Original author: BUBBLE
In January 2025, the advent of DeepSeek R 1 caused a shock in the AI world, and it also truly changed the Crypto AI ecosystem. In the past cycle, Crypto AI mainly revolved around AI Agent, and DeepSeek R 1 and its open source strategy completely changed the rules of the game: extremely low training costs and breakthrough adaptive training methods make the vision of decentralization of the AI industry no longer an empty talk, but a reality within reach. This change has far-reaching impacts. The total market value of the Crypto AI market has shrunk significantly, and many AI tokens have experienced a 70% callback, but is this really a crisis? Or does it mean a complete reshuffle of Crypto AI? Is DeepSeek the terminator that shatters the Crypto AI narrative, or the game breaker that accelerates its entry into the era of practical application?
Wild Growth of DeepSeek
The development of DeepSeek can be traced back to 2021. At that time, the hedge fund Huanfang, which focuses on quantitative trading, began to recruit AI talents on a large scale. It was rare for quantitative companies to switch to AI, and most of the recruits were AI researchers who explored cutting-edge directions, including large models (LLM) and literary graph models. Although there are rumors that Huanfang made the transformation in order to better utilize the companys idle GPU resources, most of the reasons should still be to seize the commanding heights of cutting-edge AI technologies such as large models.
By the end of 2022, Huanfang had attracted more and more top AI talents, mainly students from Tsinghua University and Peking University. Stimulated by ChatGPT, Liang Wenfeng, CEO of Huanfang, decided to enter the field of general artificial intelligence and founded DeepSeek in early 2023. However, with the rapid rise of AI companies such as Zhipu, Dark Side of the Moon, and Baichuan Intelligence, DeepSeek, as a pure research institution and lacking star founders, faces huge difficulties in independent financing. Therefore, Huanfang chose to divest DeepSeek and fully fund its development. Although this decision is extremely risky, DeepSeek does not need to be subject to profit commitments or valuation pressures from the financing party. At the same time, it has a relatively sufficient reserve of GPU resources, allowing the team to focus on technological breakthroughs, and a group of innovative young people can run rampant in a paradise. At this moment, DeepSeek is more like a research institute than a company.
Just like the early days of OpenAI, no one would have thought of how a company that studies robot hands playing Rubiks Cube would eventually develop ChatGPT, and no one could have thought of how Magic Cube, a quantitative company, would use DeepSeek to break through the current AI bubble. The former took 7 years, and the latter only 2 years. In November 2023, DeepSeek LLM with 67 billion parameters and performance close to GPT-4 was launched, DeepSeek-V2 was launched in May 2024, and DeepSeek-V3 released in December of the same year performed on par with GPT-4 o and Claude 3.5 Sonnet in benchmark tests. DeepSeeks rapid technological leap is not due to the companys financial resources or high education, but after a technological singularity occurs, ChatGPT affects the worlds AI industry, and large and small singularities accelerate in any soil that can satisfy imagination until the next key singularity appears.
Finally, in January 2025, DeepSeek accelerated through the singularity and opened that door with their first generation of large model with reasoning capabilities, DeepSeek-R 1, with training costs far lower than ChatGPT-O 1 and excellent performance.
Distributing the key to the Stargate to the world with open source
Just one day after DeepSeek R 1 was released and the open source model was announced, US President Trump officially announced the start of a $500 billion ultra-large-scale investment Stargate plan at a White House press conference. OpenAI, SoftBank, Oracle and investment company MGX jointly established a joint venture called Stargate to build a new artificial intelligence infrastructure for OpenAI in the United States.
This level of investment is even comparable to the Manhattan Project, which is intended to use the power of the whole country to push closed-source AI to a climax with algorithm stacking, monopolize the AI market, and ensure the leading position of the US domestic AI industry. However, when the plan was released, no one would have thought that a few days later, this open source model on the other side of the ocean would not open its door, not only bringing a hammer to the door to smash the wall, but also giving hammers to others.
As an open source model that can rival the top closed source models, DeepSeeks new training architecture has triggered a chain reaction, making it difficult for closed source AI to move forward. Closed source models that cannot outperform DeepSeek R 1 will be directly eliminated by the capital market. Even Marc Andreessen, the founder of A16z, OpenAIs investor, has publicly stated that more attention should be paid to open source AI rather than closed source AI. In the industry, whether it is supporting the possibility of AGI or supporting AI as an upgraded version of the SaaS industry, it is believed that the harm of closed source is far greater than that of open source. Whether it is black box, industry monopoly, information security, or capital manipulation of attention, any of them is a very dangerous development direction.
Although some industry insiders suspect that V3s mixed expert technology MoE requires a huge data set and is suspected of using OpenAIs model for distillation, and that the reinforcement learning-based method in R1s reinforcement learning RL requires a lot of hardware resources, and is suspected of falsifying the number of training chips used, it does not affect the industry structure reform it brings.
The open source of DeepSeek R 1 breaks the closed-source large model business logic of OpenAI in terms of training architecture, and uses the logic of allowing the model to evolve by itself to avoid the large investment in computing power and data labeling of the traditional paradigm. Although the training model is still a blind box, the cost of the blind box is much lower.
At the AI hardware level, DeepSeeks V3 open source directly challenges NVIDIAs market dominance. NVIDIAs GPU moat is largely based on its underlying parallel computing platform and programming model CUDA. Its extensive ecosystem and enough developers make the learning cost of using non-NVIDIA chips for training too high. The high threshold purchasing conditions and political restrictions have caused a split in the development of global AI.
For us, in the short term, the US AI stock market has shrunk significantly, the total market value of Crypto AI has almost dropped, and the market has entered a bear market. But in the long term, the most recognized AI industry is moving towards an open source, transparent, and decentralized development path. From any perspective, the combination of Crypto and AI will be more tacit.
Crypto AIs redemption, move forward! Move forward! Move forward by any means necessary
During this round of Crypto AI bubble burst, many AI concept tokens received a 70% callback, and the Crypto AI market shrank significantly. Some people joked that 5.5 million US dollars can train a large model. The market value of these AIs exceeds 500 million US dollars. Why buy Crypto AI? Indeed, Crypto is a market dominated by funds, not products. 90% of AI tokens have no practical significance.
But in fact, with the improvement of the regulatory system of the crypto market, the crypto market is still the most suitable soil for small and medium-sized AI companies to start businesses. Compared with ChatGPT O 1, DeepSeek brings 1/100 of the model cost and model training method, which will bring more than 10,000 times the ecological growth of the current market.
To put it bluntly, what DeepSeek brings to crypto is a decentralized training model, which makes Depin-type projects more rational, makes the training process and information feeding more transparent, and makes the reward mechanism for data set contributors to obtain value more reasonable, making it easier for both supply and demand sides of model training to settle. The development of the surrounding ecology of the AI industry, which is more than 10,000 times larger, has further improved the industry richness of the downstream of Crypto AI. When enough competitive and creative product narratives appear in the market, as long as one of them really breaks the circle, external funds will naturally flow back into Crypto. The market has suffered from PVP for a long time. A series of celebrity coin harvests after TrumpCoin broke the original abundant liquidity and positive feedback balance of the AI market. Therefore, the bubble punctured by DeepSeek is actually a greater benefit.
Currently, many Crypto AI projects have either quickly integrated DeepSeek or updated its architecture, including ElizaOS, Argo, Myshell, Build, Hyperbolic, Nillion Network, infraX, etc. Some of these projects have been optimized directly on the product side through DeepSeek.
Myshell
V3, R1 and even the image generation model Janus-Pro were added to the production flow of chatbots and application plug-ins. Myshells technicians completed the model integration in almost half a day. As one of the few projects in the blockchain that has always insisted on polishing its products and even made a name for itself in Web2 AI products but has been reluctant to issue coins, the open source of DeepSeek will bring good news to Myshell users on the cost side. Lower costs will bring more Agent developers to Myshell, whose products are already perfect.
Argo
Argos developer Sam Gao DeepSeeked Argos important functions in the early stages of product design. As a workflow system, Argo built LLM into the standard DeepSeek R 1 and handed over the original workflow generation work to DeepSeek R 1. Because of the WorkFlow, the token consumption and the amount of context information will be very huge (average >= 10k Tokens), and Argo also integrated CoT (Chain-of-Thought) into the WorkFlow thinking process. The open source of DeepSeek not only reduces the cost of workflow products, but also allows LLM to be deployed locally in Argo, and the privacy and security of users can also be guaranteed.
Before DeepSeek R 1 came out, Argo had integrated its model initial training logic Chain-of-Thought CoT into Argos Agent Workflow production process. Especially for tasks such as meme trading and market trend analysis, Argo customized its workflow using Graph-of-Thought (GoT), a novel approach that builds reasoning as a graph where nodes represent LLM ideas and edges represent dependencies between these ideas.
Argo chose GoT, the only Crypto AI Workflow that currently uses this model, to achieve a more reliable and transparent process. This innovative approach directly affects the security and trust of transactions on the Argo platform. Integrating the Mind Map (GoT) into the Web3 AI agent puts 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 critical in decentralized finance (DeFi).
It is worth noting that Argo core developers Sam and Shaw co-wrote a paper titled EraseAnything: Enabling Concept Erasure in Rectified Flow Transformers on how to remove undesirable concepts from large-scale text-to-image diffusion models without compromising the overall generation performance of the model. They received help from DeepSeek researcher Xingchao Liu.
Hyperbolic
Hyperbolic Labs also took the lead in announcing the hosting of the DeepSeek-R 1 model on its GPU platform. Users can rent Hyperblic GPU resources to run the DeepSeek-R 1 model locally or in a designated data center without sending sensitive data to DeepSeeks servers. This approach not only protects data privacy, but also takes advantage of the excellent reasoning performance of the DeepSeek model. At the same time, through Hyperbolics decentralized computing network, users can obtain the efficient reasoning capabilities of the DeepSeek model at a lower cost. For startups, super individual entrepreneurs, or even pure AI efficient users, it will be a very competitive solution.
The bursting of this round of bubble has indeed dealt a heavy blow to the Crypto AI market, and many AI Tokens have lost their hype value. But in essence, DeepSeek is not destroying Crypto AI, but forcing the market to accelerate its evolution. After DeepSeek R 1, the future of Crypto AI will no longer rely solely on speculation, but will be reconstructed around decentralized AI computing, economic incentive mechanisms for model training, fair distribution of AI resources, and practical products. The real challenge is whether Crypto can use the technological revolution brought about by DeepSeek to build a truly valuable AI ecosystem, rather than just creating concepts and hype.
This is not the end, but evolution. Crypto AI needs to move faster and more aggressively. / Accelerate
