Nvidia (NVDA) vs 10Y Treasury Yield
Nvidia (NVDA) closed FY26 Q3 with $57.0 billion in quarterly revenue, up over 60% year-over-year, with the data center segment generating roughly $52 billion or 59% of total revenue per Nvidia's earnings release. SPY (SPDR S&P 500 ETF) tracks the broad U.S.
Also known as: Nvidia (NVDA) (STK_NVDA, Nvidia) · 10Y Treasury Yield (10Y yield, 10 year treasury, TNX)
Why This Comparison Matters
Nvidia (NVDA) closed FY26 Q3 with $57.0 billion in quarterly revenue, up over 60% year-over-year, with the data center segment generating roughly $52 billion or 59% of total revenue per Nvidia's earnings release. SPY (SPDR S&P 500 ETF) tracks the broad U.S. large-cap index where Nvidia itself sits as one of the largest weights. The pair captures whether Nvidia is leading the AI capex cycle (NVDA outperforming SPY) or whether the AI trade is broadening to suppliers and customers (NVDA in line or lagging SPY). The 2023 to 2024 stretch saw NVDA outperform SPY by roughly 200 percentage points cumulative; Q1 2025 produced NVDA's worst quarter in a decade at -22%, signaling regime stress that resolved over the balance of 2025.
What the NVDA-vs-SPY momentum pair is actually measuring
Nvidia trades on Nasdaq under ticker NVDA. The company reported FY26 Q3 (calendar Q3 2025) revenue of $57.0 billion, up more than 60% year-over-year, with $52 billion of that in the data center segment that supplies AI training and inference accelerators (H100, H200, B100 Blackwell). The SPDR S&P 500 ETF (SPY) tracks the S&P 500 with $560 billion plus in assets. Nvidia is one of the top-three weights in SPY at roughly 6 to 7% of the index, so a long-NVDA short-SPY trade is not a clean isolation; it is a bet on Nvidia outperforming the other 500 names by enough to overcome the embedded long-NVDA exposure in the SPY short.
Momentum here is measured by the ratio of trailing returns. The simplest version is 12-month NVDA total return minus 12-month SPY total return. The professional version, used at quantitative desks and described in Asness, Moskowitz, and Pedersen (2013, Journal of Finance), is the 12-1 momentum signal: 12-month return excluding the most recent month, which removes short-term mean reversion. NVDA's 12-1 momentum versus SPY's 12-1 momentum is the cleanest way to see whether Nvidia is structurally leading the index.
The 2023 to 2025 momentum cycle
Nvidia returned roughly +239% in calendar 2023 and +171% in calendar 2024 versus SPY at +26.3% and +25.0% respectively per index providers. The cumulative two-year NVDA outperformance over SPY was approximately 360 percentage points. The catalyst was the ChatGPT release in November 2022 that triggered hyperscaler AI capex acceleration: Microsoft Azure capex rose from $33 billion in FY22 to a guided $80 billion in FY26; Meta capex rose from $32 billion in 2022 to a guided $65 to $75 billion in 2025; Amazon AWS capex rose to a guided $100 billion plus in 2026; Google capex stepped up to $75 to $85 billion in 2025. Almost all of that incremental capex flowed to Nvidia accelerators in the 2023 to 2024 window.
Nvidia entered Q1 2025 with consensus expectations priced for continued 2x-plus revenue growth and the stock returned -22% in Q1 2025, the worst single quarter in a decade. The drawdown was triggered by the late-January DeepSeek R1 release that demonstrated competitive AI training results at a fraction of Nvidia-stack training cost, raising questions about whether the AI capex cycle had over-built. NVDA recovered through Q2 to Q4 2025 as data-center revenue accelerated to $52 billion in FY26 Q3, but the volatility of the relative pair stepped up: realized NVDA-SPY pair volatility rose from 25% annualized in 2024 to 40% annualized in 2025.
Duration sensitivity, real yields, and the AI-multiple anchor
Nvidia trades at a forward P/E that has ranged from 25x to 50x across 2023 to 2026, which makes it a high-duration equity. The standard duration formula for an equity is approximately the inverse of the dividend yield plus risk-premium adjustments; for a non-dividend-paying high-multiple stock like Nvidia, duration is around 25 to 35 years versus SPY closer to 18 to 22 years. A 100bp rise in DFII10 (10-year TIPS yield, the cleanest real-rate proxy on FRED) should compress NVDA's discount-rate-driven valuation roughly 1.5x as much as SPY's.
But the 2023 to 2024 episode broke this textbook duration relationship: DFII10 rose from 1.20% in early 2023 to 2.42% in October 2023 while NVDA returned 239%. The reason is that earnings revisions outran discount-rate compression: consensus 2024 revenue estimates for Nvidia rose from $35 billion in Q1 2023 to over $125 billion by year-end, a 250% upward revision that swamped any duration-driven multiple compression. When the AI-earnings-revision cycle slows or reverses, duration sensitivity reasserts. Q1 2025 captured this: DFII10 held in the 1.9% to 2.2% range while NVDA fell 22% on flat real yields, suggesting earnings-revision risk rather than duration was the dominant driver.
How Convex Risk Appetite Index reads the pair
Convex Risk Appetite Index (CRAI) blends VIX, ICE BofA HY OAS (BAMLH0A0HYM2), and equity put-call ratios. Nvidia's relative outperformance versus SPY is sensitive to CRAI in a non-linear way. When CRAI is in its top quartile (risk-on), NVDA tends to outperform SPY because momentum capital crowds into the highest-beta AI exposure; when CRAI moves to its lowest quartile (risk-off), NVDA tends to underperform SPY because high-beta tech experiences the largest forced unwinds. The August 5, 2024 yen-carry-unwind episode produced exactly this: NVDA fell -6.4% on the day versus SPY at -3.0%, with NVDA's higher beta amplifying the cross-asset shock.
The CRAI-conditioned read of the NVDA-SPY pair has been reliable since 2023: 30-day forward NVDA-SPY excess returns regressed on CRAI quartile produce a roughly 8% spread between top and bottom quartile of CRAI. Allocators use this as a tilt rule rather than a binary signal: in top-quartile CRAI, slightly tilt toward NVDA over SPY; in bottom-quartile CRAI, tilt toward SPY-ex-NVDA. The current April 2026 CRAI reading sits in the upper half of its 12-month range, consistent with continued NVDA leadership but elevated drawdown risk.
The Q1 2025 drawdown and what it taught about regime stress
NVDA fell 22% in Q1 2025, the worst quarter in a decade, on a combination of the DeepSeek R1 release (January 27, 2025), tariff escalation rhetoric on semiconductor imports, and consensus AI-capex normalization concerns. SPY returned approximately +1% over the same quarter, producing roughly 23 percentage points of underperformance. This is the largest single-quarter underperformance in NVDA's modern history.
The drawdown was a regime-stress test for the AI trade. Three things happened simultaneously: hyperscaler capex guidance held flat or rose (Meta raised from $60 to $65 billion to $65 to $75 billion in late January 2025; Microsoft confirmed $80 billion FY26 capex in late January 2025; Amazon guided $100 billion plus for 2026 in early February 2025); Nvidia's revenue accelerated through the year ($57 billion in FY26 Q3 versus $35 billion in FY25 Q3); and yet the stock retraced 22% in the first three months. The lesson is that NVDA-SPY momentum can decouple from underlying fundamentals on positioning unwinds, and the pair's fundamental anchor reasserts only over multi-quarter windows.
Practical takeaway and current April 2026 configuration
The pair currently reads as Nvidia leading SPY on momentum but with elevated downside skew. Nvidia FY26 Q3 results posted $57 billion revenue and over 60% year-over-year growth, which keeps the fundamental case intact. NVDA enters mid-2026 trading at a forward P/E in the high 20s, below the 35x peak of 2024 but above the 18x SPY multiple. Realized NVDA-SPY pair volatility ran around 35% annualized over the trailing six months, elevated relative to the 2024 baseline of 25%.
The trade structure is to hold the pair as a beta tilt rather than a high-conviction directional bet. Long NVDA short SPY captures continued AI capex acceleration; short NVDA long SPY captures broadening of the AI trade to suppliers (TSMC, ASML, AVGO) and customers (Meta, Microsoft, Amazon, Google). Position sizing should account for 35% pair vol plus the embedded NVDA weight in the SPY short (roughly 6 to 7% of any short-SPY position is itself short-NVDA). The cleanest expression for AI-capex-cycle continuation is long NVDA against an equal-weighted S&P 500 short (RSP) rather than against the cap-weighted SPY, which removes the NVDA-on-NVDA offset.
Conditional Forward Response (Tail Events)
How 10Y Treasury Yield has historically behaved in the 5 sessions following a top-decile or bottom-decile daily move in Nvidia (NVDA). Computed from 1,242 aligned daily observations ending .
Following these triggers, 10Y Treasury Yield rises 0.59% on average over the next 5 sessions, versus an unconditional baseline of +0.50%. 124 qualifying events; 10Y Treasury Yield closed positive in 51% of them.
Following these triggers, 10Y Treasury Yield rises 1.12% on average over the next 5 sessions, versus an unconditional baseline of +0.50%. 125 qualifying events; 10Y Treasury Yield closed positive in 57% of them.
Past behavior in the tails is descriptive, not predictive. Mean response is the simple arithmetic mean of compounded 5-day forward returns following each trigger event; baseline is the unconditional mean across the full sample window. Edge measures the gap between the two.
90-Day Statistics
Explore Each Metric
Related Scenarios & Forecasts
Get daily macro analysis comparing key metrics delivered to your inbox. Stay ahead of market-moving divergences.
Frequently Asked Questions
How much has Nvidia outperformed SPY since the AI trade started?+
Nvidia returned approximately +239% in calendar 2023 and +171% in calendar 2024 versus SPY at +26.3% and +25.0%. Cumulative two-year outperformance was roughly 360 percentage points. Q1 2025 saw NVDA fall 22%, the worst quarter in a decade, on DeepSeek R1 disruption and AI-capex normalization concerns. NVDA recovered through 2025 as data-center revenue accelerated to $52 billion in FY26 Q3. The current April 2026 forward P/E is in the high 20s, below the 35x 2024 peak but above SPY's 18x.
Is Nvidia really duration-sensitive, or does the AI cycle override it?+
Nvidia trades at a high forward P/E (25x to 50x across 2023 to 2026), giving it equity duration around 25 to 35 years versus SPY at 18 to 22 years. Textbook says a 100bp rise in DFII10 (10-year TIPS yield, the real-rate proxy) should compress NVDA roughly 1.5x as much as SPY. The 2023 to 2024 episode broke this: DFII10 rose from 1.20% to 2.42% while NVDA returned 239% because earnings revisions ran ahead of discount-rate compression. When earnings revisions slow, duration reasserts: Q1 2025 saw NVDA fall 22% on flat real yields.
What was the DeepSeek R1 release and why did NVDA drop?+
DeepSeek R1 was released on January 27, 2025 as an open-source AI reasoning model that demonstrated competitive results at a small fraction of Nvidia-stack training cost. NVDA fell roughly 17% on January 27, 2025 alone, the largest single-day market-cap loss in U.S. history at the time, on concerns that the AI capex cycle had over-built relative to actual compute requirements. The drawdown extended through Q1 2025 to 22% before recovering as hyperscaler capex guidance held or rose: Meta raised to $65 to $75 billion, Microsoft confirmed $80 billion FY26, and Amazon guided $100 billion plus for 2026.
How should I size a long-NVDA short-SPY pair trade?+
Recognize that NVDA is roughly 6 to 7% of SPY itself, so a dollar-neutral long-NVDA short-SPY trade has embedded long-NVDA exposure in the SPY short side. The clean expression is long NVDA against an equal-weighted S&P 500 short via RSP (Invesco S&P 500 Equal Weight ETF) where NVDA is roughly 0.2% rather than 6 to 7%. Realized NVDA-SPY pair volatility ran around 35% annualized over the trailing six months versus 25% in 2024, so position sizing should target lower notional than the volatility might suggest if the historical reference is 2024.
Which Convex composite is most relevant for NVDA versus SPY?+
Convex Risk Appetite Index (CRAI), built from VIX, ICE BofA HY OAS, and equity put-call ratios, is the most relevant. Nvidia's relative outperformance is non-linearly sensitive to CRAI: top-quartile CRAI produces roughly 8% of forward 30-day NVDA-SPY excess return versus bottom-quartile CRAI. The August 5, 2024 yen-carry-unwind illustrated the asymmetry: NVDA fell 6.4% versus SPY at -3.0%. Allocators use CRAI as a tilt rule: in top-quartile CRAI, tilt toward NVDA; in bottom-quartile CRAI, tilt toward broader SPY-ex-NVDA exposure.
How concentrated is the AI trade in Nvidia versus the rest of SPY?+
In April 2026, the AI-leveraged stocks (Nvidia, Microsoft, Meta, Amazon, Google parent Alphabet, Broadcom, Apple) constitute roughly 33 to 35% of SPY by market cap. Nvidia alone is 6 to 7%. The Magnificent 7 broadly account for over 90% of S&P 500 earnings growth contribution in 2024 to 2025 per FactSet. The NVDA-SPY momentum pair is therefore a concentrated bet within a concentrated index; broadening of the AI trade beyond Nvidia (to TSMC, ASML, AVGO suppliers and to Meta, Microsoft, Google customers) tends to compress the pair even if Nvidia continues to perform well in absolute terms.
What is the historical base rate for Nvidia leading SPY by this much?+
Looking at the post-1999 daily history of NVDA-SPY relative returns, 24-month rolling outperformance of 360 percentage points is in the 99th percentile of all rolling 24-month windows. The closest analogs are Cisco Systems in 1998 to 2000 (CSCO outperformed SPY by approximately 280 percentage points over 24 months ending March 2000) and Apple in 2007 to 2009 (AAPL outperformed by approximately 180 percentage points around the iPhone launch). The post-2000 Cisco outcome is the bearish historical analog: CSCO retraced 86% from peak through 2002. The post-2007 Apple outcome is the bullish analog: AAPL went on to multiply substantially from the relative-momentum peak. Which analog applies depends on whether AI capex acceleration is durable beyond 2027, which is the central question for the pair.
Related Comparisons
Explore Across Convex
Data sourced from FRED, CoinGecko, CBOE, and other providers. This page is for informational purposes only and does not constitute financial advice. Past performance does not guarantee future results.