Demand for Vertiv's cooling kit held; the share price did not
This tests the standing view that AI build-out spending runs on power and cooling, not on chip supply.
Drafted with Claude Fable 5.1 · read and committed by the publisher How we work →
What would change our mind
recognized revenue failing to catch up with deferred revenue over the next two quarterly reports; or the hyperscalers cutting their capital-spending plans in their next quarterly reports; or a data-center project of 100 megawatts or more deferred citing surplus capacity.
Corrected 13 October 2026: an earlier version gave two share-price falls of 14 and 17 percent, citing the House View. The House View removed those figures for lack of a primary source, so they are removed here.
Dry run, told honestly. Before this publication had a single reader, it ran a trial on itself: it wrote down what it thought about Vertiv, dated it, and waited for the evidence. This is the result. The claim survived. The price did not.
The question
When a company sells less than it had guided but its business is plainly getting better, which one was wrong — the business, or the expectation?
Which part of the House View this bears on
Two parts. The variableVariable: something that is moving and that the frame depends on. We describe it; we do not forecast it. we call The market is paying less for good news — over the summer, companies that beat expectations and raised their forecasts saw their shares fall anyway, and we hold that this is the market paying less for every dollar of earnings, not the businesses failing (the House View; our claim on it). And the watchpointWatchpoint: a claim we have put on the record with its date, what would prove it wrong, and the date we will check. Each one has its own page. on power and cooling — that the demand for the equipment under the artificial-intelligence (AI) build-out runs on capital spending, not on chip supply, and does not slow when rented compute goes on sale (Spare computing capacity does not stop new power construction). Vertiv is the clearest single test of both, which is why it was the trial.
Strong results are earning smaller rewards, because money costs more.
What is known
As of the second-quarter report, dated 29 July 2026 (Vertiv Holdings, Form 8-K, Exhibit 99.1) :
- Vertiv sold $3,274 million of equipment and services in the quarter, up 24 percent from a year earlier. Setting aside acquisitions and currency, the business grew 18 percent on its own .
- That was less than the company itself had guided: in April it had guided second-quarter sales of $3,250 million to $3,450 million, a midpoint of $3,350 million .
- Earnings under generally accepted accounting principles (GAAP) came to $1.27 a share, up 53 percent. The company's adjusted figure was $1.52, up 60 percent . The bridge between them adds back amortization of acquired intangibles ($0.19 a share) and a contingent payment tied to an acquisition ($0.07), less a restructuring credit ($0.01) . The release applies no tax effect to those add-backs, which flatters the adjusted figure slightly; the items themselves are routine.
- Cash generated by the business after capital spending — the company's adjusted free cash flow — was $925 million, up 234 percent from a year earlier .
- Deferred revenue reached $3,634 million, from $1,815 million at the end of 2025 and $1,257 million a year earlier . Deferred revenue is money customers have already paid for equipment Vertiv has not yet delivered. It is the closest thing to a backlog figure the company still publishes.
- Guidance was raised for the second quarter in a row: full-year adjusted earnings of $6.65 to $6.75 a share, up from $6.30 to $6.40 in April .
- On the company's own forecast, the stock went from about 51 times full-year adjusted earnings on 1 June (a close of $323.39 over the $6.35 midpoint then in force) to about 39 times on 31 August ($258.72 over $6.70) .
Above: what Vertiv recognized as sales in each quarter, against the money customers had paid ahead at each quarter's end, second quarter 2025 to second quarter 2026, from the company's five quarterly releases .
One fact about disclosure matters as much as any number. Vertiv's results release of 11 February 2026 still reported orders (up about 252 percent on an organic basis in the fourth quarter) and a backlog of $15.0 billion, up 109 percent from a year earlier . Its first- and second-quarter 2026 releases report neither . The figures above are what remains to judge demand by.
Names that bear on this
Vertiv — disclosed deferred revenue of $3,634 million at 30 June 2026 and a second raise to full-year guidance (8-K, 29 July 2026) . Which way it cuts: demand confirmed; the timing of recognition is the open question. What resolves it: the third-quarter report, with deferred revenue and organic growth.
Powell Industries — reported revenue of $312 million for its fiscal third quarter, up 9 percent, against new orders of $934 million and a backlog of $2.4 billion, up 69 percent, including a data-center order above $400 million for on-site power generation (8-K, 3 August 2026) . Which way it cuts: the same pattern — orders landing, revenue lagging — in a company that still reports backlog. What resolves it: its fiscal fourth-quarter report; the last two came on 19 November 2024 and 18 November 2025 .
No company here is being recommended; see the position disclosure at the end.
The case for: the business was right and the spreadsheet was early
Start with the two numbers that moved the most, because they are the same event seen from two sides. Sales came in short of the company's own midpoint. Cash came in at more than three times the year-earlier level. A company does not generate that much more cash on lighter sales unless customers are paying ahead of delivery — and deferred revenue doubling in six months says exactly that. Orders are real; they are landing in the accounts later than the guidance assumed. That is a timing story, and timing stories resolve on their own.
The second argument is the guidance. A management team that raises its full-year forecast twice in a row, in a quarter where it fell short of its own sales midpoint, is telling you it can see the deliveries it has sold. Vertiv no longer reports orders or backlog in its releases, so this is the signal it has chosen to give instead, and it has not wavered.
The third is the frameThe frame: what we think is going on in the world, written down, dated, and revised when the evidence asks. The House View is the frame in full; this picture is the frame at a glance. itself. Power and cooling sit downstream of the capital budgets of the largest companies in the world, not of how many graphics chips are for sale this month. On 1 July a report, unconfirmed by the company concerned, said one of the largest buyers of computing capacity was preparing to rent its surplus to outsiders, and the shares of two companies that rent out chips fell sharply that day (the House View: AI compute is now two businesses). Our claim is that spare rented capacity moves work around; it does not cancel new power and cooling construction, which follows capital spending, not chip scarcity (Spare computing capacity does not stop new power construction). Vertiv's quarter is that sentence with numbers attached.
AI compute has split into a rented market and a government-funded one.
On this reading, what the stock lost over the summer was not the business. It was the price people were willing to pay for each dollar of it — the thing we are watching across the whole market under The market is paying less for earnings, not doubting the businesses — and a price that falls while the business improves is the discount rate moving, not the thesis failing. The discount rate is the return investors demand for waiting on future earnings.
The case against: a thesis that cannot be proven wrong is not a thesis
The strongest honest objection is about the evidence, not the company. Vertiv stopped reporting the two numbers — orders and backlog — that would settle the question directly. Deferred revenue is a proxy, and a flattering one: it rises when customers pay early, which they may do for reasons that have nothing to do with how much they will order next year. A proxy chosen after the real number disappeared deserves less weight than the real number did.
The second objection is that "timing, not demand" is what every management team says after a sales shortfall. It is sometimes true. It is also unfalsifiable in the quarter it is said. If recognized revenue does not catch up within two reports, the timing story becomes a demand story, and nobody announces the moment it changed.
The third objection is the one the trial got wrong. At about 51 times the company's own forecast for this year's earnings on 1 June, the stock was priced for everything to go right on schedule. A business can be excellent and still be priced for more than it can deliver; the multiple fell by about a quarter over the summer while the business improved, and both of those things are true at once. The trial forgave the valuation in writing. The market did not.
And the fourth objection is the frame's own: if the market has decided to pay less for every dollar of earnings — because the price of money has risen, or because appetite for risk has fallen — then it does not matter how good the quarter was. Being right about the business while wrong about the discount rate is still being wrong about the price.
What we do not know, and the date we will
- Whether recognized revenue catches up with deferred revenue. Vertiv's third-quarter report is the first reading; its investor calendar had not posted the date as of 11 October 2026, and its last three third-quarter reports came on 25 October 2023, 23 October 2024 and 22 October 2025 . The fourth-quarter report is the second reading. Two reports without catch-up and the timing story is over.
- Whether Vertiv's customers keep spending. The capital-spending plans of the largest cloud and platform companies (the hyperscalersHyperscalers: the largest cloud companies (Microsoft, Alphabet, Amazon, Meta); their capital spending is the demand under the AI build-out.) in their next quarterly reports are the demand signal Vertiv no longer publishes itself.
- Whether the summer pattern — beat, raise, fall — repeats across the market in the October-to-November reporting season, or whether companies that beat now hold. That is The market is paying less for earnings, not doubting the businesses's own test, and its review date is 20 November 2026.
- Whether any large data-center project is deferred or canceled citing surplus capacity. We have found none as of this writing. Spare computing capacity does not stop new power construction reviews on 15 November 2026.
What this does to the frame
It sharpens Spare computing capacity does not stop new power construction and leaves The market is paying less for earnings, not doubting the businesses standing. The quarter confirmed that power-and-cooling demand runs on committed capital, not on chip scarcity: deferred revenue doubled in the first half of the year, and the company raised its guidance again on 29 July, after the 1 July report that one of the largest buyers of computing capacity was preparing to rent out its surplus (the House View: AI compute is now two businesses). That is the watchpoint's claim, with a number on it. On The market is paying less for earnings, not doubting the businesses, the quarter is one data point in the pattern, not a resolution; the resolution comes from the next cycle.
What changes is not in the House View. It is in how this publication writes. Three rules came out of the trial and are now standing:
- A claim that depends on price states the price above which it no longer holds. The trial forgave a valuation that was already stretched at the start, and that was the signal that mattered.
- A claim that depends on a company's voluntary disclosure names its fallback at the time of writing. Vertiv's orders and backlog disappeared under the claim; deferred revenue and organic guidance are the proxies now written down in Spare computing capacity does not stop new power construction.
- A sales shortfall alongside a cash beat is a timing signal, not a disconfirmer, and what would prove the claim wrong is written down in advance — so the post-mortem does not have to argue it.
The orders were real. The price was the mistake. Both go on the record.
How this piece was made
Sources. 9 numbered sources, in the Record.
Corrected. 13 Oct 2026: the note sits at the top of this piece. The corrections log →
Related claims
Tell the publisher
Something we should look at? Suggest a company, a topic or a feature.
Record
Written according to the Grant Bishop standard — machine-researched, machine-checked, human-judged; then read and committed by the publisher. Every number traces to a filing or a dated market record; every claim carries a date and what would prove it wrong.
- Drafted by
- Claude Fable 5.1 under Grant Bishop's methodology
- Verification
- multi-source verified, machine-drafted, human-gated
- As of
- 2026-10-11
- Published
- Oct 12, 2026, 11:39 ET · 2026-10-12 15:39 UTC · changelog
- Bears on
- What would change this
- recognized revenue failing to catch up with deferred revenue over the next two quarterly reports; or the hyperscalers cutting their capital-spending plans in their next quarterly reports; or a data-center project of 100 megawatts or more deferred citing surplus capacity.
- Position disclosure
- The publisher holds index-level U.S. equity exposure and no direct position in any single name discussed.
- Not advice
- General commentary for a general audience. Not advice for your situation.