifted from trading Search to trading Cloud growth/margin + antitrust headlines. That is now incomplete. Since June the stock has traded on capital structure: it fell into and through an $84.75B equity raise, and the largest single day of the period was a Buffett soundbite, not an operating datapoint. Tonight is the first operating print since the market re-priced Alphabet as a company that issues equity.

---

Phase C — partial re-run

Management & capital allocation (this is where the change is)

The people are unchanged (Pichai CEO; Ashkenazi CFO; Page & Brin retaining voting control via Class B; Saraci newly appointed Principal Accounting Officer). The capital-allocation regime is not.

The single sharpest capital-allocation fact: in the space of five quarters Alphabet went from returning $62.2B/year to shareholders via buyback to issuing $84.75B of equity — a swing of roughly $147B/year in the direction of capital travel. The dividend was still raised 5% to $0.22/quarter through it.

Dilution arithmetic [all estimates, inputs primary]:

• Diluted share count ≈ 12,246M.
• New common, underwritten + greenshoe: 29,278,642 Class A + 29,278,642 Class C = 58.6M shares.
• Berkshire private placement: 14,212,035 + 14,359,656 = 28.6M shares.
• Mandatory convertible, at conversion ~May 2029: $19.25B ÷ ~$355 ≈ ~54M share-equivalents, before capped-call offset and before the variable conversion-rate mechanics (actual count depends on the share price at conversion, so this is an order-of-magnitude figure only).
• ATM, if fully drawn at ~$350: ~114M shares.
• Total ≈ 255M shares ≈ ~2.1% of the current base.

That is modest as a percentage — but it is a reversal of sign, not a reduction in magnitude. The 2026-06-18 dossier's forward model assumed a 1.5–2% annual share-count shrink as an EPS tailwind. That input is now wrong in direction.

One nuance in the ATM that materially changes how it should be read. Alphabet states the $40B ATM is not primarily capex funding: proceeds will be used "primarily to facilitate, for a period of time, an administrative change to how it meets tax obligations associated with vesting of employee equity awards," with "approximately $30 billion of ATM program proceeds… used to meet these 2026 calendar year tax obligations" [primary: 8-K 2026-06-04, Ex. 99.1]. Mechanically, Alphabet is moving from funding RSU tax withholding out of corporate cash (net share settlement) to funding it by selling shares into the market. The share-count effect is roughly a wash versus the old method; the cash effect is a ~$30B one-year preservation of corporate cash, redeployable to capex. So the honest split of the $84.75B is: ~$34.75B underwritten + $10B Berkshire = ~$44.75B genuinely new capex funding, plus a ~$30B cash-preservation manoeuvre, plus ~$10B of ATM residual for general corporate purposes. Anyone quoting "$84.75B raised to fund AI capex" is quoting Alphabet's headline, not Alphabet's own use-of-proceeds paragraph.

Forensic red flags (three new items, two carried)

Carried unchanged from 2026-06-18: zero SEC Litigation Releases, zero AAERs 2021–2026 — clean on the accounting-fraud axis. The full competition-law map (US v. Google Search remedy and appeals, EDVA ad-tech pending, EC €3.0B ad-tech, EC legacy cases) is carried by reference; no resolution located since 2026-06-18.

New / re-scored, all filing-grounded:

1. Earnings quality deteriorated further, and is now quantified. Q1-2026 OI&E $37.7B, of which $36.9B was net gains on equity securities, primarily unrealised marks on non-marketable holdings [primary: 8-K 2026-04-29 Ex. 99.1; 10-Q cash-flow add-back $36,804M]. That is ~93% of income from operations ($39,696M) arriving as unrealised marks on private stakes. Net income rose 81% while operating income rose 30%. The non-marketable book is about to grow further via the $40B Anthropic commitment. Marks on a private counterparty that is simultaneously the largest reported contributor to your revenue backlog are the highest-scrutiny line in these financials. No allegation of impropriety — the disclosure is complete and unambiguous; the point is that headline EPS is not a measure of this business.

2. Off-balance-sheet exposure expanded on three fronts simultaneously [primary: 10-Q 2026-Q1, Notes 3/4/5/10]:

• $75.6B of not-yet-commenced data-centre lease payments (2026–2031, terms to 25 years) — not on the balance sheet today, ~3.5× the commenced lease book.
• Credit backstops to data-centre VIEs, accounted for as credit derivatives, unconsolidated. Maximum exposure disclosed only as "the financial risk over the remaining period of the arrangements, as reflected by the credit derivative notional value" — no aggregate notional is given in the MD&A text retrieved.
• Backstop agreements to energy-infrastructure VIEs, accounted for as financial guarantees, unconsolidated, with maximum exposure "limited to the potential amount of future payments." The 10-K accounting policy already contemplated this: "we provide reimbursements for costs incurred by third parties during power generation project development phases if specified trigger events occur". Alphabet is now extending credit into the AI-datacentre and power supply chain, off balance sheet, in service of contracts booked into its own backlog. Quantifying the notional is an open item.

1. Useful lives are now the highest-leverage estimate in the company, and remain unchanged so far. Q1-2026 depreciation on property and equipment $6.5B, up 44% YoY, against a capex base guided to roughly double [primary: 10-Q]. Alphabet explicitly flags useful lives as a critical estimate subject to change from "expected technology advancements, and future network deployment plans" [primary: 10-Q, Risk Factors]. No change to useful-life assumptions was disclosed in the Q1-2026 10-Q. Any extension announced tonight or later is a margin lever, not a margin outcome — watch item #7 above.

2. Segment comparability broke in Q1-2026. Wiz ($29.5B) folded into Google Cloud on 2026-03-11; Intersect ($5.9B) allocated across segments from 2026-03-10 [primary: 10-Q]. Cloud's Q1-2026 revenue and 32.9% margin therefore include ~three weeks of Wiz. Neither the 10-Q text retrieved nor the press release quantifies the contribution. Cloud "organic" growth is not separately disclosed — an open item, and one that matters more tonight, when Q2 will carry a full quarter of Wiz.

3. Concentration risk migrated from the revenue statement to the backlog. The 10-K's standing disclosure — "No individual customer or groups of affiliated customers represented more than 10% of our revenues in 2023, 2024, or 2025" — is a revenue test and remains true. It says nothing about the backlog, where a single reported counterparty is described in press reporting as >40%. Alphabet does not disclose backlog concentration. A revenue-diversified company can have a highly concentrated forward book, and this one now appears to.

---

Phase D — re-run as mechanics (no verdict)

Forward model — inputs restated, projection NOT re-issued

The 2026-06-18 EPS ladder (bear ~$12.0 / base ~$13.0 / bull ~$14.0 for FY2026) is withdrawn, not updated. Two of its explicit inputs are now falsified:

Re-issuing a three-year EPS ladder tonight — hours before the print that resets the base, with Wiz contribution unquantified, useful lives unconfirmed, and the ATM's timing undisclosed — would be manufacturing precision. Deliberately not done. No Brier forecast logged (dispatched run; and a forecast on an unresolved base is not a calibration signal).

What is offered instead — the funding identity, which is the arithmetic that actually governs the model:

TTM operating cash flow (to 2026-03-31) $174B [primary: 8-K Ex.99.1]
FY2026 capital expenditure guidance $180–190B [primary: 8-K Ex.99.1]
 ──────────
Gap before dividends −$6 to −$16B
Dividends (~$0.22/qtr × ~12.25B shares) ~−$10.8B
 ──────────
Gap before any buyback −$17 to −$27B

Funded by, over the last ~12 months:
 Equity raised (announced, June 2026) $84.75B [primary]
 Debt raised (six currencies) >$85B [primary]
 ──────────
 Total external capital ~$170B

The line that matters for the "Who Pays the Rent" spine: Alphabet's 2026 capex guidance exceeds its trailing-twelve-month operating cash flow. The most cash-generative advertising business ever built no longer funds its own buildout from operations. That is a filed, arithmetic fact, not an interpretation — and it is why the June raise happened. Whether that is prudent, aggressive, or irrelevant is not assessed here.

The two readings, stated as what each requires (NOT a bull/bear verdict)

Presented as mechanics — what each reading must be true for. No view is taken on which is correct.

For the reading "the buildout is earning its capital," these must all hold:

• Cloud revenue growth stays near the Q1 pace (~$22B+ in Q2) while capex steps to ~$48–51B/quarter.
• Cloud operating margin holds ≥ ~30% on the larger base — i.e. incremental capacity comes on at or above installed-base economics.
• Backlog grows, not merely holds, meaning new bookings exceed the ~$22B/quarter being recognised.
• Depreciation growth stays below revenue growth, with useful lives unchanged.
• Operating cash flow accelerates toward the capex line, shrinking the external-funding requirement rather than widening it into the 2027 step-up.

For the reading "this is a financing story dressed as a growth story," these must all hold:

• The backlog's quality is weaker than its size — concentrated in counterparties that are themselves pre-profit and, in the largest reported case, partly funded by Alphabet, with cancellability terms not disclosed.
• Capex is committed (leases, TPU supply agreements, credit backstops) ahead of the revenue that services it — the $75.6B of not-yet-commenced leases and the "significant majority to be recognized in 2027" language both point this way.
• The equity raise, at 6.25% preferred and into a falling share price, is a cost-of-capital signal rather than an opportunistic one.
• Depreciation and the Alphabet-level AI R&D line (+78% YoY) compress margin faster than Cloud expands it.
• Reported earnings quality continues to depend on unrealised marks on private counterparties inside the same ecosystem.

The point of stating both this way: every one of the ten items above resolves against a specific, disclosed line in tonight's release or the 10-Q that follows it. Neither reading requires a forecast. This dossier takes neither.

Structural questions a skeptic would put to the filings (questions, not conclusions)

1. What fraction of the $462.3B backlog is cancellable, and on what notice? Alphabet excludes cancellable contracts from backlog by policy — so the disclosed figure is, by its own definition, non-cancellable. What are the termination-for-convenience and milestone provisions inside it? Not disclosed.
2. What is the aggregate notional of the credit-derivative and financial-guarantee backstops to data-centre and energy-infrastructure VIEs? The 10-Q describes the accounting but the retrieved MD&A text gives no aggregate. Not quantified here.
3. How much of the ~$462B backlog is a counterparty Alphabet is also funding? The $40B commitment and the reported ~$200B backlog contribution have never been reconciled in a filing — the 10-Q does not name the counterparty.
4. What did Wiz contribute to Cloud's Q1-2026 revenue and 32.9% margin, and what will a full quarter contribute to Q2? Not disclosed.
5. Why was the capex-funding equity raised at a 6.25% mandatory-convertible coupon by a company with a $25B untapped commercial-paper programme, $10.5B of undrawn credit facilities, and a >$100B cash-and-securities position? The instrument choice is information.
6. If Q2 capex prints near the ~$48–51B/quarter run-rate the guide requires, what is the 2027 number the company has already said will "significantly increase" — and against a $174B operating-cash-flow base, what closes that gap?
7. What happens to reported earnings if the non-marketable book marks down as violently as it marked up ($36.9B in one quarter)?

Management questions — carried, with one closed and four added

The fifteen questions in the previous dossier stand. Question 8 is now answered by the filings, not by management: "does the TPU become a merchant product?" — Yes. The 10-Q discloses agreements to supply "multiple gigawatts of TPU hardware to customers who require or provide on-premises infrastructure," with revenue recognition beginning later in 2026 and the significant majority in 2027 [primary: 10-Q 2026-Q1]. TPU is no longer only a captive cost advantage; it is a merchant hardware line with its own backlog, its own credit exposure, and — per reporting — a Broadcom-mediated supply channel.

Four questions added by this refresh:

1. What proportion of the $462.3B Cloud backlog is attributable to your three largest counterparties, and what proportion of those counterparties are entities in which Alphabet holds an equity interest or has extended credit support?
2. What is the aggregate notional of the credit backstops extended to third-party data-centre and power-infrastructure entities, and under what trigger events do they become payable?
3. Given a $25B untapped commercial-paper programme and a >$100B liquidity position, why was $16.75B of the June raise structured as 6.25% mandatory convertible preferred rather than senior debt?
4. With 2027 capex guided to "significantly increase" from $180–190B against $174B of trailing operating cash flow — what is the multi-year funding plan, and at what point does the equity issuance stop?

---

Tonight's scorecard — the numbers to read, in order

Read in this sequence. Items 1–3 determine the regime; 4–6 determine whether the regime is earning; 7–9 are the quality checks.

Secondary, all from Q1-2026 bases [primary: 8-K 2026-04-29 Ex. 99.1]: Search & other vs. $54,190M (Q2-25 derived); YouTube ads vs. $9,796M; Network vs. $7,354M (declining line); subscriptions count vs. 350M; Alphabet-level activities vs. −$5,391M; Other Bets loss vs. −$2,100M; tokens/minute vs. 19B (June); operating cash flow vs. $45.8B.

---

Falsified by this refresh — corrections to the previous dossier

1. **"Cloud backlog ~$155B (+46% YoY)" — WRONG AND STALE.** The correct figure at the time of writing was **$462.3B (Google Cloud) at 2026-03-31**, disclosed in `filings/10-q-2026-q1.md` — **a file already on the shelf and cited elsewhere in that same dossier.** The dossier reached for a web summary of a Sept-2025 number when a primary source two quarters newer and 3× larger was in its own directory. **This is a textbook silent-underread: a read that returned less than the truth and looked like a healthy quiet result.** The lesson generalises — when a figure and a `` file cover the same metric, the file wins and must be checked first.
2. "buybacks stepped down as capex stepped up" — understated. Buybacks went to zero in Q1-2026 [primary: 10-Q].
3. Lens 11 input "Buybacks ~1.5–2% annual share-count shrink" — wrong in sign. Alphabet is a net issuer; ~2.1% dilution from announced programmes.
4. "long-term debt $10.9B → $46.5B… still trivial against the cash pile" — superseded. Senior notes carrying value $79.1B at 2026-03-31; >$100B total debt pro-forma after the Q2 offerings [primary].
5. Lens 5's "$35.7B in the single quarter… annualize and you are north of $140B" — wrong frame. Guidance is $180–190B; Q1 was the low quarter, not the run-rate.
6. Lens 14 Q8 (TPU as merchant product) — answered YES by the filings.
7. Entirely absent from the prior dossier, all disclosed in filings it had on disk or filed within its window: the Wiz close ($29.5B, into the Cloud segment), Intersect ($5.9B), the $40B private-company commitment, Waymo's $16.0B raise (majority Alphabet-funded), the $31.1B Q1 note issuance, the $75.6B of not-yet-commenced data-centre leases, the credit backstops to data-centre and energy VIEs, the GFiber contribution, and the precise $36.9B equity-mark figure.
8. Not falsified, explicitly re-verified as still standing: the antitrust map (Search remedy + appeals; EDVA ad-tech pending with no ruling located; EC €3.0B on appeal), the zero-SEC-enforcement finding, the full-stack moat structure, and the supply-chain chokepoints.

---

Value-capture trace — who actually captures the money

Added 2026-07-22 per the standing money/Quinn steer: follow the money through DOCUMENTS, end at a business decision. Mechanism only — no directional call. Every leg below is filed or explicitly labelled as press-sourced.

The chain, in order of cash movement:

1. Alphabet raises $84.75B equity + >$85B debt over 12 months [primary: 8-K 2026-06-04, 8-K 2026-06-05, 10-Q Q1-2026]. Buybacks go to zero — the shareholder stops being paid so the buildout can be. Capital providers capture: 6.25% on the mandatory convertible, 6.125% on hundred-year paper. That is the price of the money, and it is the first hard number in the chain.
2. Alphabet commits $40B to "a private company" — $10B now, $30B milestone-contingent; the 10-Q declines to name it [primary: 10-Q Q1-2026]. Press says Anthropic ``.
3. That party commits ~$200B back to Google Cloud/TPUs ``, landing inside the $462.3B Cloud backlog at 2026-03-31 [primary: 10-Q].
4. Alphabet credit-backstops the data centres and power serving that capacity [primary: 10-Q, VIE disclosures] — so it carries the counterparty risk on its own customer's obligations.
5. Alphabet spends $180–190B buying the physical layer [primary: 8-K Ex-99.1 2026-06-04].

Where the margin actually lands — and it is not with the buyer. Alphabet's TTM operating cash flow is $174B against that guide: capex now exceeds operating cash flow before dividends. Amazon is the same shape — TTM FCF $1,232M, −95% YoY [primary: 10-Q]. The buyers are cash-flow-negative on the buildout. The counterparties are not:

• Memory — Jassy, on the record: "the cost of components, particularly memory, has skyrocketed. There is just not enough capacity" [primary: AMZN Q1-2026 call], with memory volatility newly added to Amazon's formal safe-harbour language. Samsung's preliminary Q2 operating profit is ~19x YoY [primary: Samsung preliminary release 2026-07-07 — PRELIMINARY only; full results 2026-07-30].
• Advanced packaging — CoWoS and N3 reported sold out through end-2026 with lead times into 2027 ``.
• Power and plant — DLR's Q1 new leasing was its second-highest quarter ever, ~70% above the next highest, with the largest single lease in company history at 200 MW [primary: DLR 10-Q Q1-2026].
• Capital providers — see leg 1.

The portable law: in this buildout, the party announcing the spend is not the party capturing the margin. Announced capex is a transfer, and the filings say where it lands — into memory, packaging, power and interest. A reader deciding where to be positioned should be reading the suppliers' pricing power and the buyers' funding costs, not the buyers' capex headlines.

The unresolved leg, stated plainly: the ~$200B customer commitment that anchors the backlog is single-sourced to press and appears in no filing. If it is real, the backlog is demand. If the counterparty is substantially funded by Alphabet's own $40B commitment, part of the backlog is Alphabet paying itself, and the trace above is partly circular. The filings do not resolve this, and no number on tonight's print will resolve it either — it resolves only if a counterparty names the commitment in its own disclosure.