CoreWeave — The Thin Floor

The Thesis in One Security. Its Revenue Is More Contracted Than Its Peers' — but Its Floor Covers Only a Seventh of the Price, and Everything Above That Is a Levered Bet on a GPU's Economic Life.

TON618 Capital Research · As of July 16, 2026 · Download PDF

A valuation, not a trade. Fair value is an estimate with unusually wide bands here, not a price target. Data: CoreWeave's FY2025 10-K, Q1-2026 10-Q, S-1, and earnings releases (SEC EDGAR, CIK 0001769628), and market data as of the July 16, 2026 close ($72.91, Schwab). Fact ("Data shows…") is separated from opinion ("We believe…"). Figures are DISCLOSED (in a filing), REPORTED (release/press), CONSENSUS (sell-side), or ESTIMATED (ours). This note is one of two single-name tests designed by our thematic note The Neocloud Fuse (July 16, 2026); a companion valuation of Oracle publishes alongside it, and §10 sets the pair side by side. The companion The Depreciation Question (July 2026) supplies the useful-life analysis §9 leans on. Independently fact-checked before release; the Source & Verification Note documents that pass and what we could not verify.

§0 Verdict

What this note is, and is not. It is a valuation of the equity, built to answer the question our thematic note posed: when an AI-infrastructure buildout is financed with debt, what is actually left for the shareholder? The finding is not that CoreWeave is a fraud or a fantasy — its revenue is among the most thoroughly contracted of any fast-growing company's — but that at $72.91 you are paying almost entirely for growth whose value swings nearly sevenfold on the one assumption nobody can pin down. (A companion valuation of Oracle publishes alongside this note and runs the same test on an issuer with the opposite architecture; §10 sets the two side by side.)

§1 The Story in Plain English

Read this if you read nothing else.

CoreWeave rents out AI computers. It buys Nvidia GPUs by the hundred-thousand, houses them in data centers, and rents them to companies that need to train and run AI — above all Microsoft, which was 67% of its revenue last year. It funds the GPUs mostly with debt secured against the rental contracts, which is why it now carries about $25 billion of it. It went public in March 2025 at $40 a share, ran to $187 by that June on AI euphoria, and has since fallen to $72.91 — still nearly double the IPO price, but 61% off the top.

Here is the unusual thing about CoreWeave, and the reason it is worth valuing carefully. Its revenue is more locked-down than most fast-growing companies'. Customers sign multi-year take-or-pay contracts — they pay whether or not they use the capacity — and prepay 15–25% up front. So the money coming in is genuinely contracted. That is the opposite of the worry we raised about this whole sector in our thematic note: for CoreWeave, demand is not the problem.

The problem is what is left for the shareholder after the bills. And the bills are enormous. A GPU depreciates fast — Nvidia ships a much better chip roughly every year — so the accounting charge for wearing out the hardware eats up nearly half of every dollar of revenue. Then there is the interest on the $25 billion of debt. Put those two together and CoreWeave's celebrated "60% profit margin" (a figure that conveniently ignores both) collapses to about 1% once depreciation is counted. The company is growing explosively and burning cash — over $20 billion of negative free cash flow expected this year — funded by yet more debt and stock sales.

So what is the equity actually worth? Start with the safest part: the contracted book. If CoreWeave signed no new business and simply collected on the contracts it already has, paid its debts, and wound down, the shareholders' slice would be worth only about $10 a share — because the margins, after depreciation and interest, are thin, and the debt is large. That $10 is the floor. It is just 14% of the $72.91 price. Everything else — the other $63 — is a bet that CoreWeave keeps signing profitable new business, faster than its GPUs go obsolete, without the debt catching up with it.

That bet might pay off handsomely: much of the growth is already contracted (a new $21 billion Meta deal, Anthropic, OpenAI, an Nvidia partnership to build five gigawatts of capacity by 2030). Or the treadmill and the leverage might grind the equity down. Our central estimate is about $60 a share — modestly below the price — but the honest truth is a range from $17 to $114. CoreWeave is not a stock you value to a point. It is a levered option on how long an AI chip stays useful.

§2 One Business, One Bet

A pure play with no second engine — whatever floor exists must come from the contracts already signed.

Most large companies that make an enormous bet carry it on top of something older and safer — a legacy franchise that keeps earning whether or not the bet works, and that gives the equity a floor an analyst can value without believing anything about the bet. CoreWeave is not built that way. It is a single business: buy GPUs, house them, power them, rent them out. There is no installed base of sticky enterprise software underneath, no services annuity, no dividend. There is a three-year-old company growing at a rate almost without precedent, financed to the hilt — and nothing cushions the equity except the contracts the business has already signed.

That purity is exactly why our thematic note nominated CoreWeave as its test case, and it cuts two ways:

• **On the demand side, CoreWeave is safer than the thematic note feared.** Our thematic note worried that AI compute contracts are "ceilings" written by circular, cash-burning counterparties. For CoreWeave that worry is least applicable: its anchor is Microsoft (67%), not a lab; its contracts are take-or-pay with cash prepayments; OpenAI is not yet even a 10% customer. The revenue is real and contracted. This partly falsifies the thematic note's surface claim — and we say so plainly.
• On the economics side, CoreWeave is the thesis in its rawest form. The thematic note's deeper point was that the market lends against these assets cheaply (CoreWeave's own secured, non-recourse facility prices at Treasuries + 200bp) and against the company dearly (its unsecured notes at ~Treasuries + 534bp) — a ~334bp wedge — and that the equity is the thin residual behind that wedge. CoreWeave is the security where you can see that residual directly. This confirms the thematic note's core claim, in one ticker.

So we value CoreWeave the way its structure demands: the contracted book as a floor (§3), the margin that determines everything (§4), and the growth option that is most of the price (§5) — then weight three futures (§6) and check the answer against a DCF that shares none of the scenario model's assumptions (§7).

§3 The Contracted Floor: A Thin, Wasting Annuity

What CoreWeave is worth if it stops growing — the honest downside anchor.

CoreWeave discloses $98.8 billion of remaining performance obligations (RPO) as of March 31, 2026 — contracted, largely take-or-pay revenue it has not yet recognized, with a weighted duration of roughly five years and a recognition schedule (for the $98.8B figure) of ~36% within 24 months, ~39% in months 25–48, and ~25% thereafter. Valuing its run-off is the fairest way to establish a floor — what the equity is worth on the business already signed, before any belief about growth.

• The run-off math. Take the $98.8B of contracted revenue across its ~7-year tail, apply CoreWeave's ~56% adjusted-EBITDA margin (the take-or-pay contracts genuinely support it), and discount at 11% — a lower rate than the growth business because this cash is contracted. That is ~$42B of present value. Subtract the ~$9.6B of construction-in-progress capex still required to bring the contracted capacity online, the interest on the debt as it amortizes against these contracts, and the $22.6B of net debt that sits senior to the equity. What remains for shareholders is roughly $6 billion — about $10 a share. [ESTIMATED from DISCLOSED RPO, margin, capex, and debt]
• Why the floor is thin. The contracts are real, but three things hollow out the equity's share of them: the margin after depreciation is modest, the capex to complete the build is large, and — decisively — $25B of debt gets paid before the shareholder does. A take-or-pay contract is a wonderful thing for a lender; for the residual equity holder behind $25B of that lender's claims, it is worth much less.
• The number in context. Ten dollars of a $72.91 price. An investor who turns out to be wrong about everything CoreWeave does from today forward still owns the contracted book — but the contracted book, after the debt is paid, covers about a seventh of what they paid. Every other dollar depends on business that does not yet exist.
• What the floor is not. It is not a liquidation value (GPUs and data centers would fetch something in a wind-down, though a fire-sale of two-generation-old GPUs is not where you want to be), and it is not a floor the market will necessarily respect in a panic. It is the going-concern value of the contracted book, and it says: strip out future growth, and only ~$10 of the $72.91 is accounted for.

§4 The Margin Is the Whole Argument

CoreWeave's "60% margin" and its ~1% margin are the same company. The gap is the thesis.

CoreWeave reports an adjusted EBITDA margin near 56–60%. It also reports a GAAP net margin of −23% (FY25) and −36% (Q1-26). Both are true, and the distance between them is the entire valuation debate.

• What the add-backs remove. Adjusted EBITDA adds back, each quarter, roughly $1.15B of depreciation and $0.54B of interest. That is how a −36% GAAP net margin becomes a 56% "adjusted" one. But depreciation is the cost of the GPUs wearing out (or going obsolete), and interest is the cost of the debt that bought them — for a company whose entire business is buying depreciating hardware with borrowed money, these are not "adjustments." They are the business.
• The number management does not headline. CoreWeave discloses an **adjusted operating income figure — which adds back stock comp but not depreciation. It fell from 17% of revenue (Q1-25) to ~1% (Q1-26) as the depreciation on its growing GPU fleet caught up with it. FY2026 guidance is only $900M–1.1B of adjusted operating income on $12–13B of revenue — a ~7–8% margin.** [DISCLOSED / REPORTED] That ~1–8% is far closer to the economic truth than the 56%.
• The crux, as a table. The steady-state net margin — what CoreWeave earns after real depreciation and interest, once revenue catches up to the installed base — depends almost entirely on the economic life of a GPU:

• Where the truth probably sits, and the nuance that matters. CoreWeave depreciates its GPUs over six years (§9). Many argue an AI GPU's economic life is three to four, as Nvidia's annual cadence pushes prior generations to cheaper inference work at lower rental rates. But — and this is the point that keeps CoreWeave from being a simple short — its revenue is contracted take-or-pay for ~5 years. For the contracted book, the relevant life is the contract term, not the obsolescence term: Microsoft pays the agreed rate whether or not the H100s are still cutting-edge. So the depreciation-treadmill risk is largely a renewal and residual-value risk at year 5+, and a risk on the un-contracted growth capex — not an in-contract risk. That is why our base case sits at a 4.5-year effective life (~14% margin), not the 3.5-year bear. The treadmill bites hardest on the growth CoreWeave has not yet contracted — which, per §5, is most of what you are paying for.

§5 What You Are Paying For: The Growth Option

Decompose the price, and 86% of it is a bet on business CoreWeave has not yet earned.

The cleanest way to see CoreWeave is to split the $72.91 price into what is signed and what is hoped for:

• Contracted floor: ~$10 (14% of price) — the run-off value of the book it already has (§3).
• Growth option: ~$63 (86% of price) — everything else. The value of CoreWeave signing new profitable business, faster than its assets depreciate, without the leverage catching up.

The floor is a seventh of the price; the option is six-sevenths. That is the single most important sentence in this note. It means CoreWeave is not really being valued as a business with a growth kicker; it is being valued as a growth option with a thin book of collateral attached.

• Why the option could be worth it. The growth is not blue-sky. Much of it is already contracted forward: a new ~$21B Meta commitment (April 2026), a multi-year Anthropic agreement, the OpenAI MSAs, and an Nvidia partnership to build >5 GW of "AI factories" by 2030 (CoreWeave has >3.5 GW contracted, >1 GW active today). And the market plainly will finance the assets — its secured, non-recourse facility prices at Treasuries + 200bp. A bull can reasonably argue the option is closer to its intrinsic value than to zero.
• Why the option is dangerous. Every dollar of that growth is financed at 7–15% and deployed into assets that lose value fast, with the equity standing last in line behind $25B of debt. The market pays ~$63 for the growth option; we value that same option at roughly $50 (our $60 central estimate less the $10 floor) — so we think it is worth having, just not at the price. And because it is levered, the distribution around that $50 is violent: it is worth ~$100 in the bull and roughly nothing in the bear (§6).
• The market-implied read. At $72.91, the market is paying roughly the base-case value for the growth option. That is not an unreasonable price for the option on its own terms — but it means the price carries no discount for the chance the base case is wrong, and beneath it sits only a 14%-of-price floor if it is.

§6 Three Futures for CoreWeave

The scenarios, and why the distribution is so wide.

We value CoreWeave on its normalized earnings power around FY2028 (management targets a >$30B revenue run-rate exiting 2027), discounted back at a high rate for a levered, volatile equity. Each scenario turns on the economic GPU life and the revenue the growth delivers. (Full arithmetic in the accompanying models; every input traces to the filings.)

• Bull — $114 a share (28% weight). The treadmill is mild (effective ~5-year life), renewals hold at good rates, and revenue compounds to ~$34B by FY2028 as the contracted pipeline (Meta, Anthropic, OpenAI, Nvidia) converts. Net margin reaches ~17%; the market rewards a profitable hyper-grower with a mid-teens multiple. This is a genuinely good outcome and it is not a fantasy — most of the revenue is already signed.
• Base — $64 a share (44% weight). The effective life runs below book (~4.5 years), margins settle thin-but-positive (~14%), revenue reaches ~$30B. The equity is worth roughly today's price, less a little. This is the most likely single outcome, and it is why the stock is not obviously mispriced to a casual eye.
• Bear — $17 a share (28% weight). The treadmill bites (effective ~3.5-year life), renewal pricing compresses as older GPUs glut the market, the margin thins to ~7%, and — critically — the leverage turns a disappointing operating result into an equity wipeout. With $7.5B of debt due inside twelve months and net debt at 7× EBITDA, a growth stumble forces dilutive equity raises or distressed refinancing. The equity does not fall 20%; it falls ~77%. The left tail is fat by arithmetic, not by mood.

Why the tilt is down, argued from the data (per our house standard, a skew must be evidenced). The base case is roughly the price, so this is not a dramatic call. The modest downward tilt comes from three disclosed facts, not from sentiment: (1) the contracted floor covers only 14% of the price, so there is minimal downside protection; (2) the base and bull require the effective GPU life to hold at 4.5–5 years, above the 3–4 years our Depreciation Question note argues for the growth capex; and (3) the leverage — $25B of debt, $7.5B current, 7× EBITDA — makes the bear a near-wipeout rather than a drawdown, which pulls the probability-weighted mean down even at unchanged odds. The counterweights are real and keep the tilt modest: contracted take-or-pay revenue, a Microsoft anchor, a large signed pipeline, and demonstrated cheap access to asset-level debt. The honest landing is a central estimate a notch below the price, wrapped in the widest distribution we have published.

§7 Cross-Checks: A DCF and the Comps

One method is a guess. Three are a triangulation — and the spread here is the message.

The scenario model (§6) is one lens. We ran a consolidated DCF and the peer multiples independently, so the conclusion has to survive methods that do not share the scenario model's assumptions.

• The consolidated DCF says ~$54 — and the reason it sits below the earnings method is that it charges the cash burn in full. An unlevered free-cash-flow model on CoreWeave's growth path (revenue to ~$53B by FY2032, adjusted-EBITDA margin ~54%, capex tapering from $33B toward replacement) discounted at a 12% WACC lands at ~$54 a share. It is below the scenario model's $65 because the DCF pays, up front, for the tens of billions of negative free cash flow CoreWeave burns in FY2026–FY2027 before the build inflects. The two methods bracket a central estimate near $60.
• The DCF is violently sensitive — which is itself the finding. At an 11% WACC it gives $80; at 13%, $34. At 3% terminal growth, $41; at 5%, $71. A valuation whose answer moves by more than 2× on ordinary changes in the discount rate is telling you the truth: almost all of CoreWeave's value is terminal — it is a bet on a distant, uncertain steady state, which is exactly what "86% of the price is a growth option" means, restated in DCF form.
• The multiples say "reasonable" — but only the lenses that ignore what ails it. At $72.91 CoreWeave trades ~14× FY26 run-rate adjusted EBITDA and ~5× revenue — unremarkable for a company growing revenue ~2.4× a year. But it has no GAAP earnings, so there is no P/E to cite, and EV/EBITDA and EV/Revenue flatter CoreWeave precisely because they exclude the depreciation treadmill and the $25B of debt — the two facts that define the equity's risk. On the metric that would matter — earnings, once they exist — the price already embeds the base case. We give the multiples no weight.
• Where it lands. Scenario $65, DCF $54, central ~$60 (−18%); a contracted floor of $10 beneath it; and a scenario range from $17 to $114. Every method agrees on the sign — no discount to fair value — and the DCF insists the tilt is a little more negative than the earnings method alone. But the dominant output is not the point estimate. It is the width, and the thinness of the floor under it.

§8 The Balance Sheet Is the Fat Left Tail

Leverage is why the bear case is −77% and not −20%.

CoreWeave's balance sheet is not one risk factor among many — it is the risk, the reason the downside scenario is an equity wipeout rather than a drawdown.

• The debt load. Total debt reached $24.9B at March 31, 2026 (net debt ~$22.6B), up $3.5B in a single quarter. Net debt is roughly 7× adjusted EBITDA — investment-grade software companies run 1–2×. The equity market cap (~$40B) is only ~1.75× the net debt; the enterprise is more debt-financed than most public equities an investor will encounter.
• The refinancing wall. $7.5 billion of debt is current — due within twelve months — and $6.1B of principal falls due in the remainder of 2026 alone, against a company burning free cash flow. CoreWeave must roll this over in size, repeatedly. It has done so (the $8.5B DDTL 4.0 facility, the $4B convertible, the unsecured notes, equity from Nvidia and Jane Street) — and the thematic note's own finding is that the assets finance cheaply. But a market that will fund the collateral at Treasuries+200bp is not the same as one that will fund the company on demand; the unsecured notes cost ~9.6%, and in a stress the window narrows exactly when the wall arrives.
• The dilution vector. CoreWeave funds its build with debt and equity — the $2B Nvidia placement, the $1B Jane Street placement at $109, the $4B convertible. Every equity raise to plug the FCF gap dilutes the per-share value, and our bear case assumes the share count grows toward ~680M as the company issues into weakness. Leverage plus dilution is why $17, not $50, is the bear.
• The concentration. Microsoft is 67% of revenue. That is a strong counterparty and a genuine de-risker of demand — but it is also a single point of failure. Any renegotiation, build-out slippage, or shift in Microsoft's own capex plans (it has publicly trimmed data-center leases elsewhere) would hit CoreWeave harder than a diversified peer.
• The litigation footnote. A securities class action (Masaitis v. CoreWeave, D.N.J., filed January 12, 2026) and consolidated derivative suits are pending, with no accrual taken. Not thesis-defining, but on the risk ledger.

The balance sheet does not doom CoreWeave — the contracted book and the cheap asset-level financing are real mitigants. But it removes any margin for error, and it is the mechanism that converts a merely-disappointing operating outcome into the −77% bear.

§9 The Depreciation Treadmill

Why "adjusted EBITDA" flatters, and why this ties to our companion note.

Our companion note The Depreciation Question (July 2026) argues that AI-infrastructure builders' six-year server lives overstate current earnings, and it flags CoreWeave as its most extreme case — the one company where charging depreciation on a realistic economic life flips the sign of the result.

• The policy. CoreWeave depreciates computing equipment (GPUs) over six years, straight-line, a life it extended from five to six effective January 2023 — and has not shortened despite Nvidia's Blackwell/GB200/GB300 cadence. [DISCLOSED: FY25 10-K]
• The gap. A six-year book life against a widely-argued three-to-four-year economic life understates depreciation. On CoreWeave's ~$21B gross technology-equipment base, moving to a four-year life would add on the order of $1B+ a year of non-cash charge — and that base is growing fast. Because CoreWeave is barely profitable at the operating line already, a more aggressive (more realistic) depreciation schedule pushes the reported result convincingly negative.
• The ramp still coming. CoreWeave carries ~$9.6B of construction-in-progress — ~23% of gross PP&E — not yet being depreciated (as of March 31, 2026). As that capacity is placed in service, the depreciation line climbs further, precisely as the bull case needs margins to expand.
• The saving grace, restated. As in §4, the take-or-pay contracts blunt this for the contracted book — revenue is locked over the ~5-year term regardless of obsolescence. So the treadmill's real damage is to residual values, renewal rates, and the returns on un-contracted growth capex. That is the honest scope of the risk: not that the current book is worthless, but that the growth — 86% of the price — earns less than it appears once the hardware's true useful life is charged against it.

§10 The Companion Test: CoreWeave Beside Oracle

Published together with our Oracle valuation. One comparison, in one place — the pair was designed as a single experiment.

Our thematic note designed two single-name tests of the same question — when an AI-infrastructure backlog is financed with debt, what is left for the shareholder? — and deliberately chose issuers with opposite architectures. Oracle carries its AI bet on top of a mature software franchise; CoreWeave is the bet. Read together, the two valuations say something neither says alone:

The trade-off runs in opposite directions. Oracle has the weaker demand story (an undisclosed counterparty behind roughly half its backlog, contracts that are ceilings) but the thicker cushion — a majority of its price is covered by a business that needs no AI thesis at all. CoreWeave has the stronger demand story (contracted, prepaid, anchored by a hyperscaler) but the thinner cushion — a seventh of the price, behind $25B of debt. An investor in Oracle is mostly exposed to whether the AI revenue is real; an investor in CoreWeave is mostly exposed to whether the AI economics are good. Both prices sit near their respective base cases; neither offers a discount. The pair, taken together, is the thematic note's conclusion made concrete: the market has priced these securities for the base case of a buildout whose load-bearing numbers — one counterparty's solvency, one chip's useful life — remain undisclosed.

§11 What Would Change Our View

The falsifiers, stated so the reader can hold us to them.

• Renewal economics. The single most valuable datum would be the rate at which a first-generation contract renews as the GPUs age. Renewals at or near original rates validate the ~5-year effective life and the bull; renewals at steep discounts (or non-renewal) confirm the treadmill and the bear.
• A path to positive free cash flow. CoreWeave has not guided to a positive-FCF year. A credible crossover date — with capex tapering as the contracted build completes — would materially de-risk the leverage and lift the base toward the bull.
• The economic-life disclosure. Any move by CoreWeave (or its auditors) to shorten the GPU life toward four years would confirm our treadmill concern and pressure the reported margins the bull relies on.
• Refinancing terms. How the $7.5B current maturity is rolled — at what spread, secured vs unsecured, with how much equity — is a live, near-term test of the balance-sheet risk in §8. Cheap, secured rollover argues the market still backs the assets; expensive or equity-heavy rollover confirms the fat tail.
• Microsoft. Any expansion or contraction of the 67% relationship moves the stock more than any other single event.
• What would confirm us. Continued adjusted-operating-margin compression as depreciation catches up; the FCF burn widening rather than narrowing; equity issuance to fund the build; the short interest (already rising, ~81M shares) proving prescient.

§12 House View

The call. CoreWeave is worth a central estimate of ~$60 against a $72.91 price — roughly 18% downside — but the central estimate is the least important number in this note. The important ones are that the contracted floor is only ~$10 (14% of the price), that ~86% of the price is a levered growth option, and that the honest fair-value distribution runs $17 to $114. Two methods triangulate the center — normalized earnings power $65, a consolidated DCF $54 — and both, with the thin floor beneath them, say the same thing: CoreWeave is priced for its base case with almost no protection if the base case is wrong — modestly rich, on a thin floor. Confidence: low-to-moderate — the lowest in this series, because the entire equity value above ~$10 rests on two things we cannot pin down: the economic life of a GPU and CoreWeave's ability to keep refinancing $25B of debt.

Why the skew is down, and modest (per our standard, argued not asserted). The tilt down is earned by three disclosed facts — the 14%-of-price floor, the effective-life assumption sitting above the Depreciation Question range, and the leverage that makes the bear a wipeout. It is kept modest by three equally real facts — take-or-pay contracted revenue, a large already-signed growth pipeline, and demonstrated cheap access to asset-level debt. The honest landing is a notch below the price, inside a very wide band.

The steelman against us. The bulls have a strong case, and it is not the "wrong-lens multiples." It is this: CoreWeave's revenue is genuinely contracted, its growth is genuinely pre-signed (Meta, Anthropic, OpenAI, Nvidia), the market genuinely finances its assets cheaply, and if GPUs prove to have five-to-six-year useful lives — plausible if inference demand keeps older chips utilized and power scarcity keeps all capacity full — then the treadmill never bites, the margin holds near 17%, and $114 is the honest number, not $60. We hold our more cautious line because the equity sits behind $25B of debt on a 14%-of-price floor, because the effective life the price requires is above what the hardware cadence suggests, and because a business this levered should be valued with the fat left tail fully weighted. But of the three notes in this series, this is the one where we hold our view with the least conviction — by design, because CoreWeave is the security where the least is knowable.

What the analysis implies. Three understandings fall out of the work. First, CoreWeave is the thesis in one security — the purest, most levered expression of neocloud economics, with the demand risk lowest and the capital-structure risk highest. Second, the floor is thin — strip out future growth and only ~14% of the price is accounted for. Third, the equity is a levered option on a GPU's economic life — a number the company will not disclose and the industry cannot agree on — which is why we publish a range from $17 to $114 and treat any single figure, ours included, as the midpoint of a genuinely wide distribution.

§13 Disclosures

Information only. This material is published by TON618 Capital for information purposes only. Nothing here is an offer to sell or a solicitation of an offer to buy any security, fund interest, or digital asset, and nothing here is personalized investment advice or a recommendation regarding any instrument.

Publisher's exclusion. All research is published solely as general, impersonal information of regular circulation. It is not tailored to the objectives or circumstances of any individual and is not issued in connection with compensation from any client. The Fund has no clients and distributes all research free of charge. On that basis it publishes in reliance on the publisher's exclusion from the definition of "investment adviser" under the Investment Advisers Act of 1940 (§202(a)(11)(D); cf. Lowe v. SEC, 472 U.S. 181 (1985)).

Registration & conflicts. TON618 Capital is not registered as an investment adviser or broker-dealer in any capacity. The Fund is a Bitcoin fund and may hold or transact in the securities or digital assets it discusses; material conflicts are disclosed where they exist. As of the date of this note the Fund holds no position in CoreWeave or in any other single security named herein; it maintains a strategic Bitcoin allocation. The Fund receives no compensation from any party in connection with this research.

Use of AI. Artificial intelligence is used in the creation of this research. A material conflict is disclosed here: this note values a company whose customers and partners include the maker of the AI tooling used to produce it. All methodology and data integrity are reviewed and approved before publication by TON618 Capital's Chief Investment Officer, Keyth Beck; errors may nonetheless occur, and readers should verify independently.

CFA. This report was prepared to align with CFA Institute analytical standards (methodology only). CFA® and Chartered Financial Analyst® are registered trademarks owned by CFA Institute. That reference describes the analytical framework applied; it does not imply the report was prepared, reviewed, or authored by a CFA charterholder, and the report is not issued, reviewed, endorsed, certified, or approved by — nor affiliated with — CFA Institute.

Risk & feedback. Past performance is not indicative of future results. Equities are volatile and may result in total loss of capital; a highly-leveraged, pre-profit equity such as this one carries elevated risk of permanent loss. A fair-value estimate is not a price target and carries wide uncertainty. Corrections and feedback are welcome — please direct them to CIO Keyth Beck at keyth@ton618capital.com.

Source & Verification Note

This note values CoreWeave from primary filings pulled directly from SEC EDGAR (CIK 0001769628): the FY2025 10-K (accession 0001769628-26-000104), the Q1-2026 10-Q (0001769628-26-000222), and the Q1-2026 earnings release (8-K ex-99.1, 0001769628-26-000220), with market and short-interest data at the July 16, 2026 close. All three valuation methods — the normalized earnings-power scenarios, the consolidated DCF, and the contracted-book floor — are reproducible in the two accompanying models; every input traces to a cited figure.

High-confidence, filing-verified (DISCLOSED). Revenue FY23/24/25 $229M / $1,915M / $5,131M and Q1-26 $2,078M; gross margin 71.7% FY25 falling to 65.5% Q1-26; operating loss $(46)M FY25 / $(144)M Q1-26; net loss $(1,167)M FY25 / $(740)M Q1-26; D&A $2,454M FY25 / $1,147M Q1-26 (48%→55% of revenue); interest expense $1,229M FY25; the adjusted-EBITDA bridge ($3,093M FY25, 60% margin) and the adjusted-operating-income collapse (17%→~1% of revenue YoY); RPO $60.7B (12/31/25) and $98.8B (3/31/26), the latter with a 36%/39%/25% recognition schedule (≤24mo / 25–48mo / thereafter); total debt $24.9B and net debt ~$22.6B, current portion $7.5B, $6.1B due in the remainder of 2026; capex $10.3B FY25 and $7.7B Q1-26, FCF ~$(7.3)B FY25 and $(4.7)B Q1-26; the six-year GPU life (extended from five in 2023) and the ~$9.6B of construction-in-progress (3/31/26); customer concentration (Microsoft 67% of FY25 revenue, no other ≥10%); the dual-class share structure and ~545.6M basic shares; and the pending Masaitis and derivative litigation with no accrual.

Deliberately hedged, not asserted as fact (REPORTED / CONSENSUS / ESTIMATED). FY2026 guidance (revenue $12–13B, adjusted operating income $900M–1.1B, capex $31–35B) and the ">$30B run-rate exiting 2027" target are management guidance. The named growth commitments (Meta ~$21B announced April 9 2026, Anthropic, OpenAI ~$11.9B + up to $6.5B — with an intervening ~$4B expansion, so ~$22.4B disclosed in aggregate — Nvidia >5 GW by 2030) are REPORTED from releases; the OpenAI amounts are contractual ceilings ("up to"), consistent with the thematic note, and OpenAI is not yet a 10% customer. The ~2.5-year "payback" is management's claim and is explicitly an adjusted-EBITDA payback — not a profit payback — and we treat it as such. Sell-side consensus and the ~$132 median price target are stale (they reference a ~$117 last price, predating the drop to $72.91) and are not inputs to our fair value. Short interest (~81M shares, rising) is from Massive Market Data.

The softest inputs, named plainly. First and above all, the economic life of a GPU — the single number that swings the steady-state margin from ~21% (six years) to ~1% (three years), and with it the fair value from $114 to $17. It is undisclosed, industry opinion is divided, and it is the load-bearing assumption of the entire note. Second, the contracted-book floor (~$10) is our estimate from the disclosed RPO, margin, remaining capex, and debt — not a disclosed figure — and it moves with the discount rate and the assumed run-off margin. Third, the growth/replacement capex split is not disclosed; CoreWeave presents nearly all capex as growth, so our normalization of "maintenance" capex is a judgment. Fourth, the fully-diluted share count (RSUs, options, the $4B convertible) is higher than the ~545.6M basic we anchor on; we use ~560M in the DCF and up to ~680M in the bear to reflect dilution, but the exact diluted figure was not extracted from the proxy.

Method notes. Fair value triangulates three methods, all reproducible. (1) Normalized earnings power (~$65): FY2028 revenue by scenario ($24–34B) × a steady-state net margin derived from the economic GPU life (7–17%) × a 10–16× P/E, discounted 2.5 years at 13–16% and divided by a diluted, growing share count. (2) Consolidated unlevered DCF (~$54): a seven-year FCFF build (revenue to ~$53B by FY2032, adjusted-EBITDA margin ~54%, D&A normalizing from ~50% toward ~30% of revenue, capex tapering from $33B toward replacement) at a 12% WACC and 4% terminal growth; highly sensitive to both ($34 at 13% WACC to $80 at 11%), reflecting that nearly all value is terminal. (3) Contracted-book floor (~$10): the $98.8B RPO run off at its ~56% margin and 11% discount, on the disclosed 36/39/25 recognition schedule, less remaining CIP capex and net debt. The central estimate (~$60) is the mean of methods (1) and (2); the floor is a downside anchor, and EV/EBITDA and EV/Revenue comps are shown but given no weight because they exclude the depreciation and leverage that define the risk.

Companion notes. The Neocloud Fuse (July 16, 2026) is the thematic parent that designed this test and its Oracle counterpart; The Depreciation Question (July 2026) supplies the useful-life analysis behind §9 and flags CoreWeave as its most extreme case. Read alongside the Oracle note, this pair is the intended point: the same AI-infrastructure bet, valued once with a thick floor and once with a thin one.

Version 1.0 · analyst: TON618 Equity Research. Fair value ≈ $60 vs $72.91 (scenario $65 / DCF $54; floor $10; range $17–114); modestly rich on a thin floor, risks skewed down, distribution unusually wide.