DeepSeek: Hello Crypto, big hammer 80, small hammer 40, what kind of hammer do you want?

In January 2025, the launch of DeepSeek R1 shook the AI community, and it also truly changed the Crypto AI ecosystem. In the past cycle, Crypto AI primarily revolved around AI agents, while DeepSeek R1 and its open-source strategy completely changed the rules of the game: extremely low training costs and breakthrough adaptive training methods made the vision of a decentralized AI industry no longer a mere talk, but an achievable reality. This transformation has far-reaching implications. The total market value of the Crypto AI market has significantly shrunk, with many AI tokens experiencing a 70% pullback. But is this really a crisis? Or does it signify a complete reshuffling of Crypto AI? Is DeepSeek the 'terminator' that shatters the Crypto AI narrative, or is it the 'breaker' that accelerates its entry into the practical era?

The barbaric growth of DeepSeek

The development of DeepSeek can be traced back to 2021. At that time, the quantitative trading hedge fund Huanshang began to recruit AI talent on a large scale, which was rare for quantitative companies to transition to AI. Most of the recruits were researchers exploring cutting-edge directions, encompassing fields like large models (LLM) and text-to-image diffusion models. Although there were rumors that Huanshang made this transition to better utilize the idle GPU resources within the company, the main reason should still be to seize the commanding heights of cutting-edge AI technologies such as large models.

By the end of 2022, Huanshang had attracted more and more top AI talents, mainly students from Tsinghua and Peking University. Stimulated by ChatGPT, Huanshang CEO Liang Wenfeng was determined to enter the field of general artificial intelligence and established DeepSeek in early 2023. However, the rapid rise of AI companies such as Zhipu, Moon's Dark Side, and Baichuan Intelligence, coupled with DeepSeek's nature as a pure research institution lacking star founders, made independent financing extremely difficult. Therefore, Huanshang chose to spin off DeepSeek and fully fund its development. Although this decision carried high risks, DeepSeek was free from the profit commitments or valuation pressures of financing parties. Meanwhile, it had relatively ample GPU resource reserves, allowing the team to focus on technological breakthroughs, as a group of innovative young people could charge ahead in a nurturing environment. At this moment, DeepSeek resembled more of a research institute than a company.

Just like in the early days of OpenAI, no one would have thought that a company researching robotic hands playing Rubik's Cube could eventually develop ChatGPT. Similarly, no one could have imagined how Huanshang, a quantitative trading company, could use DeepSeek to break through the current AI bubble; the former took 7 years, while the latter took only 2 years. In November 2023, DeepSeek LLM with 67 billion parameters and performance close to GPT-4 was launched. In May 2024, DeepSeek-V2 went online, and in December of the same year, DeepSeek-V3 was released, performing comparably to GPT-4o and Claude 3.5 Sonnet in benchmark tests. This rapid technological leap by DeepSeek is not due to the company's financial resources or high academic qualifications but is a result of a technological singularity occurring after 'ChatGPT influenced the world AI industry.' Various singularities are accelerating in any soil capable of meeting imagination until the next key singularity appears.

Finally, in January 2025, DeepSeek accelerated past the Singularity, using their cultivated first-generation reasoning-capable large model DeepSeek-R1, which opened that door at a training cost far lower than ChatGPT-O1 and with outstanding performance.

Using open-source to distribute the key to open the Stargate to the world.

The day after DeepSeek R1 was released and the open-source model was announced, U.S. President Trump officially announced the start of a $500 billion mega investment 'Stargate' program at a White House press conference. A joint venture named Stargate was established by OpenAI, SoftBank, Oracle, and investment firm MGX to build new AI infrastructure for OpenAI in the United States.

Such a level of investment is comparable to the 'Manhattan Project', aiming to use national strength to push closed-source AI to its peak through algorithm stacking, monopolizing the AI market to ensure the leading position of the U.S. domestic AI industry. However, at the time of the plan's release, it is unlikely they anticipated that just days later, this open-source model across the ocean would not only refuse to open the door but would also bring a hammer to smash through the wall, while simultaneously giving others hammers.

As an open-source model that can rival top closed-source models, DeepSeek's new training architecture has triggered a chain reaction, making it increasingly difficult for closed-source AI to compete. Closed-source models that cannot keep up with DeepSeek R1 will be directly eliminated from the capital market. Even Marc Andreessen, the founder of A16z (OpenAI's investor), has publicly stated that we need to focus more on open-source AI rather than emphasizing closed-source AI. In the industry, whether supporting the potential emergence of AGI or viewing AI solely as an upgraded version of the SaaS industry, it is widely believed that the harms of closed-source far outweigh those of open-source, whether due to black boxes, industrial monopolies, information security, or capital manipulation of attention — any of these are very dangerous development directions.

Although some industry insiders suspect that the mixed expert technology 'MoE' of V3 requires a huge dataset and is distilled using OpenAI's model, and that the reinforcement learning 'RL' method in R1 requires a large amount of hardware resources, leading to suspicions of fraud in the number of training chips used, it does not affect the structural reforms it brings to the industry.

The open-sourcing of DeepSeek R1 has broken the closed-source large model commercial logic of OpenAI in terms of training architecture, using a logic that allows models to self-evolve to avoid traditional paradigms' large investments in computing power and data labeling. Although training models is still like opening a blind box, the cost of the blind box has been significantly reduced.

At the AI hardware level, the open-sourcing of DeepSeek V3 directly challenges Nvidia's market dominance. The moat surrounding Nvidia's GPUs largely lies in its bottom parallel computing platform and programming model, CUDA. Its extensive ecosystem and sufficient developers make the costs of using non-Nvidia chips for training prohibitively high, while high purchasing thresholds and political restrictions create fragmentation in global AI development.

For us, in the short term, the AI sector in the U.S. stock market has shrunk significantly, while the total market value of Crypto AI has nearly been slashed, and the market has entered a bear phase. However, in the long term, the most recognized AI industry is moving towards an open-source, transparent, and decentralized development path. From any perspective, the integration of Crypto and AI will become more harmonious.

The redemption of Crypto AI, move forward! Move forward! Advance by any means necessary.

During this round of the Crypto AI bubble burst, many AI concept tokens experienced a 70% pullback, and the Crypto AI market shrank significantly. Some jokingly remarked, 'You can train a large model for just $5.5 million; with these AI market caps, why buy Crypto AI at all?' Indeed, Crypto is a market dominated by capital rather than products, and 90% of AI tokens lack practical significance.

However, with the improvement of the regulatory system in the crypto market, the crypto market remains the most suitable soil for small and medium-sized AI companies to start their businesses. The model cost from DeepSeek, which is 1/100 compared to ChatGPT O1, along with its model training methods, will drive ecological growth of over ten thousand times compared to the current market.

Specifically, what DeepSeek brings to crypto is a decentralized training model, which rationalizes Depin-type projects, makes the training process and information feeding more transparent, and establishes a more reasonable value reward mechanism for data set contributors, simplifying the settlement for both supply and demand sides of model training. The surrounding ecosystem of the AI industry, which is worth more than ten thousand times, has further enriched the downstream industries of Crypto AI. When enough competitive and creative product narratives appear in the market, as soon as one truly breaks through, external funding will naturally flow back into Crypto. The market has long suffered from PVP, and a series of celebrity coin harvests after TrumpCoin disrupted the originally abundant liquidity and positive feedback balance of the AI market, so the bubble burst by DeepSeek is actually a greater boon.

Currently, many Crypto AI projects have either quickly integrated DeepSeek or updated their architecture based on it, including ElizaOS, Argo, Myshell, Build, Hyperbolic, Nillion Network, infraX, and others. Some of these projects have directly optimized their products using DeepSeek.

Myshell

In the production flow of chatbots and application plugins, V3, R1, and even the image generation model Janus-Pro were integrated by Myshell's technicians in almost half a day's time. As one of the few projects in the blockchain space that consistently focuses on refining products, even gaining fame in Web2AI products and hesitating to issue tokens, this time, the open-sourcing of DeepSeek will bring good news to Myshell users on the cost side, as lower costs will attract more agent developers for the already well-developed Myshell.

Argo

Sam Gao, a developer of Argo, incorporated key functionalities of Argo into DeepSeek from the early stages of product design. As a workflow system, Argo integrates LLM as a standard DeepSeek R1 and delegates the generation of the original workflow to DeepSeek R1. Because of the WorkFlow aspect, the token consumption and context information volume will be huge (average >= 10k Tokens), and Argo also integrates CoT 'Chain-of-Thought' into the WorkFlow thinking process. After the open-sourcing of DeepSeek, not only did it reduce the cost of workflow products, but it also allowed for local deployment of LLMs in Argo, ensuring user privacy and security.

Before the release of DeepSeek R1, Argo had already integrated its model's preliminary training logic Chain-of-Thought 'CoT' into the production process of Argo's Agent Workflow. Especially for tasks like meme trading and market trend analysis, Argo customized its workflow using Graph-of-Thought (GoT), a novel approach that constructs reasoning as a graph, where nodes represent 'LLM thoughts' and edges represent the dependencies between these thoughts.

Given that Argo chose GoT (the only Crypto AI Workflow currently using this model), it achieved a more reliable and transparent process. This innovative method directly impacts the security and trust of transactions on the Argo platform. Integrating the thought graph (GoT) into Web3 AI agents places Argo at the forefront of AI crypto trading. The structured reasoning of CoT not only enhances the security of financial transactions but also ensures transparent and reliable decision-making, which is crucial in decentralized finance (DeFi).

It is worth noting that Sam, a core developer of Argo, collaborated with Shaw on a paper (EraseAnything: Enabling Concept Erasure in Rectified Flow Transformers) about how to remove unwanted concepts from large-scale text-to-image diffusion models without compromising the overall generative performance of the model, with the help of DeepSeek researcher Xingchao Liu.

Hyperbolic

Hyperbolic Labs also took the lead in announcing the hosting of the DeepSeek-R1 model on its GPU platform, allowing users to rent Hyperbolic GPU resources to run the DeepSeek-R1 model locally or at designated data centers without sending sensitive data to DeepSeek's servers. This method ensures data privacy while leveraging the excellent reasoning performance of the DeepSeek model. Furthermore, through Hyperbolic's decentralized computing network, users can obtain the efficient reasoning capabilities of the DeepSeek model at a lower cost, making it a very competitive solution for startups, super individual entrepreneurs, or simply efficient AI users.

This round of bubble burst has indeed dealt a heavy blow to the Crypto AI market, and many AI tokens have lost their speculative value. But essentially, DeepSeek is not eliminating Crypto AI; rather, it is forcing the market to accelerate its evolution. After DeepSeek R1, the future of Crypto AI will no longer solely rely on speculation but will need to be restructured around decentralized AI computing, economic incentive mechanisms for model training, fair distribution of AI resources, and practical products.

This is not an end, but an evolution. Crypto AI needs to move faster and more aggressively. / Accelerate