Nvidia's stock is tracking the price of renting its own chips

• Nvidia (NVDA) at $206.27 trades near 23x next-twelve-month earnings while H100 rental rates have risen 64% in six months. The data favors the long thesis that live compute pricing, not quarterly revenue, sets the fundamentals.
• NVDA and the H100 rental index move together at a 0.82 correlation over six months, and 0.69 after detrending. Even the 2020-vintage A100 got 8.5% more expensive to rent rather than collapsing toward zero.
• Falling B200 utilization is the one signal that could break the thesis. Watch whether GPU rents roll over alongside utilization into the August 26 print.

Wall Street is overwhelmingly long. Of 68 analysts tracked, 63 rate NVDA a buy or strong buy, with a mean target of $301 implying 46% upside.³

The consensus rests on the after-the-fact data point everyone watches, which is reported data center revenue. That number is enormous. Data center revenue hit $75.2B in the April quarter, up 92% year over year, and management guided the July quarter to $91B with no Chinese data center compute assumed at all.¹

The consensus is directionally right but it watches the wrong clock. Reported revenue is a backward signal that arrives once a quarter. The forward signal is the spot price of renting the chips, and that price is set continuously in a live market.

The Ornn GPU Price Index prices that market daily. Over the last six months the H100 rose 64%, the H200 rose 70%, and the Blackwell B200 rose 47%.⁴

Rents on the newest accelerators are climbing, not compressing, which is what a structural supply shortage looks like. The most overlooked figure is the A100. A chip Nvidia launched in 2020 should be sliding toward commodity pricing as newer silicon floods the market.

Instead its rental rate rose 8.5% over six months, which means the installed base is not depreciating on the schedule a bear would need.⁴

The honest bear case lives inside the same Ornn dataset that powers the bull case. While rental prices climbed, utilization fell. On a seven-day-average basis, B200 utilization slid from the mid-to-high 70s toward the mid-40s over six months and H200 from the mid-50s toward roughly 50%.⁴

The Ornn methodology is explicit that a high price index alongside falling utilization can signal demand cooling rather than supply tightening. That divergence matters because it is the early-warning variable.

If rents roll over while utilization keeps sliding, that sequence historically precedes a capex digestion phase, and Nvidia sells the picks in that gold rush.

The structural bear case is more concrete than valuation hand-wringing. Hyperscaler custom silicon is real and shipping. Google's TPU, Amazon's Trainium, Microsoft's Maia, and Meta's MTIA are designed to claw inference workloads away from Nvidia.⁶⁷

TrendForce projects ASIC-based AI servers reach 27.8% of total AI server shipments in 2026, the highest share since 2023, rising toward nearly 40% by 2030. GPUs still hold nearly 70% of shipments today.⁶

The threat is sharpest in inference, the larger long-run market and the workload most exposed to defection. Independent research compiled by Introl, citing New Street Research, projects Nvidia's share of the inference market falling from above 90% toward 20% to 30% by 2028 as custom silicon scales.⁷

Capex is still growing, not contracting, but the rate of change is the variable to watch. Aggregate 2026 capital spending across Amazon, Alphabet, Microsoft, and Meta is guided to roughly $715B to $725B, up about 77% from the roughly $410B spent in 2025.⁸

Moody's, using a broader basket that places 2026 spending near $785B, projects 2027 capex approaching $1T, a still-large but decelerating roughly 27% growth rate. A single guidance cut from any one of the four would reprice the entire complex.⁹

China remains a zero. Nvidia guided the July quarter assuming no Chinese data center compute revenue, which both de-risks the guide and concedes a market it once led.¹

The last time the market anointed a hardware cycle leader of this scale, in 2000, the stock that defined the era fell nearly 90% within two years as the buildout digested. Nvidia's earnings are real in a way that comparison flatters, but the lesson on cycle timing stands.¹³

The evidence supports the view that Nvidia's pricing power is real and live, not a quarterly story, because the spot rental price of its chips correlates with the stock at 0.82 and even six-year-old A100s are getting more expensive to rent. The counter-argument from falling B200 utilization is the one signal that could break the thesis, so the variable to watch is whether GPU rents roll over alongside utilization into the August 26 print. Until rents break, the compute market is voting with real dollars every single day.

1. Nvidia Q1 FY27 press release and financial statements, filed via SEC 8-K (accession 0001045810-26-000051), May 20, 2026. https://www.sec.gov/Archives/edgar/analytics/0001045810/000104581026000051/q1fy27pr.htm
2. yfinance (Tier 0) NVDA intraday quote, June 10, 2026. Price $206.27 (last prior close $208.19 on June 9), 52-week range $140.85 to $236.54, market cap ~$5.0T on 24.22B shares outstanding. https://finance.yahoo.com/quote/NVDA
3. finnhub price_target and recommendation_trends (Tier 0), as of June 6, 2026. Mean target $301, 68 analysts, 63 buy or strong buy. https://finnhub.io/
4. Ornn Compute Price Index (OCPI), the institutional GPU-rental benchmark distributed via Bloomberg Terminal, with ICE compute futures referencing it (history-simple, daily-index-simple, and volume-metrics endpoints). Six-month rental changes December 9, 2025 to June 9, 2026. https://ornn.com/
5. NVDA-Ornn correlations recomputed from stored series via scipy.pearsonr. H100 level 0.82 (6M), H200 level 0.85 (6M), detrended 0.69. https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.pearsonr.html
6. TrendForce, Global AI Server Shipments Forecast to Grow Over 28% YoY in 2026, January 20, 2026, with March 18 and 20, 2026 follow-ups. https://www.trendforce.com/presscenter/news/20260120-12887.html
7. Introl, Custom Silicon Inflection 2026 (citing New Street Research), February 23, 2026. Nvidia inference-market share projected to fall from above 90% toward 20% to 30% by 2028. https://introl.com/blog/custom-silicon-inflection-2026-hyperscaler-asics-nvidia-gpu
8. Tom's Hardware, April 30, 2026, and OfficeChai, May 2, 2026. 2026 hyperscaler capex ~$715B to $725B, up ~77% from ~$410B in 2025. https://www.tomshardware.com/
9. Moody's Ratings via RCR Wireless, May 18, 2026. Hyperscaler capex ~$785B in 2026 approaching ~$1T in 2027, roughly 27% growth. https://www.rcrwireless.com/
10. Wedbush and supply-chain reporting, February to May 2026. Nvidia locked roughly 60% of TSMC CoWoS advanced packaging through 2026. https://www.wedbush.com/
11. Focus Taiwan, DataCenterKnowledge, wccftech coverage of GTC Taipei, June 1 to 5, 2026. Vera Rubin arrives H2 2026, supply-constrained as it ramps. https://focustaiwan.tw/
12. moomoo earnings calendar, June 6, 2026. Next Nvidia print scheduled August 26, 2026. https://www.moomoo.com/
13. Harding Loevner, Nvidia and the Cautionary Tale of Cisco Systems. Cisco fell roughly 88% from its March 2000 peak within two years. https://www.hardingloevner.com/insights/nvidia-and-the-cautionary-tale-of-cisco-systems/