Future winners in crypto may no longer come from finding "the next token nobody has heard of." In Jason Kam’s view, the better trade is to identify the winner that can hold share over time inside a market that has already been validated.
That was the central argument from the Folius Ventures founder in a conversation with 168X on Oct. 21, 2026. Kam said crypto has not disappeared, but the industry has moved from a Wild West phase into a fintech phase, with the biggest opportunities narrowing to two themes: putting everything on-chain and bringing dollars into everything.
Folius has largely stopped making new primary crypto bets
The discussion, hosted by Mr. Z of 168X, opened with how Kam’s investment framework has changed since Folius was formed in a very different crypto market.
Kam said Folius now effectively has two parts. One is the legacy fund. The other is closer to his own holding company or a small family office, with some outside capital as well, and the flexibility to look across equities and crypto.
Since January and February this year, he said, roughly 80% to 90% of his time has gone to AI, especially public markets. On the crypto side, the work is now more about managing existing positions.
He also said Folius has made almost no new primary crypto investments since September 2024. He gave two reasons. First, the lifecycle of the old token factory model has run its course: a small team forms a project, raises money, launches a TGE, lists with low float and high FDV, then waits for unlocks. In his words, that setup makes less and less sense.
Second, Folius’ fund structure allows LPs to redeem quarterly. Kam said that kind of structure is not a natural fit for early-stage investments that may take five or eight years to exit.
As a result, he described the current period as a harvesting phase. The focus is on realizing value from what has already been funded and managing liquidity, not forcing new private deals just to keep up the image of being a VC.
Kam does not expect "crypto VC" to exist by 2030
Asked whether crypto VC will still exist as a category in 2030, Kam’s answer was blunt: "I think there definitely won’t be crypto VC in the future."
He said liquid funds and prop funds will still exist because crypto still has its own market structure, on-chain data, positioning analysis and historical patterns that can create an edge. But for firms built around the traditional model of primary crypto venture investing, the path looks much harder.
His reasoning is that as crypto becomes more formalized, the real competition shifts to firms such as Sequoia, Founders Fund and Union Square Ventures. In that environment, he asked, why would a strong founder have to take money from a boutique crypto fund?
Kam said crypto funds are likely to evolve in one of three directions:
- move into liquid strategies and OTC, becoming more like public market investors;
- become fintech VCs focused on stablecoins, payments and neobanks;
- open the mandate completely and invest across AI, robotics and biotech.
He added that firms such as Paradigm and Framework are already moving in that direction. In his view, "I understand crypto" on its own will not be enough as an investing skill set.
Crypto’s opportunity set has narrowed to two major tracks
When the conversation turned to what is left if the stories around infrastructure, Layer 2s and new Layer 1s keep fading, Kam said the answer is already well defined.
He reduced it to two things.
The first is putting everything on-chain and giving financial products a more open, less permissioned trading venue. That leads to what he described as an everything exchange: trading, speculation and a broader set of assets entering on-chain markets.
The second is the U.S. dollar as a value carrier reaching every corner of daily life around the world. From that, he said, come stablecoins, payments, payment processors and neobanks.
Kam argued that the market spent years talking about sharding, zero-knowledge systems and different application layers, but most of those stories died. One core reason, he said, is that users do not care about the technology itself. Too many products were built like a hammer looking for a nail.
Trading and money are different. Global financial assets are an enormous TAM, and global payments are another huge market. Crypto’s penetration in both remains low.
So his conclusion is not that crypto has lost its chance. It is that the opportunity set has compressed from 100 vague tracks into two very large ones.
Stop hunting for the next 100x and buy the winners
Kam said his investment approach has become much simpler: "Buy the winners."
In areas such as stablecoins, exchanges and payments, he said, the leaders are becoming easier to identify. Investors can still study new startups, but there is no reason to assume a newly formed team will outperform an incumbent that already has distribution, liquidity, engineering strength and users.
What matters more is whether a winner’s market cap is expensive or cheap relative to adoption.
He gave several examples. Robinhood can be evaluated through market cap relative to assets under custody. Hyperliquid can be looked at through valuation relative to TVL and volume. Payments companies can be judged by the transaction volume they process and how deeply they have penetrated their user base.
Kam said the market value of these companies tends to swing around adoption. In bubble periods they can become absurdly expensive. In downturns, the market can act as if they will never grow again. The real alpha, he said, may not be discovering the winner. It may be buying the winner when the market temporarily hates it.
Founders still have room, but not by copying the leader
Kam did not say entrepreneurship is over. He said founders still have a chance, but not by building the same thing as the number one player.
Mr. Z used HIP-3 as an example. He said he initially thought the design was very smart because it allowed different teams to build their own market on Hyperliquid. But in the end, TradeXYZ took most of the trading volume. That outcome, he said, was not only about mechanism design. Founder relationships, early ecosystem position and liquidity also mattered.
Kam’s point was that capital has inertia and stickiness. Once a positive loop forms, the second player has a hard time winning simply by offering the same product.
That is why founders need a genuinely different edge. If others are building traditional trading, maybe the opening is AI plus intent trading. If others are focused on pure trading, maybe social can become a new trading interface.
For Kam, finding a new entry point inside a proven market matters more than inventing a new market that nobody needs.
Bitcoin and the rest of crypto have split apart
On the next bull market, Kam said Bitcoin has to be separated completely from everything else in crypto.
His framing was simple: BTC is a digital gold trade. The rest of crypto is fintech.
He said Bitcoin used to function like the sun, with the whole industry orbiting around it. But assets such as Hyperliquid now show less correlation with BTC than in the past and increasingly trade more like U.S. equities or fintech adoption stories.
Kam also said BTC is moving away from the old four-year halving framework. Annual new supply is now very small, and compared with ETFs and global dollar liquidity, the marginal supply-demand effect of halving is getting weaker.
That leaves Bitcoin looking more like gold, rising gradually with the dollar cycle and global asset allocation. Other tokens, by contrast, can start to resemble equities that trade 24/7 if they have real cash flow, value accrual, credible management and growth.
His conclusion was that investors should stop using the same framework to value BTC and HYPE. In his view, they are now two different asset classes.
For the next 12 to 18 months, buy BTC on dips but watch three major risks
Kam said that since June this year, his base case for Bitcoin has been to accumulate on weakness over the next 12 to 18 months. If the market falls, buy gradually. If it rises, hold. If it rises too much, he said he might sell some calls.
He argued that after several cycles, the market has a rough sense of what a cycle bottom looks like. Long-term holders at these levels do not have a strong reason to sell unless they are forced to.
Still, he listed three "ghost stories" that could trigger a major BTC drawdown:
- extremely strong AI productivity. If AI pushes productivity close to unlimited growth, capital may chase productivity assets instead of a non-cash-flow asset;
- Q Day. A sudden breakthrough in quantum computing could create major FUD around the security of current cryptography. Even if technical fixes or a fork are possible, the market would likely trade the fear first;
- a real financial crisis and forced liquidation. If the AI cycle breaks and causes a liquidity event, BTC would likely fall with other risk assets.
Outside those three scenarios, his approach stays the same: if it drops, buy slowly.
AI could become the biggest liquidity source for the next crypto cycle
Kam said the crypto cycle is likely to be shaped more and more by the AI cycle because the wealth effect from AI is so large.
He offered a rough calculation. Add up the potential IPOs and liquidity events across Silicon Valley AI companies over the next two to three years, then assume employees eventually allocate 2% to 15% of those assets into crypto. On that basis, he estimated a potential inflow of $10 billion to $60 billion.
He said that figure is large enough to be compared with the cumulative net inflows into spot BTC ETFs since launch. And that estimate only covers employee liquidity. It does not include the broader wealth effect created by rising AI stocks.
His point was that strong AI performance does not mean crypto immediately becomes the best place for capital, because money is still chasing AI. But once AI wealth starts to be realized, part of it may end up in crypto.
The reverse also applies. If AI suddenly breaks down, risk assets are unlikely to be spared in the first phase. In that sense, Kam said, the biggest macro variable for the next crypto cycle may not come from inside crypto at all.
The hardest question in AI hardware is revenue per GPU
Kam said he now spends a large amount of time studying AI and semiconductors. Earlier this year, he said, the market did not understand coding agents, so buying a basket worked. Later, the market did not understand GPU demand, which also created a relatively straightforward trade. Now the key questions are much harder.
For him, the central issue is not whether token prices have fallen. It is whether revenue per GPU has fallen.
He noted that token prices have indeed been dropping, but the cost of generating a token with a GPU has also been falling quickly. Model efficiency, software optimization and chip utilization are all improving. That means lower token prices do not automatically imply weaker GPU economics.
Another variable is whether demand growth from open source and post-training outweighs demand weakness from slower growth at frontier labs.
Kam warned that by the time the market can clearly see GPU rental yields moving lower, smart money may already have traded that shift one or two quarters earlier. The top in AI, he said, will not be as simple as watching a moving average. The real turn will be hidden in revenue per GPU, channel checks and hyperscaler earnings.
He is more comfortable buying hyperscalers than chasing chips
If semiconductors are becoming harder to trade, Kam said the part of AI he is more comfortable buying now is the hyperscalers.
Mr. Z noted that as open-source models improve, enterprises do not need to run every task on the most expensive frontier model. Cloud service providers can integrate different open-source models into their cloud stacks and offer enterprise deployment.
Kam said that points to Microsoft, Amazon and Google as the companies with the clearest distribution advantage. If he had to buy AI now, he said he would look at MSFT, AMZN and GOOG, and even some cloud ETFs.
Semiconductors are harder, in his view, because expectations are already very full while the direction of revenue per GPU remains uncertain.
He also stressed that this is not a trade to buy and ignore for 10 years. The pace of open-source progress, the frontier lead of closed-source models, token prices and GPU economics are all still moving. But at this stage, the rotation logic from hardware to cloud service providers looks clearer than blindly buying every chip name.
In China AI, Kam would back DeepSeek, Tencent and ByteDance
Looking at China over the next few years, Kam said the three names he most wants to back are DeepSeek, Tencent and ByteDance.
What attracts him to DeepSeek is the clarity of its strategy: make the model as open as possible, as cheap as possible and as strong as possible, then push cost-performance to the limit.
With Tencent, he said the company does not necessarily need to build the strongest frontier model in the world. Its real assets are WeChat, gaming and ad distribution. AI can improve ad efficiency and cut game development costs, while the monetization channels remain firmly in Tencent’s hands.
ByteDance, he said, is a different case. It has strong consumer distribution and is also one of the most aggressive companies in China when it comes to compute investment.
Kam’s broader point was that China does not need to surpass the U.S. in frontier models first in order to create major AI equity value. The model is the capability. Distribution is the machine that turns capability into money.
ByteDance’s hidden alpha is not just Doubao
Kam said the most underappreciated part of ByteDance is not simply Doubao.
The first piece is video generation. If ByteDance’s video generation model keeps improving, he said, the gap between AI video and traditional film production could narrow sharply over the next year or two.
In the past, a piece of content needed actors, cameras, lighting, post-production and a full production team. In the future, he said, a few people plus a large amount of compute may be enough to produce long-form and short-form video, changing the entertainment cost curve.
The second and more important point is that ByteDance is already an app factory. Kam said the company is very good at product development, growth and distribution, and TikTok has already shown it can push products to a global audience quickly.
If AI keeps lowering the cost of building apps, then companies that can repeatedly produce apps and distribute them to hundreds of millions of users may become even more valuable. In his framing, ByteDance’s edge is not just that it has a model. It has model, data, distribution and a growth engine.
AI plus crypto should not be forced into a narrative
On where AI and crypto truly intersect, Kam was cautious: "Nobody knows."
He said many serious AI researchers do not care about crypto at all. Instead, it is often crypto-native people trying to figure out how to attach AI to a token.
That does not mean there is no opportunity. Kam said areas such as privacy, private inference and Venice have real demand. Payments, stablecoins and financial markets could also see more machine-to-machine transactions as agents become more common.
But he said he would not assume a project has value just because it labels itself AI plus crypto. The problem with the category, in his view, is that the narrative is easy to tell while product-market fit is hard to find. Crypto only deserves a place if it actually makes an AI product better, not if AI is inserted just to justify a token.
In a K-shaped downturn, PDD plus agents could be defensive
Kam said he looks at A-shares, Hong Kong stocks and U.S.-listed Chinese stocks separately because each has its own capital flow issues, but some individual companies are extremely cheap.
He said he personally holds a position in PDD. The appeal is not only valuation. He thinks the company may have a natural advantage in a K-shaped downturn.
If consumers become more price-sensitive, and agents help users search across the internet for the lowest price and highest-efficiency product, then a platform such as PDD, which has already optimized the low-cost supply chain, could be strengthened rather than weakened.
Kam pointed back to the 2007 to 2009 U.S. financial crisis, when dollar stores did not collapse in the same way as higher-end consumption. So if AI eventually drives a split in employment, income and consumption patterns, low-cost distribution plus agents could offer an interesting form of downside protection.
The three things he most wants to study over the next 12 months
At the end of the conversation, Kam named the three areas he most wants to study over the next 12 months.
First, he wants to understand who will capture the profit from cheaper intelligence. As tokens get cheaper and AI capability becomes more like a commodity, someone will buy that cheap intelligence and convert it into large enterprise value. He said that could be private equity, software companies or a completely different business model.
Second, if AI weakens in the medium term because revenue per GPU falls, what should investors buy then? Kam said many people know how to be long AI today, but far fewer have thought seriously about who wins in the post-AI cycle.
Third, he said his conviction is strong that AI will unlock robotics and the physical world. The next market phase will take robotics more seriously. The question is not whether robotics becomes a narrative, but which company can turn AI intelligence into physical productivity.
So for Kam, the real research agenda for the next 12 months is this: who captures cheap intelligence, who wins after the AI cycle, and who takes robotics.
168X’s five takeaways from the interview
Mr. Z closed the episode by reducing Kam’s framework to five points:
- traditional crypto VC is losing its reason to exist, and what remains may be fintech VC, liquid funds or investment platforms that can work across AI, robotics and public markets;
- crypto’s biggest opportunities have narrowed to two tracks, putting everything on-chain and bringing dollars into everything, so buying proven winners in weak periods may matter more than chasing 100 new stories;
- BTC is separating from the broader crypto fintech ecosystem and becoming an independent digital gold trade, with AI wealth effects, Q Day and liquidity stress more important than halving over the next 12 to 18 months;
- AI’s biggest effect on crypto may not be AI agent tokens, but how much wealth AI creates and how much of that wealth eventually flows into crypto;
- the next stage of alpha may be less about whether AI keeps going up and more about who monetizes cheap intelligence, who wins after the AI cycle and who wins in robotics.
The conversation ended with Mr. Z thanking Jason Kam for joining 168X for a discussion that ranged across crypto VC, Bitcoin, AI, semiconductors and Chinese technology stocks.
