External innovation and internal dilemma: How cryptocurrencies navigate uncertainty

Author: Zeke, a researcher at YBB Capital

1. AI is inevitable, but what about Crypto x AI?

At the beginning of 2025, a "nuclear bomb" dropped by DeepSeek, developed by Magic Square Quantitative, shocked the AI ​​world. China only used 2048 NVIDIA H800 GPUs to train its AI model, which cost $5.58 million (about one-tenth of the cost of Meta) to face GPT-4, Llama 3.1, etc. in benchmarks such as MMLU and GPQA, and even slightly surpassed in complex reasoning, Chinese semantic understanding and other fields. Although the United States has imposed a chip blockade on China for many years, DeepSeek's computing power has greatly broken the US monopoly. China has embarked on a technical path that suits its national conditions but is not inferior to the world's top models. This open source, low-cost, and homogenized approach has broken through the US computing defense at an astonishing speed.

For many years, Chinese technology products have lagged behind in performance and have been labeled cheap and low-end. I believe this is the intuitive impression of many people on Chinese Internet companies. But DeepSeek is different. Putting aside the subjective feeling of whether it has surpassed ChatGPT, the fact that this matter has been frequently mentioned by American political and technological circles speaks for itself. It is obvious that China has now transformed from a technology follower to a challenger, and its global chain effect is far-reaching.

Although my wallet was the first to be affected, which can be attributed to the misjudgment of the development of traditional AI, I still want to share my views, especially on the impact of DeepSeek on the crypto industry.

1. NVIDIA is the biggest loser in this incident.

First, the demand for AI computing power is questioned; second, NVIDIA's unified hardware and software computing architecture CUDA is bypassed. If you are familiar with the field of AI, you must know that CUDA is an important cornerstone for the development of modern AI. When using NVIDIA GPUs, large model developers generally rely on CUDA to complete their work. CUDA has lower technical requirements for developers because it has pre-packaged functions and developers don't have to worry about too many details, but this comes at the expense of execution efficiency.

Because CUDA is a general programming framework, the flexibility of model training is limited. DeepSeek's solution is to directly use PTX (an intermediate instruction set framework designed by NVIDIA for GPUs) to bypass the hardware's restrictions on training speed, thereby shortening the training time. Other models require 10 days of training, while DeepSeek only needs 5 days to complete. This also means that if DeepSeek intends to adapt to China's domestically produced GPUs in the future, the hardware adaptation process will be more seamless, and it is expected to shake NVIDIA's dominance in the AI ​​chip market. (This paragraph is excerpted from a report on the DeepSeek training process by Mirae Asset Securities of South Korea)

In addition to the possibility that NVIDIA's stock price may fall, which will severely impact the crypto market that is closely related to the US stock market, I personally think that in the long run, this is actually beneficial to decentralized computing projects. First, more personal GPUs will be able to contribute excess computing power. Second, if DeepSeek's compact open source model approach is successful, it will force many AI companies to open source their models, thereby increasing the demand for local deployment and secondary development computing power. From the hardware requirements of DeepSeek R1, from a minimum of 1.5B parameters to a maximum of 70B, from NVIDIA GeForce GTX 1660 Super to 40, 50 series, and even professional-grade A100, H800 GPUs have the opportunity to contribute excess computing power. For decentralized computing projects that are currently somewhat underutilized, this could be a potential breakthrough - provided that latency is low enough.

2. AI Framework Projects: Hot Crypto Tracks Before DeepSeek's "Nuclear Bomb"

Before DeepSeek dropped the "nuclear bomb", AI framework projects were the hottest emerging track in the crypto field and the last topic I covered before the Lunar New Year. However, with DeepSeek's breakthrough, most of these projects are now rapidly approaching zero. After all, DeepSeek is on par with OpenAI at a cost of less than $6 million, while our AI framework project, FDV, is as high as billions of dollars and has not yet produced anything that can really be regarded as a practical AI agent.

Since the birth of blockchain, people have been almost obsessed with the pursuit of assetization. At present, the tolerance for assetization in the crypto field is already very high. For AI framework projects that are not even on the chain, they only need an open source GitHub repository and a social account to issue tokens. This "codebase-based token" approach inevitably has the risk of being completely wiped out by the "two-way mooring" attack of traditional AI companies.

In the golden age of artificial intelligence development, traditional Internet companies are unlikely to stop at DeepSeek as the only weapon. Against the backdrop of Sino-US competition, the development of artificial intelligence will only accelerate, and the key issue is how Crypto can be combined with the upstream and downstream of AI to highlight the advantages of decentralization without being knocked down by unexpected AOE. Generally speaking, we can divide the Crypto x AI technology stack into four layers: computing power, data, middleware, and applications. In the current layered structure, I don't see the necessity for Crypto to participate. But from a future perspective, privacy and security may be strong points, as AI agents have become a reality of replacing or assisting human work. Ensuring the privacy of work and personal data processed by AI may be a challenge that traditional Internet companies cannot solve. In addition, if AI agents have payment permissions, ensuring the security of wallets will become a problem. Using blockchain as a compliance and audit layer for AI models may be a key direction for future development.

On the other hand, incentive mechanisms also play a vital role. In addition to stimulating computing power and model sharing, incentive mechanisms can also teach AI how to interact with the virtual world. Unlike LLMs, which have decades of global internet data, teaching AI the right behavior requires constant human annotation. For example, teaching a visual model to recognize animals and cars is not something that can be outsourced to a group of college students. To create an AI agent that can interact with the virtual world, a large decentralized network of individuals is needed to teach the AI. This is a potential direction. I discussed this in more detail in my past article. What else can incentive mechanisms drive? Combined with DePin, teaching AI agents to interact with the physical world, incentivizing AI to gain attention, incentivizing AI secondary creation (Bittensor's incentive model is a good example), or having AI automatically adjust token incentive mechanisms - these are all fascinating possibilities. This leads to another question from my previous article: When a decentralized project grows and enters the mainstream, how should deflation and inflation be managed? Is it by simple code rules, or by a few people on the project team, or by those key people? Of course, we still have governance tokens. But governance tokens are meaningless until the "sybil problem" is solved. Democratic voting will never be reflected in governance proposals, because a16z can veto the vote of a large community with just a few wallets, so what is the point of voting?

Of course, we can't gather a group of high-end AI talents like traditional Internet companies, nor can we buy or rent large-scale GPU clusters for training. Trying to replicate DeepSeek in the blockchain is a daydream. The role of Crypto is to bring irreplaceable decentralized features to another field, just as we once brought financial freedom to the world. AI is the inevitable narrative of mankind, but the key question is: What role can Crypto play in it?

3. Wordcoin: The first discussion on the crypto utopia project

This is the first time I mentioned Wordcoin in an article. This crypto utopia project initiated by Sam Altman still seems ridiculous to me. The decision to record your irises is like choosing between state surveillance and corporate surveillance, similar to the choice between the red pill and the blue pill in The Matrix.

However, the idea of ​​universal basic income or financial inclusion no longer seems to be a joke at this stage. DeepSeek's AI agents are able to be deployed locally to compete with top models, and they have begun to appear in hospitals and government agencies in China. According to McKinsey's 2024 forecast, up to 50% of jobs may be replaced by AI in the next six years. Future versions of Wordcoin may even be distributed by the government. If this trend intensifies, related universal financial tokens may appear and be repeatedly hyped. If given another five or six years, this may coincide with Trump's presidency. Will the crypto president issue such a token? I think it is very likely.

4. Elon Musk and the future of AI research funding

Given Elon Musk's recent remarks, AI may dominate the Nobel Prize in the next 25 years. I believe that blockchain-based fundraising (or even contributions of computing power, storage, methods, and other resources) to drive AI research will be more interesting and effective than the current decentralized science (DeSci) movement. Perhaps I can call it decentralized artificial intelligence science, or DeAIS for short.

2. Meme Coin is no longer Meme

Before, when we discussed MemeCoins, the focus was on subculture, community consensus, and viral effects. Now, sitting in front of GMGN, I analyze conspiracy groups, major addresses, and insider trading by developers. When CA was released between various "shitcoin" groups, it was time to start. Today's meme coins are more ridiculous than ever. At the current stage of Pump.fun, Don't expect to find a stable coin to sleep on - when you go to the bathroom, the chart may have plummeted.

The simplification of the threshold for asset issuance and the high anonymity of blockchain have pushed this casino-like culture to the extreme. An unknown person can turn the crypto market into an ATM. The evolution of meme is becoming more and more random - not just "codebase coins", any event, any person, and even any AI can become a coin. Without a cultural core and a consensus-bound community, the so-called top projects are forgotten after a few weeks. The celebrity coin craze set off by Trump lasted only a month. Under a tweet from President Milley, billions of dollars flowed back from Sol to the outside world, marking the retreat of meme. Milley's response was simple: delete the tweet and reply "I don't know."

The rapid development of AI has taken away too much of the world's attention. The development of technical factions has stagnated. Retail investors have given up value investment and can only bet that they are the lucky ones whose currency prices have been raised. The increasingly scarce liquidity has been drained again and again. The red K-line of CEX and DEX every day, the contempt of traditional capital and laymen for altcoins are all reflected in the continuous loss of liquidity.

Third, you can't find a sword by carving a boat

The law of cycles has obviously failed, and all "small tricks" cannot save the defeat. BTC's bull market does not necessarily mean that altcoins will also rise, but if BTC falls, altcoins will definitely fall. We must update our understanding of altcoins. The altcoin market is no longer a market where a white paper can support a project. Large projects listed on first-tier CEXs must be mature enough to support their token prices.

Looking back at the growth of tokens over the past seven years, there were less than 2,000 tokens listed on major exchanges in the market in 2017, but by 2024, the number of tokens had approached 25,000 (data source: Coingecko, including offline tokens). The exponential expansion of the number of tokens is essentially an irreversible evolution of the blockchain's low-entropy value system to a high-entropy noise system. When each token in 2017 was still carrying the ideal of "disrupting the world", by 2024, the token had evolved into a liquidity exit chip. The emergence of more tokens has not brought more innovation or real-world applications, but the valuation of star projects has greatly increased the demand for market liquidity.

As mentioned earlier, without external recognition, retail investors cannot support the valuation of these projects. Most altcoin projects reach their peak market price when they are listed, and Binance becomes their final destination.The crypto world needs a revolution. Star projects should prove the rationality of their large financing. Bybit's pilot project to disclose financial information may be a potential solution. But in my personal opinion, the market needs a deep bear market to reshape the valuation system and listing standards of altcoins.

Fourth, chaos

I once saw a glimmer of hope in Ton, thinking that the beginning of consumer-level applications of cryptocurrency had arrived. But this fleeting light disappeared with the fading of the Tap to Earn trend. Five years ago, the liquidity mining spawned by DeFi pushed the cryptocurrency world to an unprecedented peak. Five years later, the only field we are still successful in is DeFi.

Now when I communicate with people in the industry, the topic is very simple: "Have you bought BTC? Have you shorted it? Do you have CA?" Everyone is confused and can't find the direction. Now except BTC, buying any coin makes people sleepless. "Diamond hand" is no longer a compliment now, and not holding BTC is more like a fool.

Opening various blockchain media apps on mobile phones feels like reading the New York Times and tabloid gossip. All kinds of phenomena reflect that most of the hopes of this circle are now pinned on policies and attention. From the perspective of VC, our future may lie in investing only in tool products, and asset issuance platforms can only become people who sell shovels and collect rent in order to survive.

Conclusion

Obviously, this is not the reality we want to see. Although the current state of cryptocurrency seems to be lost in a fog of confusion, DeepSeek's success proves that innovation and reform are still the most effective way to break through stagnation. Cryptocurrency is currently in the best position ever, with favorable policies, attention, funding, and solid infrastructure. In the near future, many altcoin ETFs may inject new liquidity into the market.

We are clearly in the mainstream, but trapped in our own walled city.

Behind the retreat of Meme Coins may be a turning point, and the future of mankind may not only revolve around AI.