Did Our Score See NVIDIA's Surge Coming?
The score hit A in 2020. The stock exploded in 2023. A lead-lag study — with the caveats stated first.
6 min read · 1 company · 10 years of scores · Not investment advice
The fair critique
A model that gives its highest score — 87/A — to the world's most valuable company is not telling anyone anything new. NVIDIA at the top of the table is a description of the present, not a discovery. The question that matters is whether the score saw it coming: was it high before the surge, or only after?
We have ten years of annual Blitz Index estimates for NVIDIA. Here they are:
The score crossed into A territory in 2020. The stock's historic surge — the ChatGPT-driven repricing — began in 2023. Three years earlier, on the basis of the data-center business crossing $10B, CUDA lock-in deepening, and gross margins expanding toward 65%, the model was already at 81/A.
The 2022 dip
Then 2022: the stock fell roughly 50% in the tech drawdown. The score fell too — to 79/B. Not because the model "predicted" the crash, but because the business inputs genuinely softened: gaming weakness, the crypto hangover, margin pressure during the inventory correction. Then 2023: back to 87/A, where it has stayed.
So the sequence reads: score rises (2017–2020) → price surges (2023) → score holds (2023–2025), with score and price falling together in the 2022 drawdown. At the key inflection, the score led the price by roughly three years.
Data-center revenue had crossed $10B and was growing ~50%. CUDA had become the standard language of GPU computing — a B1 network-effect moat deepening, not widening. Gross margin was marching from 59% toward 65%+. None of this required knowing about large language models. The business fundamentals behind the 2023 surge were measurable in 2020, and the model measured them.
What the flat years said
The 2017–2019 plateau — 69, 69, 67 — is as instructive as the 2020 jump. These were the years NVIDIA was "just" a gaming GPU company with a promising data-center side business. The score held above the 65 threshold throughout, grading it a successful blitzscaler even when the stock was merely compounding rather than exploding. In retrospect, the model was patient: it did not need the AI narrative to recognize the CUDA moat deepening year by year.
That patience cuts both ways. A 67/B in 2019 would not have told anyone a historic repricing was three years out — it said the business was sound and the moat was widening. The surge required an external catalyst, large language models, that no business-quality model can foresee. The score's job is to identify the loaded spring, not to predict when it releases. Investors who confuse the two are making the exact error the ARK backtests documented: mistaking a business grade for a price call.
The caveats — stated plainly
The annual scores were reconstructed in 2026 with knowledge of what happened — hindsight bias is real. The one true blind data point, scored at T=2016-12-31 with no knowledge of the future, came out at 66/B against the estimate's 64/B. Encouraging — but it is one point.
NVIDIA is the success story. The ARK backtests showed this same model could not pick winners at the 2021 valuation peak (r = −0.15). A score that "saw NVIDIA coming" in reconstruction is not proof it will see the next one in real time.
In 2020, NVIDIA at 81/A also traded at a demanding multiple. The model would have said "great business" — it would not have said anything about whether the stock was cheap. That distinction is the entire lesson of the ARK tests, and it applies here.
What this does establish
The Blitz Index is not merely a lagging indicator that stamps high scores on already-successful companies. The fundamentals that drove the 2023 surge were visible in the model's own inputs three years early — and the model's worst miss on NVIDIA (the 2022 dip to 79/B) was directionally honest about the business cycle.
Whether that measurement can be made reliably, in real time, before the surge — across dozens of companies, not one — is the question the v1.1 research track exists to answer. This study is the opening of that inquiry, not its conclusion.
Method: Blitz Index v1.0 (frozen Oct 2026). Annual scores are estimates reconstructed 2026-10-11 (labeled as such in history.json); the 2016 blind score (66/B) used only information available at T=2016-12-31. Price references are approximate public-market history.
Not investment advice. A historical methodology study. It contains no buy, sell, or hold opinion on any security. Past validation does not predict future results.