The disclosure premium — what a contract book is worth in 2026

Companion article to Issue 23 // 14 August 2026 // R. Lauritsen

The disclosure premium — what a contract book is worth in 2026

Friday's issue argued that the market has started grading disclosure rather than results. This is the part an email couldn't carry: how large the effect is, how much of it survives honest scrutiny, a five-column test for telling a contract book that counts from one that's marketing, and the two ways this ends — which imply opposite trades.

8 min read // Sectors: memory and storage, semiconductor equipment, AI cloud infrastructure, model providers

THE CLAIM, STATED PRECISELY

Here is the version of the thesis I'm willing to defend, which is narrower than the tape suggests.

Sandisk's customers were exactly as committed on 5 August, when the stock fell about 8% on a record quarter, as they were on 13 August, when it rose 13.67%. The contracts existed throughout. What changed in between is that investors could finally see them in a form precise enough to model: $93.9bn of minimum contracted revenue, ten agreements, eight customers, weighted-average life beyond four years, $16.5bn of guarantees, and coverage of more than half of fiscal 2027 bit shipments.

So the effect is not that markets have developed a taste for paperwork over economics. It is that contracted demand is economically meaningful and was previously unobservable — and the repricing lands at the moment of observability, not the moment of commitment. That distinction sounds academic and was worth roughly thirteen points. It also carries a practical consequence: the gain accrues to whoever owns the stock before the disclosure, not after it. Which is why the interesting names below are the ones that haven't published yet.

HOW BIG IS IT — AND HOW MUCH DO WE ACTUALLY KNOW?

Three observations, tagged by how much causal weight each can bear. They are not equivalent, and Friday's issue treated them as though they were.

[MEASURED — same company, eight days] Sandisk's fiscal Q4, reported 5 August: revenue $8.97bn, up 372%; adjusted EPS $39.25 against a $34.52 estimate; a record 84.6% gross margin. The stock fell about 8%. Its Investor Day, 13 August: the contract disclosure above, plus a fiscal 2028–2030 model targeting mid-to-high-teens revenue growth, roughly 80% non-GAAP gross margin and adjusted free cash flow near half of revenue. The stock rose 17% intraday and closed up 13.67%. Same business, same demand curve, eight trading days. This is the only genuine control in the set.

[OBSERVED — cross-company] Applied Materials reported the same evening: revenue $9.12bn against $8.99bn expected, adjusted EPS $3.50 against $3.39, then guided October revenue to $10.25bn ± $500m against a $9.55bn consensus, describing its demand visibility as unprecedented and citing customer conversations running to 2030. It fell 2.48%, then five to six per cent more on Friday.

[OBSERVED — cross-company] CoreWeave, 11 August: revenue $2.58bn, up 112%, net loss widening to $626m on $640m of interest expense — and a $104.2bn backlog, up 246%, with contracts beyond 48 months rising from 10% to 21% of the mix. The stock rose about 21% on Wednesday.

[INFERRED] Across those three, the observed reaction spread between publishers and non-publishers ran roughly twelve to twenty-two points depending on the pairing. That is a description of what happened. It is not a measurement of a disclosure premium, and the difference between those two sentences is the difference between analysis and advocacy.

THE FIVE-COLUMN TEST

Not every backlog number moves a stock, and the ones that don't are more instructive than the ones that do. Five properties separate a disclosure the market can model from one it discounts. This is the reusable part of the piece — it works outside semiconductors and outside this cycle.

1. Quantified. A number, not an adjective. "Sold out through 2026" is a characterisation; "$93.9bn across ten agreements" is a disclosure. SK Hynix has been saying the first thing for a year, which is roughly why it still trades as a memory cyclical with a good story attached.

2. Contracted. Guidance is management's opinion about the future; a remaining performance obligation is a customer's signature on it. Applied Materials produced the best forecast in semiconductors this week and it was worth minus two and a half points, because a forecast carries the seller's risk and a contract carries the buyer's.

3. Duration. CoreWeave's contracts beyond 48 months went from 10% to 21% of the mix, and the market rewarded that shift as much as the headline size. Length is what converts a backlog from a working-capital fact into a valuation input, because it lets an analyst extend the model past the current cycle.

4. Coverage. What share of future output the book actually covers. Sandisk: more than half of fiscal 2027 bits, roughly two-thirds of fiscal 2028. Without a coverage ratio, a large contract number tells you nothing about how much of the business is still exposed to spot pricing — and the exposed half is precisely what a cyclical de-rating attacks.

5. Counterparty backing. The one almost nobody discloses, and to my eye the most under-discussed number of the week: $16.5bn of cash deposits and financial instruments standing behind $93.9bn of contracts, about 17%. A contract book with no guarantees is a list of intentions from customers who may themselves be funding in the high-yield market. A contract book with deposits behind it is a different asset entirely.

THE LEDGER, SCORED

Ten names run through the five columns. ✓ = disclosed and quantified, ~ = partial or qualitative only, ✗ = absent. Columns are Number, Signed, Duration, Coverage, Backing.

Moves are the week to Thursday 13 August unless marked. MSFT is July's print, included because it set the standard. CRWV's move is Wednesday's regular session; AMAT's second leg is Friday's pre-market and open. Scores count ✓ only, and reflect our reading of public disclosure — another analyst could mark them differently.

The bottom six rows behave exactly as the thesis predicts: nothing scored, nothing paid — including two companies that beat and one that guided seven per cent above consensus. It's the top four that make this interesting, because they don't line up.

Micron scores 4/5 and wasn't paid. Microsoft scored 2/5 and was paid 15.5%. A naive version of this thesis — better disclosure, better return — is falsified by its own ledger, and I'd rather say that than quietly drop the rows. Two readings survive. Either scale substitutes for completeness, and $678bn of remaining performance obligation moves a stock regardless of how few columns it fills. Or disclosure is priced once, at the moment of novelty, and Micron's June filing was absorbed in June, with no Investor Day staged around it to make the market look twice.

Those readings diverge in a way you can act on. If scale is what matters, Micron is fairly priced and there is nothing to do. If novelty is what matters, Micron's contract book is a re-rating waiting for a venue — and Sandisk has just demonstrated that staging the venue is worth thirteen points.

WHAT WOULD BREAK THIS

Two exits, on different clocks, pointing in opposite directions. Conflating them is how the position sizing goes wrong.

A bigger number. Nvidia on 26 August is the only print with the mass to reset the cohort in one session, and it publishes no contract book. If a beat and guide gets bought there, the market has shown that at the top of the stack results still work unaided, and this shrinks to a mid-cap phenomenon. Binary, dated, hedgeable — and the single most likely way to be wrong.

Commoditised disclosure. The subtler exit is arithmetic. Any information edge disappears once it becomes universal, and the incentive to produce one is now roughly thirteen points. If the October prints bring a wave of newly quantified backlogs, publishing stops being a differentiator and the effect compresses towards zero. Note which way that cuts: it is bullish for today's punished side, which gains a disclosure it lacked, and bearish for today's paid side, which loses its scarcity.

The two exits imply opposite trades. The first favours owning the currently punished names into 26 August with defined risk. The second favours owning the publishers now and being ready to rotate out of them in October. They wear the same thesis and they are not compatible — which is the argument for sizing this as a tilt rather than a position, and for distrusting anyone who tells you it's obvious.

THREE DATED SIGNALS

1. The reaction to Nvidia, not the result — 26 August. Watch the grade, not the paper. A beat that gets sold at the top of the stack confirms the effect where it has never been tested; a beat that gets bought means results still clear the bar unaided for the largest company in the world. The same call carries two binaries already on the board from Issues 21 and 22: does the $250bn OpenAI guarantee appear as a disclosed contingent liability, and are reduced-memory Rubin Ultra configurations confirmed or denied?

2. Micron's fiscal Q4 — late September. The closest thing to a clean experiment, for the reason set out above: Micron's contract book is already public, so quantifying it again carries little new economic content. If the stock moves anyway, the effect is genuinely about disclosure and the framework holds. If it doesn't, Sandisk's repricing was mostly the NAND sizing and the capital-returns commitment, and I was reading the wrong variable. This is also where last week's drift stat — the hardware cohort's −13.4% average one-month post-earnings drift — gets marked.

3. The imitation count — the October prints. Our test, defined here so it can be checked: count the S&P 500 technology companies that introduce a quantified contracted-revenue, backlog or RPO metric they have not previously disclosed. Three or more and the effect begins compressing, and the trade rotates from the publishers to the newly publishing. Fewer than three and disclosure stays scarce, which keeps it intact into year-end. Sub-test worth tracking alongside: how many of those disclosures include a counterparty-backing figure. My expectation is close to none.

How we'll score it. All three signals are dated, so this note can be marked like a slate: 26 August, late September, October. Each gets scored in the Friday issue as it lands, and the drift number keeps its standing line in the Scoreboard until it resolves. If the framework fails a test, you'll read it here first, in these terms.

Stay curious — and stay qualified.

— R. Lauritsen

Editor, iPrompt Signals

METHODOLOGY AND SOURCES

Reporting comes from named sources; estimates and inferences are tagged in the text, and the tags are load-bearing — [MEASURED] is used only for the same-company Sandisk comparison, [OBSERVED] for cross-company reactions that cannot support causal claims, and [INFERRED] for our own reasoning. Sandisk's Investor Day figures — $93.9bn total contract value at floor pricing, ten New Business Model agreements across eight customers, $16.5bn of financial guarantees, coverage ratios, the NAND market sizing and the fiscal 2028–2030 financial model — are from the company's 13 August 2026 Investor Day presentation and IR materials, with analyst reads via Goldman Sachs and JPMorgan as reported. Sandisk's fiscal Q4 results (5 August 2026) are from the company release. Applied Materials' fiscal Q3 results and October guidance are from its 13 August 2026 release and earnings call, with market reaction via Seeking Alpha, Benzinga and Yahoo Finance. CoreWeave's Q2 2026 figures are from its 11 August 2026 release and investor presentation, as reported by CNBC and Investing.com. Micron's Strategic Customer Agreement figures are from its fiscal Q3 2026 earnings call and slides (24 June 2026). Microsoft's remaining performance obligation is from its July 2026 results. The five-column test, the ledger scores, the imitation-count threshold of three companies and the twelve-to-twenty-two-point reaction-spread estimate are our own construction, tagged at point of use. Prices are Thursday 13 August closes unless marked; fiscal and calendar periods are distinguished throughout.

Disclaimer: For informational purposes only. Not financial advice. iPrompt Signals is not a registered investment advisor. Conduct your own research and consult a qualified financial professional before making investment decisions.