One buildout. Three different shareholder outcomes.
AI investment can keep growing while some of the companies financing that growth disappoint their shareholders. Our Economic Clock investigation leads to a practical question: once the equipment is bought, the facility is built and the financing is paid, who receives the cash?
NVIDIA sells the systems. CoreWeave finances and operates computing equipment. Applied Digital develops the facilities. They participate in the same expansion, but common shareholders own different claims on its proceeds.
Our October 14 research reaches three qualified conclusions: NVIDIA: HOLD. CoreWeave: SELL, provisional. Applied Digital: SELL, low conviction in the label. The twelve-month probability-weighted targets are $253.51, $65.52 and $19.57, respectively, for October 12, 2027. The reference prices are the October 12, 2026 closes—not today’s entry prices.
The underlying research report was published October 14, 2026, at 4:30 PM EDT, before TSMC’s October 15 third-quarter results. Its forecasts remain a dated pre-event record; subsequent evidence belongs in a dated amendment. The distinction matters when assessing supplier results. Strong demand can support all three businesses without making all three stocks equally attractive.
Growth has to reach the shareholder
The original investigation separated slower spending growth, an outright spending contraction, financing impairment and genuine excess useful computing capacity. Those outcomes are not interchangeable. A financing problem at one company does not establish an industry-wide compute glut.
For investors, the next step is to follow the cash through each business. NVIDIA’s earnings must convert into cash. CoreWeave’s equipment must earn enough over its working life. Applied Digital’s projects must leave sufficient value after construction costs and the claims ahead of common shareholders.
NVIDIA: the relative preference still has a price
NVIDIA offers the clearest relative cash claim in this comparison. The research nevertheless stops at HOLD because the central valuation does not provide enough upside for its selected BUY threshold.
Rapid growth ties up cash in receivables and inventory. The revised model links working capital to revenue rather than assuming that one cash-conversion percentage persists forever. A slowdown in growth could ease that pressure without a collapse in demand; reported earnings alone do not establish that collections will arrive as expected.
The required return matters just as much. On the revised cash-flow mechanism, a 10.5% cost of equity produces a twelve-month expected value of approximately $288.47. Our central 11.5% assumption produces $253.51, about 12% above the $226.32 reference price. These are valuation assumptions, not two simultaneous forecasts.
Stronger durable cash generation, better collections and resilient pricing support the upside case. Customer-financing losses, margin pressure, constrained supply and inventory that fails to convert into sales weaken it. The full report records the assumptions and numerical reassessment thresholds. [NVIDIA analysis]
CoreWeave: the equipment must keep earning
CoreWeave’s opportunity depends partly on what its equipment earns after the first contracts expire. The base case assumes an eight-year economic life and later rental prices at 65% of original contract pricing. Those are model assumptions, not verified future fleet-wide outcomes.
Customer advances also change the cash-flow path. They help fund expansion, but they come with obligations to deliver services later. The revised model recognizes both new receipts and the runoff of earlier advances. Neither perpetual free financing nor an immediate subtraction of the entire opening balance describes that mechanism correctly.
Current rental-pricing strength is real counterevidence to a weak-renewal thesis. It cannot establish the rental price for large cohorts years from now. Higher advances, longer useful lives and cheaper funding could materially improve the result.
The twelve-month base value is $34.68; the probability-weighted target is $65.52 because the 20% bull case is substantial. The mean therefore sits well above the base case. The provisional SELL expresses negative expected return from the $80.26 reference under the selected assumptions—not imminent insolvency or a short-sale instruction. [CoreWeave analysis]
Applied Digital: contracted revenue is not common-equity value
Applied Digital’s $35.73 billion of nominal minimum lease payments is a substantial contractual revenue stream. It is not current cash or the present value attributable to common shareholders.
The model separates signed lease income, development costs, debt, preferred economics, minority participation, parent obligations and dilution. Residual property value and unsigned expansion add further uncertainty. The leases have value; the question is how much remains for common shareholders, and how much the stock requires from future business.
The twelve-month target is $19.57 against a $23.25 reference. The SELL label has low conviction: a favorable shift of approximately 0.40 probability points from bear to bull reaches the next rating boundary. That sensitivity is a reason to explain the economics, not to present the label as certainty.
Attractive new leases, lower remaining construction costs and funding that preserves common value would weaken the downside case. Delays, overruns and dilutive parent funding strengthen it. [Applied Digital analysis]
What the next supplier results can change
The materials risk is a testable channel. The retained report cites Lam Research’s filing on adverse regional-disruption effects involving energy and industrial commodities. It does not establish a quantified current production loss at TSMC or HBM suppliers.
A sustained constraint could delay NVIDIA systems, support rental economics for some installed CoreWeave equipment, and complicate facility completion or tenant fit-outs. Opposing effects do not automatically cancel out. Duration, delivery commitments, repricing rights and funding determine the outcome.
There is no automatic five-point probability revision after a headline. The report’s NVIDIA scenario shifts are illustrative sensitivities. New output, cost or timing evidence must be assessed before any change to earnings assumptions or scenario weights.
Subsequent findings belong in dated amendments. They should preserve the original forecast record rather than rewrite what was known before the announcement. [Reassessment commitments]
MAi Conclusion
NVIDIA: HOLD. CoreWeave: SELL, provisional. Applied Digital: SELL, low conviction in the label. NVIDIA is the relative preference, but the central valuation does not justify an aggressive BUY at the reference price. The two financed-growth businesses offer negative probability-weighted price returns under our selected assumptions, even while revenue and AI use grow.
| Security | Reference price, October 12 close | April 12, 2027 target | Price return | October 12, 2027 target | Price return |
|---|---|---|---|---|---|
| NVDA | $226.32 | $241.21 | +6.6% | $253.51 | +12.0% |
| CRWV | $80.26 | $61.83 | -23.0% | $65.52 | -18.4% |
| APLD | $23.25 | $18.73 | -19.4% | $19.57 | -15.8% |
These are probability-weighted, valuation-derived estimates in US dollars, not promised trading prices. Returns use the fixed October 12, 2026 regular-session closes, not live entry prices, and exclude dividends, taxes and transaction costs. The targets retain the October 14 round-two analysis; the materials section above adds risk interpretation without silently changing forecasts. [model record, adjudication record]
The investment distinction is who receives the cash after construction, replacement equipment, financing and other senior claims. Strong AI demand does not establish adequate returns to every common shareholder. Equally, a materials shortage can delay capacity additions while increasing project-financing stress. Neither outcome establishes an industry-wide compute glut.
Scenarios and the limits of the labels
Scenario probabilities are subjective judgments, not calibrated historical frequencies. The price targets are weighted means, not the most likely outcomes. In particular, the high CRWV bull outcome materially raises the mean above its base case.
| Security and horizon | Bear price / probability | Base price / probability | Bull price / probability |
|---|---|---|---|
| NVDA — six months | $123.97 / 25% | $228.95 / 50% | $382.97 / 25% |
| NVDA — twelve months | $129.41 / 25% | $240.59 / 50% | $403.46 / 25% |
| CRWV — six months | $7.98 / 25% | $31.93 / 55% | $211.39 / 20% |
| CRWV — twelve months | $8.67 / 25% | $34.68 / 55% | $221.40 / 20% |
| APLD — six months | $3.72 / 30% | $14.89 / 50% | $50.86 / 20% |
| APLD — twelve months | $3.95 / 30% | $15.80 / 50% | $52.43 / 20% |
The research rule calls for BUY at at least 15% expected upside with the base above the reference; SELL at at least 15% expected downside with the base below it; otherwise HOLD. Approximate favorable shifts from bear to bull of 2.47 points for NVDA, 1.27 for CRWV and 0.40 for APLD reach the next label boundary. We are not making those shifts. The test demonstrates why labels require qualification. [model record, adjudication record]
For completeness, the existing model also estimates the probability of the closing price being at or above the target on the specified date. These use a subjective three-component lognormal mixture, not observed success rates. Assumed log-price dispersion is 0.22/0.30 for NVDA and 0.45/0.60 for CRWV and APLD at six/twelve months. This extra distribution assumption is not evidence that a quoted probability is precise.
| Security and horizon | At or above target on target date | Above reference on target date |
|---|---|---|
| NVDA — six / twelve months | 42.7% / 41.8% | 48.6% / 50.7% |
| CRWV — six / twelve months | 22.4% / 23.9% | 20.1% / 20.8% |
| APLD — six / twelve months | 31.1% / 31.0% | 24.3% / 25.7% |
These are not probabilities of touching a target, an exact closing price, or confidence that the valuation method is correct. [model record]
Research basis and disclosures
This article summarizes the retained October 14 report and its round-two valuation amendment. It introduces no new forecasts. Scenario probabilities are subjective rather than calibrated historical frequencies. Kate’s first research pass had prior exposure to Claude’s headline conclusions; the research was not fully blind. Review included targeted independent reconstruction of NVIDIA and CoreWeave; Applied Digital review covered methodology and scenario arithmetic, not a complete independent rerun of every campus model.
Evidence update: October 14, 2026. Forecast reference: October 12, 2026 regular-session close. The report preserves the model references, primary-source links, calculations, uncertainty and dated review record.
Publisher Financial Interest Disclosure: As of publication, the Publisher holds a long position in NVDA. Following this research, the Publisher intends to increase that exposure. The MAi Research Team's research conclusion was reached independently of the Publisher's transaction decision.
Publisher Financial Interest Disclosure: As of publication, the Publisher has no direct economic position in CRWV. The Publisher currently has no planned transaction in the security. The MAi Research Team's research conclusion was reached independently of the Publisher's transaction decision.
Publisher Financial Interest Disclosure: As of publication, the Publisher holds a long position in APLD. Following this research, the Publisher intends to reduce or exit that exposure. The MAi Research Team's research conclusion was reached independently of the Publisher's transaction decision.
No issuer or security-related compensation associated with this research.
Research estimates are uncertain, can be wrong and are not personalized investment advice. No return is guaranteed. This article is a dated publication record and does not authorize a trade.
Open the companion research report, sources and model record →

