External Innovation and Internal Dilemma: Crypto Navigating in the Fog

Author: YBB Capital Researcher Zeke

I. AI is Inevitable, but Crypto x AI?

At the beginning of 2025, the DeepSeek developed by Fanrong Quantitative Institute dropped a "nuclear bomb" in the AI circle. A Chinese AI model, using only 2048 NVIDIA H800 GPUs and costing $5.58 million (about one-tenth of Meta's cost), directly competed with GPT-4o and Llama 3.1 in benchmark tests like MMLU and GPQA, even slightly surpassing these top models from Silicon Valley in dimensions such as complex reasoning and Chinese semantic understanding. The U.S. has imposed a chip blockade on China for many years, yet on the computational power sandbox of DeepSeek, it encountered dramatic deconstruction. Chinese AI, confined within its own walls, has finally developed a technological path that aligns with national conditions while rivaling cutting-edge technologies, with a set of open-source, low-cost, and homogeneous combinations directly breaching the computational moat of the U.S.

Chinese tech products, which have always been slightly inferior to American ones, often give me a stereotype of being cheap and plagiarized. I believe this is also the intuitive perception of most people regarding Chinese internet companies' products. However, this time DeepSeek is indeed different. I won't discuss whether the user experience is superior to ChatGPT, as that is a highly subjective feeling. Just from the tense reactions frequently mentioned in American politics and by tech giants, it is clear that China is no longer playing the role of a technology follower. The global shock triggered by this is also quite strong.

Although my wallet was the first to be affected, and to some extent, this is indeed due to a misjudgment of traditional AI development, I still want to share my views on the impact of DeepSeek on Crypto.

1. NVIDIA is the most severely injured in this incident. First, the demand for AI computational power has been questioned, and second, NVIDIA's unified computing architecture "CUDA" has been bypassed. Friends familiar with the AI field should be no stranger to CUDA, which is one of the important cornerstones driving modern AI development. When large model developers use NVIDIA GPUs, they generally rely on CUDA for research and development. Using CUDA has lower requirements for developers because it encapsulates some functions, allowing users to ignore many details, but it certainly sacrifices execution efficiency.

Since CUDA is a general-purpose programming framework, it leads to a loss of flexibility when training models. DeepSeek's approach is to directly use PTX (the intermediate instruction set framework designed by NVIDIA for GPUs) to bypass hardware limitations on training speed, shortening training time. While other models take 10 days to train, DeepSeek can complete it in 5 days. This also means that if DeepSeek intends to adapt to domestically produced GPUs in the future, it will be more adept at hardware adaptation, which could shake NVIDIA's position as the leader in AI chips. (This paragraph is sourced from a report by Korea's Mirae Asset Securities analyzing DeepSeek's training process.)

In addition to the stock price drop having a strong impact on the crypto market, which is closely related to U.S. stocks, I personally believe that in the long run, this is actually a good thing for decentralized computational power projects. First, more personal GPUs can also play a role in the future. Secondly, if DeepSeek's small but beautiful open-source model path can succeed, it will force many AI companies to open-source their projects. The demand for computational power needed for local deployment and secondary development will become increasingly strong. Looking at the hardware requirements of DeepSeek R1, from the minimum parameter of 1.5B to the maximum parameter of 70B, there are opportunities to contribute excess computational power from the lowest NVIDIA GeForce GTX 1660 Super to the 40 series, 50 series, and then to professional-grade A100 and H800 GPUs. For currently transparent decentralized computational power projects, which have somewhat limited supply, this may be a chance for a turnaround (of course, the premise is that the latency can be kept low).

2. AI framework projects were the hottest emerging track in Crypto before DeepSeek dropped its "nuclear bomb," and it was the last track I wrote about before the Spring Festival. Now, under the bombardment of DeepSeek, they are almost all converging towards zero. After all, others can compete with OpenAI with less than $6 million in costs, while our leading projects with FDV in the billions have yet to produce anything that can be called a practical AI Agent.

Since the inscription era, we have been almost obsessively pursuing assetization. The current level of tolerance for assetization in the crypto circle is quite open; completely off-chain AI framework projects only need an open-source GitHub repository and a social account to issue tokens. The cost of such "library-based tokens" is that one must be prepared to accept a dimensionality reduction attack from traditional AI companies one day.

In the golden age of AI development, the big moves that traditional internet companies can make will not be limited to just this DeepSeek. The development of AI in the U.S.-China game will only accelerate, and the key question is how Crypto can integrate with which directions in the upstream and downstream of AI to highlight the advantages of decentralization without being killed by an inexplicable AOE. Broadly dividing the technology stack of Crypto x AI, we can categorize it into computational power layer, data layer, middleware layer, and application layer. In today's layered structure, I still feel that there is no necessity for Crypto's existence. If we consider privacy and security from a future perspective, it may be a good angle. After all, AI Agents replacing or assisting human work has already become a reality today. How to ensure the privacy of work materials and personal data processed by AI may be a problem that traditional internet companies cannot solve. Furthermore, if an AI Agent has payment permissions, how to ensure the security of the wallet will also become an issue. Using blockchain as the compliance audit layer for AI models should be our main development direction in the future.

On the other hand, there is the question of where incentives should be directed. Through incentives, we can not only stimulate the computational power layer and model sharing but also teach AI how to interact with the virtual world. Unlike LLM training, which naturally possesses decades of global internet corpus knowledge, teaching AI how to act correctly requires humans to continuously label the correct positions. It is similar to teaching a visual model to recognize what an animal is and what a car is, and this cannot be accomplished simply by outsourcing to a group of college students. Creating an AI Agent that can interact with the virtual world requires a complex decentralized network, teaching AI through countless individual humans, which is also a direction. I have discussed this in more detail in my previous articles. What else can be done around incentives? Combining with DePin to teach agents to interact with reality, incentivizing AI to gain attention, incentivizing secondary creation related to AI (for example, Bittensor's incentive model is a good example), and having the token incentive mechanism automatically adjusted by AI (this stems from a question I raised in a previous article, which was: when a decentralized project becomes immensely large and enters the mainstream, how should it handle deflation and inflation? Should it rely on simple code rules, or listen to a few people or dozens of project members? Or those key figures? Oh, right, we also have governance tokens. However, governance tokens are meaningless before solving the witch problem. Democratic voting can never reflect in governance proposals. After all, a16z can veto a large community's approval votes with just a few wallets, so what is the point of voting? And so on.

We indeed cannot aggregate a group of high-end AI talents like traditional internet companies, nor can we purchase or rent a large-scale GPU cluster for training. Recreating a DeepSeek in the blockchain is naturally a pipe dream. The significance of Crypto lies in endowing another field with irreplaceable decentralized characteristics, just as we once granted financial freedom. AI is an inevitable narrative for humanity, but what role can Crypto play in it?

3. This is the first time I mention Wordcoin in my article. This crypto utopia project initiated by Sam Altman seems somewhat absurd even now. Whether to input iris data means you need to make a choice between state surveillance and corporate surveillance, which feels like being in The Matrix, choosing between the red pill and the blue pill.

However, the concept of universal basic income or inclusive finance may no longer be a joke at this stage. The ability to locally deploy a model that rivals cutting-edge models has already allowed agents to appear in hospitals and government units in China. According to McKinsey's 2024 forecast report, 50% of jobs may be replaced by AI within the next six years. Future versions of Wordcoin may even be uniformly distributed by the government. If this process continues to intensify, related tokens for inclusive finance may emerge in large numbers and be repeatedly speculated upon. Considering a five to six-year window, this aligns with Trump's term, so will this crypto president issue a similar token? I think it is highly likely.

4. According to recent remarks by Elon Musk, AI may sweep the Nobel Prize for 25 years. Therefore, raising funds through blockchain (even contributing computational power, storage, methods, and various resources) to promote AI research may be more interesting and effective than the current DeSci. Perhaps I can call it Decentralized AI Science? That is DeAIS.

II. Meme Coin is No Longer Meme

In the past, when we analyzed Meme Coins, we talked about subculture, community consensus, and dissemination effects. Now, sitting in front of GMGN, I analyze conspiracy groups, head addresses, and developers' insider trading every day. When a CA is sent to various dog coin groups, it is the charge moment. Today's memes are more outrageous than ever before; at this stage of Pump.fun, not to mention finding a token that allows you to sleep soundly, you might just go to the bathroom, and the K-line could plummet.

The continuous simplification of asset issuance thresholds and the high anonymity of blockchain have fueled this crazy casino culture. An unknown ragtag group can treat the crypto circle as an ATM. The evolution direction of memes has also become increasingly arbitrary; not only the aforementioned library-based tokens but any event, any person, or even any AI can become a token. Without a cultural core and consensus, the so-called leading projects can be completely forgotten in just a few weeks. The celebrity coin craze that started with Trump lasted only a month. Under a tweet from President Milei, hundreds of millions of funds flowed from Sol back to the outside world, and thus memes began to recede. President Milei's response was simple: he deleted the tweet and replied to everyone, "I was not informed."

The rapid development of AI has taken away too much attention from this world, making it difficult for tech enthusiasts to make progress. Retail investors who abandon value investing can only gamble in a pig-killing plate, hoping to be among the lucky few. The increasingly scarce liquidity is repeatedly extracted, reflecting in the daily red K-lines in Cex and Dex, as well as the disdain of traditional capital and outsiders towards altcoins.

III. Carving the Boat Cannot Retrieve the Sword

The law of cycles has clearly failed, and all the ideas of carving boats to retrieve swords are futile. A bull run in BTC does not mean a bull run in altcoins, but a bear run in BTC certainly means a bear run in altcoins. Our understanding of altcoins must be reformed; the altcoin market is no longer one that can be supported by a white paper. A large project listed on a top Cex must be mature enough to support its own token price.

Looking back at the growth of tokens over the past seven years, there were less than 2000 tokens listed in the 2017 market, while the number of tokens listed in 2024 is approaching 25,000 (data sourced from CoinGecko, including delisted tokens). The exponential growth in the number of tokens is essentially an irreversible evolution from a low-entropy value system to a high-entropy noise system in blockchain. When each token in 2017 still carried the ideal of "disrupting the world," the tokens of 2024 have evolved into chips for liquidity exit. The emergence of more tokens has not brought more innovation and implementation, but the high valuations of star projects have exponentially increased the demand for market liquidity.

As mentioned above, without recognition from the external world, retail investors cannot support the valuations of these projects. Most altcoin listings often represent historical peaks, and Binance is the last stop. The crypto circle needs a revolution; star projects should justify their massive financing. Bybit's trial of publicly disclosing project financial reports may be a solution. However, in my personal opinion, the market needs a deep bear phase to reshape the valuation system and listing standards of current altcoin projects.

IV. Confusion

I once saw a glimmer of hope on Ton, believing that the beginning of consumer-grade applications in Crypto had arrived, but this brief light faded away with the trend of Tap to Earn. Five years ago, the liquidity mining derived from DeFi brought the crypto circle to an incredibly glorious peak. Five years later, the only successful field we have is still just DeFi.

The topics I discuss with people in the circle are very simple: Did you buy BTC? Did you short it? Got a CA? Everyone is confused; we can no longer find a correct direction. Besides BTC, buying any token does not allow for a sound sleep, which is the current state of the crypto circle. "Diamond Hands" is not a compliment today; if you don't buy BTC, it feels more like a synonym for being foolish.

When I open various chain media on my phone, it feels like reading The New York Times and gossip media. Various phenomena reflect that the hopes of this circle are mostly pinned on policies and attention. From a VC's perspective, we may only invest in tool-type products in the future, with asset issuance platforms becoming shovel sellers and rent collectors to survive.

Conclusion

Clearly, this is not the state we wish to see. Although Crypto currently seems to be lost in a fog of direction, the success of DeepSeek proves that innovation remains the most effective path to break through the deadlock. Crypto has the best policy environment, attention, funding, and solid infrastructure in history. In the near future, there will be many altcoin ETFs that can inject liquidity back into the crypto circle. We are clearly in the mainstream, yet trapped within our own walls. The retreat of Meme Coins may hide a turning point; the future of humanity may not only be AI.