GPU rental prices look like a leading indicator for Dell and Arm stock

The Ornn H100 rental price index moves Dell and Arm three weeks before the stocks move, a forward signal that holds at the 1 percent significance level. The same index has no predictive grip on Nvidia, the company that makes the chips.¹

• The Ornn H100 price index leads Dell (DELL) and Arm (ARM) by three weeks at +0.51 forward correlation to each. Rising GPU prices time the entry on those names, not Nvidia.
• Nvidia shows no forward relationship at -0.05 (p of 0.64), trading coincident with the rental market rather than lagging it. Only Dell and Arm clear both the correlation and Granger bar across the 16-name AI universe.
• The H100 index spiked roughly 64 percent in three weeks from late April and sits near 2.59 dollars per GPU-hour now. That tightening has historically front-run Dell and Arm upside by about three weeks.

GPU rental prices are the cleanest real-time read on AI compute demand that exists. When H100 capacity tightens, the rental price rises within days, well before any company reports a quarter. The Ornn H100 SXM index is that price, updated daily, with two years of history behind it.

The index is not a smooth ramp. It fell almost continuously from roughly 3.15 dollars per GPU-hour in mid-2024 to a 1.63-dollar trough in early February 2026 as supply caught up with demand.

Then it snapped. From about 1.98 dollars on April 30 it spiked to a 3.24-dollar peak on May 19, a 64 percent move in three weeks and 98 percent off the February trough. It has since eased to 2.59 dollars.¹

That spike is a demand signal. The question for capital is simple. Does the rental price move before the stocks that sell into the same demand, and if so, which ones.

Consensus treats Nvidia as the purest expression of AI compute demand. The logic: Nvidia sells the chips, so its stock should track the value of those chips most tightly. The data says otherwise.

Across a 16-name AI large-cap universe tested against the H100 index, Nvidia shows no forward relationship. Its 3-week-momentum signal carries a -0.05 correlation to forward returns, a value that cannot be told apart from zero (p of 0.64).⁵

If anything the chipmaker's stock moves coincident with or slightly ahead of the index. That is consistent with Nvidia being the discounting mechanism rather than the thing being discounted.

The names that the rental price actually predicts sit one layer downstream. Dell builds the AI servers that house the GPUs, and Arm licenses the CPU cores that attach to a growing share of GPU clusters.

Both show forward correlations above 0.50 at a three-week lag, and both pass a Granger causality test that Micron fails despite its correlation. Dell posts an F of 4.2 and Arm an F of 6.6, each significant below the 1 percent level.⁵

The relationship rests on two years of data covering a single AI capital cycle. A correlation that holds through one expansion can break in a contraction, and the H100 index has only traded through a falling-then-spiking regime, never a sustained crash.

If GPU rental prices fall hard because supply finally overwhelms demand, the same three-week lead that delivered Dell and Arm upside would deliver downside. A falling-knife signal is harder to trade than a rising one.

There is also a sample-size problem. The non-overlapping four-week robustness test leaves only about 25 independent observations per name, enough to reject the null at the 1 percent level but not enough to rule out regime dependence. The significance levels are computed per name and are not adjusted for testing sixteen names at once, so a single-name result near the threshold should be read with that multiplicity in mind. On that basis Arm's relationship (p below 0.001) is the more robust of the two, while Dell's (p of 0.008) is the more sensitive to a multiple-comparison correction.

The contemporaneous correlation between weekly H100 moves and weekly stock moves is near zero, which is the point. The signal is in the lead-lag, not the co-movement. That same property means a single noisy month in the rental index can flip the signal and whipsaw a position.

The April-to-May spike itself shows the fragility. The index ran from 1.98 to 3.24 dollars in three weeks, then gave back roughly a fifth of that move within days, bouncing between 2.63 and 3.20 dollars through early June.

A signal that volatile demands tight risk control. The same momentum that fires the entry can reverse before the three-week downstream response arrives.

The economic story can also reverse causality in a way the statistics cannot fully separate. Rising rental prices may not cause Dell and Arm to rise so much as both may respond to a third driver such as hyperscaler capex announcements, or to a company-specific catalyst like an earnings surprise, with the rental market reacting a few days faster than equity analysts update models.

If that is the mechanism, the edge is real but fragile, because it depends on equity markets staying slow to reprice the same news the rental market has already absorbed. The moment systematic funds start trading the Ornn index directly, the three-week lead compresses toward zero.

Micron illustrates the fragility. It correlates with the rental price at +0.32 but fails the Granger test (F of 1.1, p of 0.36), meaning its relationship is coincident rather than predictive. A coincident signal offers no timing edge at all.

The Ornn H100 rental price is a three-week leading indicator for Dell and Arm, the downstream names that fill the racks and license the cores, and it carries no predictive grip on Nvidia. The edge holds at the 1 percent significance level but rests on a single capital cycle, so the largest risk is a regime change where supply overwhelms demand and the same lead delivers downside. The signal favors Dell and Arm on rising H100 prices and warns that once funds trade the index directly, the three-week lead compresses toward zero.

1. Ornn AI Inc., Ornn Compute Price Index (OCPI), H100 SXM daily rental series (primary data, accessed June 12, 2026). Supports the H100 rental price levels and the early-May 2026 spike-then-ease narrative. Public corroboration: Crypto Briefing, “H100 rental prices slide after early May surge, says Ornn AI Inc,” May 27, 2026. https://cryptobriefing.com/h100-rental-prices-slide-ornn-ai/
2. Yahoo Finance (via yfinance), equity prices, market capitalization, forward P/E, and revenue growth, as of market close June 12, 2026. Supports the price, market-cap, forward-P/E, and revenue-growth columns in the Key Data Table and the figures cited in the body. https://finance.yahoo.com
3. Financial Modeling Prep (FMP), one-year price performance, as of market close June 12, 2026. Supports the 52-week return column in the Key Data Table (DELL +247.8%, ARM +168.1%, MU +719.1%, NVDA +41.8%). https://site.financialmodelingprep.com
4. Dell Technologies, first quarter fiscal 2027 results, May 28, 2026. The Form 8-K (Exhibit 99.1) press release supports the financial results, the $16.1B AI-optimized server revenue (up 757%), and the raised ~$60B FY2027 AI-optimized server revenue guidance: https://www.sec.gov/Archives/edgar/analytics/0001571996/000157199626000021/exhibit991earnings8kq1fy27.htm. The record $51.3B AI backlog (CFO: “our backlog now sits at $51.3 billion”) and the constraint statement (“demand for AI and traditional servers is exceeding supply; primary constraints are memory (DRAM, NAND), CPUs”) are from the Q1 FY2027 earnings call, May 28, 2026: https://www.fool.com/earnings/call-transcripts/2026/05/28/dell-dell-q1-2027-earnings-call-transcript/
5. ProCap Insights quantitative analysis, 3-week-momentum forward correlation and Granger F-tests on weekly data over June 2024 to June 2026 (97 weekly observations; about 25 independent observations per name under four-week non-overlap). Tested universe of 16 names is AMD, AMZN, ARM, AVGO, CRWV, DELL, GOOGL, META, MSFT, MU, NVDA, ORCL, PLTR, SMCI, TSM, VRT; newly listed names are tested over their available trading history (CoreWeave listed March 28, 2025). Granger causality (nested OLS, three weekly lags) Dell F=4.23 p=0.008, Arm F=6.59 p=0.000, Micron F=1.09 p=0.357, Nvidia F=1.16 p=0.329. Dell and Arm are the only two names clearing both the +0.50 forward-correlation and the Granger-causality bar. https://procapinsights.com/methodology