The index is not an index. It is a press release with a chart attached.
That is the first thing that occurred to me when I saw the headline circulating last week: Pokémon cards, up roughly 40x since January 2019, closing in on NVIDIA's 41x over the same window. The framing is designed to provoke. An alternative asset, born from a children's trading card game, allegedly matching the greatest equity run of the decade. Two numbers, placed side by side, doing the work of an argument.
But a ratio without a methodology is not evidence. It is marketing. And having spent years auditing price feeds, yield curves, and custody arrangements, my instinct is always the same: before you argue about the conclusion, you stress-test the foundation. So I did. What I found was not a story about Pokémon cards beating semiconductors. It was a story about how a low-frequency, last-sale-priced, survivorship-biased basket of assets can be made to look like a benchmark.
Math has no mercy. Neither should readers.
Context: how a trading card index gets built
The numbers come from Card Ladder, a collectibles data platform that tracks audited sales across major marketplaces. The 40x figure refers to its Pokémon index, a basket of what the platform treats as representative cards. The 41x figure is NVIDIA's total return over the same period. The comparison was amplified by a Blockworks chart placing trading card games alongside crypto, equities, and commodities, and then picked up by crypto media outlets hungry for a narrative that flatters alternative assets.
On its face, the story has real emotional pull. Pokémon is a genuine cultural phenomenon with a thirty-year history, a demographic bulge of now-affluent thirtysomethings who grew up with it, and a supply of early printings that is genuinely finite. There is a real asset here. The question is whether the 40x number describes that asset, or describes the way the asset was measured.
This matters because the article that carried the comparison was not a research note. It was a consumer-facing explainer with a risk disclaimer bolted onto the end. Its central claim — that collectibles have nearly caught a mega-cap technology stock — rests entirely on one data point that, on inspection, cannot be independently reproduced. Everything downstream of that number inherits its fragility.
Core: the five failures hiding inside a single multiple
Start with the arithmetic, because the arithmetic is the least interesting part and it already undermines the headline.
Over 7.8 years, a 40x return implies a compound annual growth rate of roughly 60.5%. A 41x return implies roughly 61.0%. The gap between them is half a percentage point. In any asset class, a 50-basis-point difference over a multi-year window is noise. It is far smaller than the measurement error embedded in a collectibles index. The honest statement is not that cards are "about to overtake" NVIDIA. It is that, under this specific calculation, the two are statistically indistinguishable — and even that statement assumes the calculation is sound.
It is not sound. Five structural defects sit underneath it, and the source material acknowledged only one.
First, start-date bias. January 2019 was not a neutral starting line. It was simultaneously a cyclical trough for collectibles and the bottom of the crypto bear market, with Bitcoin trading near $3,700. Choosing a known low as your baseline mechanically inflates every multiple that follows. Shift the start date by a single year and the implied CAGR for the card index jumps by more than eleven percentage points. A number that sensitive to an arbitrary date choice is not a measurement. It is a choice dressed as a measurement.
Second, survivorship bias. The index tracks cards that the market has already validated — the first-edition Charizards, the sealed vintage boxes, the graded grails. It does not track the thousands of ordinary cards printed in the same era that are worth approximately nothing. Comparing a curated basket of winners against a single winning company is not a comparison. A fair analogue would be the entire Pokémon card pool versus the entire Russell 3000, or the fifteen most expensive cards versus the fifteen largest companies by market cap. Neither comparison was run, because neither would produce a headline.
Third, and most damaging, stale pricing. Low-frequency assets are typically valued at their last executed sale. When a card trades once every few weeks or months, that price becomes the mark until the next trade. The result is a price series that is artificially smooth. Volatility is systematically suppressed, drawdowns are quietly absorbed into flat segments of the chart, and the asset appears steadier than it is. This is not a cosmetic issue. Any risk metric computed on a stale-priced series — Sharpe ratio, max drawdown, volatility — will be wrong, and wrong in the flattering direction. The index looks calm because nobody is quoting it.
Fourth, the asymmetry of the comparison itself. NVIDIA is also a survivor. It is the company that happened to win the AI accelerator cycle. Pitting a survivor-selected basket of cards against a survivor-selected single stock, and then treating the resulting ratio as meaningful, is a category error compounded by a selection error.
Fifth, thin-market contamination. The source material itself concedes that a handful of high-profile auctions can move the index materially. In professional index construction, anomalous prints are trimmed, capped, or excluded precisely because single transactions should not redefine a benchmark. Here, they are allowed to stand. The index is not just noisy. It is event-driven.
Put these together and the 40x number stops being a fact and becomes a artifact. It is the output of a specific set of methodological choices, none of which were disclosed, all of which point the same direction.
Now consider what the 41x on the other side actually encodes. If NVIDIA's split-adjusted price was around $3.30 in January 2019, a 41x return implies a price near $135 by late 2026. NVIDIA's post-split high in mid-2024 already approached that level. Which means the more plausible reading of the "cards caught NVIDIA" story is not that cards surged. It is that the reference asset went sideways for two years. The narrative attributes motion to the runner when the finish line simply stopped moving.
And then there is the financing cost that never appears in any of these charts. A card is not a stock. It pays no dividend, no interest, no rent. It generates zero cash flow. Holding it requires a safe, insurance, and periodic re-grading. Selling it requires grading fees that can run from twenty to a hundred and fifty dollars per card with turnaround times measured in weeks, marketplace commissions north of ten percent, auction house buyer's premiums above twenty percent, and a liquidation window that stretches from weeks to months. Round-trip friction on a collectible position can consume thirty to forty percent of nominal gains. On top of that, in the United States, collectibles are taxed at a long-term capital gains rate of up to 28%, versus 20% for equities. Apply fees and taxes to a 40x nominal return and the realized figure compresses toward the high twenties. The headline multiple is a gross number in a world where investors live on net ones.
I learned this lesson the hard way in the summer of 2020, when I modeled the yield curves of Compound and Aave and watched inflationary token emissions masquerade as organic demand. The lesson transferred cleanly: when a return looks extraordinary, decompose it into price appreciation, subsidies, and measurement artifacts. Here, the decomposition is unflattering. There is real scarcity in vintage cards. There is real demographic demand. There is also a very large measurement subsidy baked into the multiple.
Contrarian: what the bulls actually get right
It would be lazy to dismiss the entire thesis, and lazy is a failure mode I try to avoid.
The bulls are right about three things, and the skeptics who ignore them are making their own category error.
First, the underlying asset is not a token with no cash flow and no utility. Pokémon cards have genuine consumption demand — play, collection, nostalgia, community. That demand is not purely reflexive. The 1996-to-2000 cohort is now in its peak earning years, and the willingness to pay for a piece of childhood is a real preference, not a manufactured one. Holding a card also delivers what economists call convenience yield: the emotional and social return of ownership. That return is invisible in any financial model, and pretending it does not exist is its own form of dishonesty.
Second, physical scarcity in early printings is real and increases over time as cards are lost, damaged, and permanently encased. Unlike a protocol that can mint arbitrarily, the original supply is fixed. The Pokémon Company can reprint commemorative sets, but it cannot reprint a 1999 first edition. That distinction matters.
Third, the bulls are correct that traditional finance has been slow to price alternative assets. The institutionalization of collectibles — through funds, fractional platforms, and eventually tokenized custody — is a genuine secular trend, not a fad. The mistake is not believing in the trend. The mistake is using a distorted index to justify the entry price.
Where the bulls go wrong is in extrapolation. They take a specific window's optimal performance and treat it as the asset class's normal return. That is the common grammar of every bubble. The narrative does not describe the asset. It describes the best possible entry point into the asset, selected after the fact.
And the most conspicuous omission in the entire discussion is the drawdown that followed the 2021 peak. Collectibles broadly corrected fifty to seventy percent from their early-2021 highs. Anyone who entered at the top is still underwater. The 40x belongs to investors who bought in 2019 and held. For the vast majority of actual participants, the number is paper, not portfolio.
Takeaway: the real question is not cards versus chips
The genuine insight buried under this headline is not that Pokémon beat NVIDIA. It is that a non-productive, illiquid, zero-cash-flow asset with thirty to forty percent round-trip friction, a concentrated single-point grading monopoly, and an IP owner with unilateral reprint rights was successfully packaged as a benchmark comparison. The packaging worked because the number was large and the methodology was invisible.
For anyone building in the tokenization space, that is the actual lesson. The pain point the article identifies — collectibles cannot be sold in seconds, cannot be shorted, have no transparent price history — is precisely the problem that real-world-asset infrastructure claims to solve. But tokenization does not eliminate the underlying risks. It converts them. A card that trades instantly acquires the volatility of a crypto asset while retaining the custody risk of a physical object and the legal exposure of an unlicensed IP derivative. The bottleneck is not the chain. It is the license and the valuation mechanism.
So here is the question worth carrying forward. When the next comparison chart lands in your feed — some alternative asset, some multiple, some mega-cap reference point — will you ask what starting date was chosen, what got excluded from the basket, how often the price was actually quoted, and what the round-trip cost was? Or will you take the ratio at face value because it was printed next to a familiar ticker?
Don't trust. Verify the stack.