A bank forecasting a business its subject does not report leaves the reader one obvious question and no way to answer it. Alphabet publishes no TPU units, no gigawatts and no TPU revenue. The narrower question: which slice has already met a filed number?
Oppenheimer reiterated Outperform on Alphabet on 10 September and put 6.6 gigawatts and about $170 billion of cumulative TPU sales through 2028 into its Google Cloud estimates, roughly $19 billion of it in 2026. Alphabet's June quarter carries one line that moves with those sales, and it is small: inventory of $9.99 billion at 30 June against $2.44 billion six months earlier.
Alphabet reports no TPU dollars, and one balance-sheet line that moved 4.1x
The June 10-Q is where the merchant business becomes visible. Google Cloud revenue now includes "product sales, primarily the sale of TPU systems," and inventory "consists primarily of hardware related to TPU systems for sale to enterprise customers and devices."
Alphabet gives no dollar figure for either. What it gives is the stock of goods: $2,439 million at the end of 2025, $9,991 million at 30 June. The cash flow statement books that as a $7,739 million inventory build in the half, against $628 million a year earlier.
That balance bounds nothing by itself, and three things loosen it. Devices sit in the same line. Revenue is recognised on delivery at a gross margin Alphabet declines to split: CFO Anat Ashkenazi told the July call the company does "not break out margins for any specific products or infrastructure component." And Alphabet keeps building, so 30 June is not the year's supply. What the line gives the forecast is something to clear, every quarter, in public.
$19 billion in 2026 means $8 billion a quarter in the half Alphabet called small
Deliveries started in the June quarter. Cloud revenue that quarter was $24,768 million in total, GCP and Workspace and TPU together, up 23.7% on the March quarter's $20,028 million. Ashkenazi called the TPU portion "a small amount of that total agreement" and said Cloud growth "accelerated meaningfully even after excluding the impact of TPU system sales."
Take her at three values. If TPU was 5% of the June quarter, $1.2 billion, the second half has to carry $17.8 billion; at 10%, $16.5 billion; at 15%, $15.3 billion. Every path lands between $7.6 billion and $8.9 billion a quarter, in the two quarters Alphabet has described as the ramp rather than the arrival.
The company's own sentence points the other way. On the same call Ashkenazi said Alphabet expects "to recognize a relatively small portion of the revenues from our existing TPU system sales agreements this year, ramping as we exit 2026," with "the vast majority of the revenues from these agreements" landing in 2027.
Oppenheimer's shape is later still. Its three years — $19 billion, $50 billion, $101 billion — put 59% of the $170 billion in 2028, a year for which, on Alphabet's account of what it has signed, the revenue is not contracted yet. The note says as much in its own way: it calls the $170 billion incremental to Street estimates.
6.6 gigawatts implies $25.8 billion each, and nobody publishes the rate
$170 billion ÷ 6.6 gigawatts = $25.8 billion of revenue per gigawatt. Neither Alphabet nor the note publishes a per-gigawatt price, so that figure is the rate the two published numbers imply, not a disclosed one.
Gigawatts of Google TPU capacity do exist in the public record, disclosed by the counterparties rather than by Alphabet: Anthropic's million-TPU commitment, well over a gigawatt in 2026, and roughly 3.5 GW of Broadcom-supported capacity from 2027, which we set out in August. Those are capacity commitments, mostly consumed through Google Cloud. Oppenheimer's 6.6 GW is system sales into customers' own data centres, and nobody has split one from the other, so the two sets cannot be netted.
Our Cloud line runs at 10% a quarter, and the gap is worth $176 a share
The GOOGL model grows Cloud 10% sequentially, and its note gives the reason: "Do not use 24% sequential as the opening rate. That quarter includes first TPU system sales. 10% QoQ is already a very fast cloud." That path puts the December half at $56.9 billion and the full year at $101.7 billion, then $142.0 billion in 2027 and $176.5 billion in 2028.
Set Oppenheimer's TPU line against those and it is 19%, 35% and 57% of our entire projected Cloud segment in each year. The disagreement is not about TPUs. It is about the growth rate of the whole line.
Priced: at 10% sequential our base-case fair value is $311.82, against $348.06 on 25 August, the price the model was struck against. At 20% it is $376.30; at 30%, the rate the June quarter printed, $488.24, or $176.42 above base. Those figures are ours.
Two filed lines settle the 2026 slice in October: Cloud revenue and the inventory balance. Cloud near $27 billion with inventory flat leaves $19 billion unreachable. Cloud above $30 billion with inventory climbing again means the ramp arrived a quarter earlier than guided, and Oppenheimer's year is live.
Sources: Oppenheimer's 10 September note is press-reported, not opened here; TheFly via TipRanks and aktiencheck.de independently carry the 6.6 GW, the $170 billion and the 15%/30% Cloud upside; the annual split is from a 404K Research brief. Inventory, cash flow, Cloud revenue and the TPU product-sales and inventory language are Alphabet's Q2 2026 10-Q. Call quotes are from a third-party transcription of the 22 July call, non-authoritative for figures. Anthropic's and Broadcom's gigawatts are their own disclosures. Per-gigawatt rate, quarterly requirements, segment shares and every fair value are ours.
