Author: BUBBLE, BlockBeats
In January 2025, the emergence of DeepSeek R1 caused a stir in the AI world, and it also truly transformed the Crypto AI ecosystem. In the past cycle, Crypto AI mainly revolved around AI Agents, but DeepSeek R1 and its open-source strategy have completely changed the rules of the game: extremely low training costs and breakthrough adaptive training methods have made the vision of decentralizing the AI industry no longer just talk, but a tangible reality. This transformation has far-reaching implications, with the total market capitalization of Crypto AI shrinking significantly, and many AI tokens experiencing a 70% correction. But is this a crisis, or does it mean a thorough reshuffle of Crypto AI? Is DeepSeek the "terminator" that shatters the Crypto AI narrative, or the "disruptor" that accelerates its entry into the era of practical application?
The Rapid Growth of DeepSeek
The development of DeepSeek can be traced back to 2021. At that time, the quantitative trading hedge fund Phantasy began to massively recruit AI talents, which was not a common occurrence for quantitative companies to shift to AI. The majority of the recruits were AI researchers exploring frontier directions, including large language models (LLMs) and text-to-image models. Although there were rumors that Phantasy's transformation was to better utilize the company's idle GPU resources, the main reason was likely to seize the high ground of frontier AI technologies such as large models.
By the end of 2022, Phantasy had attracted more and more top AI talents, mainly from the student population of Tsinghua University and Peking University. Inspired by ChatGPT, Phantasy CEO Liang Wenfeng decided to venture into the field of general artificial intelligence and established DeepSeek in early 2023. However, the rapid rise of AI companies such as Zhishitu, Yuezhi'anmian, and Baichuan Intelligence posed a significant challenge for DeepSeek, a pure research institution without a star founder, in independent financing. Therefore, Phantasy chose to spin off DeepSeek and fully fund its development, despite the high risk of this decision. This allowed DeepSeek to focus on technological breakthroughs without the pressure of profitability or valuation from investors, and the team of young innovators could charge ahead in this fertile ground, making DeepSeek more like a research institute than a company.
Just like OpenAI in its early days, no one would have thought that a company researching robot hands playing Rubik's cubes would eventually develop ChatGPT, and no one could have imagined how Phantasy, a quantitative company, would use DeepSeek to burst the current AI bubble, with the former taking 7 years and the latter only 2 years. In November 2023, DeepSeek released its LLM with 67 billion parameters, with performance close to GPT-4, followed by the launch of DeepSeek-V2 in May 2024 and the release of DeepSeek-V3 in December of the same year, which performed on par with GPT-4o and Claude 3.5 Sonnet in benchmark tests. DeepSeek's rapid technological breakthroughs were not due to the company's financial strength or high academic credentials, but rather the result of a technological singularity, where "the impact of ChatGPT on the AI industry" accelerated the occurrence of smaller singularities in any soil that could satisfy the imagination, until the next critical singularity emerged.
Finally, in January 2025, DeepSeek accelerated through the singularity, using the first-generation large model with reasoning capabilities, DeepSeek-R1, to open the door with a training cost far lower than ChatGPT-O1 and outstanding performance.
Using Open Source to Distribute the Key to the Stargate for the Whole World
The day after the release and open-sourcing of DeepSeek R1, former US President Trump officially announced the start of a $500 billion "Stargate" program at a White House press conference. A joint venture called Stargate, formed by OpenAI, SoftBank, Oracle, and the investment firm MGX, was established to build new AI infrastructure for OpenAI in the United States.
This level of investment is even comparable to the "Manhattan Project," with the apparent intention of using algorithmic stacking to push closed-source AI to its peak, monopolizing the AI market to ensure the leading position of the domestic AI industry in the United States. However, the plan's proponents did not anticipate that just a few days later, this open-source model from across the ocean would simply not open the door, not only bringing a hammer to smash the wall, but also distributing hammers to others.
As an open-source model that can match the top closed-source models, DeepSeek's entirely new training architecture has triggered a chain reaction, making it difficult for closed-source AI to move forward, as closed-source models that cannot keep up with DeepSeek R1 will be directly eliminated by the capital market. Even Marc Andreessen, the founder of a16z and an investor in OpenAI, publicly stated that more attention should be paid to open-source AI rather than closed-source AI. Within the industry, whether supporting the potential of AGI or supporting AI as an upgraded version of the SaaS industry, the consensus is that the drawbacks of closed-source are far greater than those of open-source, be it black boxes, industry monopolies, information security, or capital control of attention, any of which are extremely dangerous development directions.
Although some industry insiders have doubts about the large data set required for V3's Mixture of Experts (MoE) technology and the high hardware resource requirements for the reinforcement learning (RL) method in R1, which they suspect may have used OpenAI's models for distillation, this does not diminish the industry-restructuring impact brought by DeepSeek.
The open-sourcing of DeepSeek R1 has broken the commercial logic of OpenAI's closed-source large models, using the logic of self-evolution to avoid the massive investment in computing power and data annotation required by the traditional paradigm. Although the training of the model is still a black box, the cost of the black box has been greatly reduced.
At the AI hardware level, DeepSeek's open-sourcing of V3 directly challenges Nvidia's market dominance. Nvidia's GPU moat is largely due to its underlying parallel computing platform and programming model CUDA, as well as its extensive ecosystem and a large enough developer base, making the learning cost for training with non-Nvidia chips too high. The high purchase thresholds and political restrictions have also led to a fragmentation of global AI development.
From our perspective, in the short term, the AI sector in the US stock market has experienced a significant contraction, and the total market capitalization of Crypto AI has nearly collapsed, with the market entering a bear market. However, in the long run, the most widely recognized AI industry is moving towards an open-source, transparent, and decentralized development path. From any angle, the integration of Crypto and AI will become more seamless.
The Redemption of Crypto AI, Forge Ahead! Forge Ahead! Advance by Any Means Necessary
During this Crypto AI bubble burst, many AI concept tokens have experienced a 70% correction, and the Crypto AI market has shrunk significantly. Some have jokingly said, "You can train a large model for $5.5 million now, so why buy Crypto AI tokens whose market value is still so high?" It is true that the crypto market is dominated by capital rather than products, and 90% of AI tokens have no real meaning.
However, as the crypto market regulatory system improves, the crypto market is still the most suitable soil for small and medium-sized AI companies to start businesses. The 1/100 large model cost and the model training method brought by DeepSeek, compared to ChatGPT O1, will lead to more than a thousandfold growth in the ecosystem.
To put it directly, what DeepSeek brings to crypto is decentralized model training, which may make projects like Depin more reasonable, making the training process and information feeding more transparent, and the reward mechanism for data contributors more reasonable, making the settlement between the supply and demand of model training easier. The more than thousandfold development of the AI industry's surrounding ecosystem has also enriched the downstream industry of Crypto AI. As long as one truly breaks through in the market, external capital will naturally flow back into Crypto. The market has long been suffering from PVP, and the series of celebrity coin harvesting after TrumpCoin has disrupted the abundant liquidity and positive feedback balance of the AI market. Therefore, the bubble burst by DeepSeek is actually a greater positive.
Many Crypto AI projects have already integrated or are about to integrate DeepSeek, including ElizaOS, Argo, Myshell, Build, Hyperbolic, Nillion Network, infraX, and more. Some of these projects have even optimized their products through DeepSeek.
Myshell
Myshell's technical team completed the model integration in just half a day, adding V3, R1, and the image generation model Janus-Pro to their chatbot and application plugin development process. As a project that has always insisted on polishing its product and has even gained a reputation in the Web2AI product space but has been reluctant to issue a Token, the open-sourcing of DeepSeek will bring good news to Myshell users in terms of cost, with lower costs allowing for more Agent developers to join the already well-developed Myshell product.
Argo
Argo's developer Sam Gao integrated DeepSeek R1 as the standard LLM for its workflow system from the initial product design stage. Argo uses DeepSeek R1 to generate the original workflow tasks. Due to the nature of the Workflow, the Token consumption and context information volume can be extremely large, with an average of >=10k Tokens. Argo also incorporates the Chain-of-Thought (CoT) into its Workflow thinking process. The open-sourcing of DeepSeek not only reduces the cost of the Workflow product but also allows for the local deployment of LLM, ensuring user privacy and security.
Before the release of DeepSeek R1, Argo had already integrated its preliminary training logic, Chain-of-Thought (CoT), into the Agent Workflow development process. Especially for tasks such as meme trading and market trend analysis, Argo has customized its Workflow using the Graph-of-Thought (GoT) model, a novel approach that constructs reasoning as a graph with nodes representing "LLM thoughts" and edges representing the dependencies between these thoughts.
Argo's choice of GoT (the only Crypto AI Workflow currently using this model) has enabled more reliable and transparent processes. This innovative approach has directly impacted the security and trustworthiness of transactions on the Argo platform. Integrating the Thought Graph (GoT) into Web3 AI agents has placed Argo at the forefront of AI-powered 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 Argo's core developer Sam collaborated with Shaw on a paper titled "EraseAnything: Enabling Concept Erasure in Rectified Flow Transformers," which explores how to remove unwanted concepts from large-scale text-to-image diffusion models without compromising the overall generation performance, and received assistance from DeepSeek researcher XingchaoLiu.
Hyperbolic
Hyperbolic Labs has also announced that it will host the DeepSeek-R1 model on its GPU platform, allowing users to rent Hyperblic GPU resources to run the DeepSeek-R1 model locally or in a specified data center, without the need to send sensitive data to DeepSeek's servers. This approach ensures data privacy while leveraging the excellent reasoning performance of the DeepSeek model. Through Hyperbolic's decentralized computing network, users can obtain the efficient inference capabilities of the DeepSeek model at a lower cost, which will be a competitive solution for startups, solo entrepreneurs, or simply efficient AI users.
This burst bubble has indeed dealt a heavy blow to the Crypto AI market, with many AI Tokens losing their speculative value. However, DeepSeek is not eliminating Crypto AI, but rather forcing the market to evolve faster. After DeepSeek R1, the future of Crypto AI will no longer rely solely 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 product development. The real challenge is whether Crypto can leverage the technological revolution brought by DeepSeek to build a truly valuable AI ecosystem, rather than just creating concepts and speculation.
This is not the end, but the evolution. Crypto AI needs to move forward faster and more aggressively. / Accelerate
