DeepSeek Unveils the Burst AI Bubble: Blessing or Curse for Crypto AI?

In January 2025, the emergence of DeepSeek R1 caused a stir in the AI world, and it truly transformed the Crypto AI ecosystem. Over the previous cycle, Crypto AI was mainly focused on AI Agents. However, DeepSeek R1 and its open-source strategy completely changed the rules of the game: extremely low training costs, a breakthrough adaptive training approach, making the decentralized vision of the AI industry no longer just talk but a reachable reality. This transformation had a profound impact, with the total market capitalization of Crypto AI shrinking significantly, and many AI tokens experiencing a 70% pullback. But is this really a crisis? Or does it signify a thorough reshuffling of Crypto AI? Is DeepSeek truly the "Terminator" that shattered the narrative of Crypto AI, or is it the "Game-Changer" accelerating its entry into the era of practicality?

The Wild Growth of DeepSeek

The development of DeepSeek can be traced back to 2021. At that time, the quantitative trading-focused hedge fund Fantasia began massive recruitment of AI talent. It was not common for quantitative companies to shift to AI, and most of their recruitment was exploring cutting-edge directions, including AI researchers in areas such as Large Language Models (LLM) and Graph Models. Although there were rumors that Fantasia's transformation was to better utilize the company's idle GPU resources, the main reason was likely to seize the commanding heights of cutting-edge AI technologies.

By the end of 2022, Fantasia had absorbed more and more top AI talents, mainly from Tsinghua University and Peking University. Stimulated by ChatGPT, Fantasia's CEO, Liang Wenfeng, decided to enter the general artificial intelligence field and established DeepSeek in early 2023. However, the rapid rise of companies like ZhiPu, Dark Side of the Moon, and BaiChuan Intelligence, posed a challenge to DeepSeek. As an independent research institution lacking a star founder, securing independent funding proved to be extremely difficult. Therefore, Fantasia chose to spin off DeepSeek and fully fund its development. Despite the high risks associated with this decision, DeepSeek did not need to succumb to profit commitments or valuation pressures from funding sources. Additionally, it had a relatively sufficient reserve of GPU resources, allowing the team to focus on technological breakthroughs. A group of innovative and young individuals could sprint freely in a carefree environment. At this moment, DeepSeek resembled more of a research institute than a company.
Similar to the early days of OpenAI, no one would have thought that a company researching robots learning to solve a Rubik's Cube by hand would eventually develop ChatGPT. Similarly, no one could have imagined how Fantasia, a quantitative trading company, would pierce through the current AI bubble using DeepSeek. The former took 7 years, while the latter took only 2 years. In November 2023, DeepSeek launched the DeepSeek LLM with 670 billion parameters, performance close to GPT-4. In May 2024, DeepSeek-V2 was released, and in the same year in December, DeepSeek-V3 performed on par with GPT-4o and Claude 3.5 Sonnet in benchmark tests. DeepSeek's rapid technological leap was not due to the company's financial strength or high qualifications but rather a technological singularity that occurred after "ChatGPT Influenced the World AI Industry." Points of all sizes accelerate in any soil that can satisfy the imagination until the next critical point occurs.

Finally, in January 2025, DeepSeek accelerated through the singularity, opening that door with their first-generation reasoning-capable large model, DeepSeek-R1, which was trained at a much lower cost and exhibited superior performance compared to ChatGPT-O1.

Using Open Source to Distribute the Key to the Interstellar Gateway Worldwide

Just one day after the release of DeepSeek R1 and the publication of the open-source model, U.S. President Trump formally announced the start of a $500 billion ultra-large-scale investment plan called the "Interstellar Gateway" project during a White House press conference. A joint venture named Stargate was established by OpenAI, SoftBank, Oracle, and the investment firm MGX to build new artificial intelligence infrastructure for OpenAI in the United States.
This level of investment is even comparable to the "Manhattan Project," aiming to leverage the full force of the nation to propel closed-source AI to its climax, monopolize the AI market, and ensure the leading position of the U.S. domestic AI industry. However, at the time of the plan's announcement, no one could have foreseen that a few days later, this transoceanic open-source model would directly shut the door, not only bringing a hammer to knock on the wall by the door but also distributing hammers to others.

As an open-source model capable of competing with top-tier closed-source models, DeepSeek's innovative training architecture triggered a chain reaction, making it difficult for closed-source AI to advance. Closed-source models unable to keep up with DeepSeek R1 will be directly eliminated by the capital market. Even Marc Andreessen, co-founder of A16z, the investor in OpenAI, publicly stated the need to focus more on open-source AI rather than focusing solely on closed-source AI. Whether supporting the potential emergence of AGI or supporting AI only as an upgraded version of the SaaS industry, industry insiders believe that the disadvantages of closed-source AI far outweigh those of open source, whether it be blackboxing, industrial monopoly, information security, or capital control of attention, as any of these directions is considered a dangerous path of development.

Although some industry insiders doubted that the hybrid expert technology "MoE" in V3 required a massive dataset, suspected of being distilled from OpenAI's model, and that the reinforcement learning "RL" in R1 based on reinforcement learning methods required a large amount of hardware resources, suspected of falsifying the number of training chip uses, this did not affect the structural reform brought about by DeepSeek R1.

DeepSeek R1's open-sourcing of the training architecture shattered OpenAI's closed-source large model business logic, employing a logic of model self-evolution to avoid the massive investment of traditional paradigm's computational power and data labeling, although the training model is still a blind box, but the cost of the blind box has been greatly reduced.

On the AI hardware front, DeepSeek's V3 open sourcing directly challenges NVIDIA's market dominance. NVIDIA's GPU moat is largely due to its underlying parallel computing platform and programming model CUDA, its extensive ecosystem, and a sufficient number of developers. This makes the cost of using non-NVIDIA chips for training intermediate learning too high, and the high barriers to entry as well as political restrictions have caused a fragmentation in global AI development.

For us, in the short term, the US stock market has experienced a significant downturn in AI, Crypto AI's total market value has almost collapsed, and the market has entered a bear market. But looking ahead, the most recognized AI industry is moving towards an open, transparent, and decentralized development path. Regardless of the perspective, the combination of Crypto and AI will become more harmonious.

Redemption of Crypto AI, Forward! Forward! By all means, forward

During this period of Crypto AI's bubble burst, many AI concept tokens have experienced a 70% retracement, the Crypto AI market has shrunk significantly, and some people jokingly say, "With $5.5 million, you can train a large model, why invest in these overvalued AI tokens." Indeed, Crypto is a market dominated by capital, not products, and 90% of AI tokens are devoid of real significance.

However, in practice, with the improvement of the cryptocurrency market regulatory framework, the cryptocurrency market is still the most suitable soil for small and medium-sized AI companies to start. DeepSeek's relative 1/100 large model cost compared to ChatGPT O1 and its model training method will bring about an ecological growth of over a thousand times compared to the current market.

In simple terms, what DeepSeek brings to the crypto world is a decentralized training model, making projects like Depin more rational, making the training process and data feeding more transparent, and making the value reward mechanism for dataset contributors more reasonable, thus making settlement between the supply and demand sides of model training easier. Furthermore, the development of the peripheral ecosystem of the AI industry by over a thousand times enriches the downstream industry richness of Crypto AI. When a sufficient number of competitive and innovative product narratives appear in the market, as long as one of them truly breaks through, external funds will naturally flow back into Crypto. The market has long suffered from PVP, and a series of celebrity coins after TrumpCoin harvesting have disrupted the balance of ample liquidity and positive feedback in the AI market. Therefore, the bubble burst by DeepSeek is actually a bigger boon.

Many Crypto AI projects have already integrated DeepSeek or are soon to do so, including projects such as ElizaOS, Argo, Myshell, Build, Hyperbolic, Nillion Network, infraX, and more. Some of these projects have directly optimized their products using DeepSeek.

Myshell

Myshell has integrated V3, R1, and even the image generation model Janus-Pro into its chatbot and application plugin development workflow. Myshell's technical team completed the model integration in almost half a day. As one of the rare projects in the blockchain space that consistently polishes its products, Myshell has built a reputation in the Web2AI product realm and has been hesitant to launch its native token. With the open-sourcing of DeepSeek, Myshell's users will benefit from lower costs, enabling more Agent developers to join the already sophisticated Myshell ecosystem.

Argo

Argo's developer, Sam Gao, incorporated DeepSeek into Argo's core functionality early in the product design phase. As a workflow system, Argo standardizes LLM as a DeepSeek R1 and delegates the original workflow generation work to DeepSeek R1. Due to the nature of workflows, the token consumption and context information will be significant, with an average of>=10k Tokens. Argo also integrates CoT (Chain-of-Thought) into the workflow thought process. The open-sourcing of DeepSeek not only lowers the cost of workflow products but also allows for LLM deployment within Argo, ensuring user privacy and security.

Prior to the release of DeepSeek R1, Argo had already integrated its preliminary model training logic Chain-of-Thought (CoT) into the Agent Workflow creation flow. Particularly for tasks such as meme trading and market trend analysis, Argo has 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.

Considering that Argo has adopted the Graph of Trust (GoT) model, which is currently the only Crypto AI Workflow using this model, it has achieved a more reliable and transparent process. This innovative approach has directly impacted the security and trustworthiness of transactions on the Argo platform. Integrating the Graph of Trust (GoT) into the Web3 AI agent has positioned Argo at the forefront of AI-secured transactions. CoT's structured reasoning has not only enhanced the security of financial transactions but has also ensured transparent and reliable decision-making, which is crucial in Decentralized Finance (DeFi).

It is noteworthy that Argo's core developers Sam and Shaw collaborated on a paper titled "EraseAnything: Enabling Concept Erasure in Rectified Flow Transformers," which discusses how to remove undesired concepts from large-scale text-to-image diffusion models without compromising the overall generative performance of the model. DeepSeek researcher XingchaoLiu assisted in this paper.

Hyperbolic

Hyperbolic Labs also announced the first hosting of the DeepSeek-R1 model on its GPU platform. Users can rent Hyperbolic GPU resources to run the DeepSeek-R1 model locally or in a specified data center without sending sensitive data to DeepSeek's servers. This approach ensures data privacy, enables the utilization of DeepSeek's outstanding inference performance, and allows users to access the efficient inference capability of the DeepSeek model at a lower cost through Hyperbolic's decentralized computing network. For startups, super-individual entrepreneurs, or just efficient AI users, this will be a highly competitive solution.

The bursting of this bubble has indeed dealt a severe blow to the Crypto AI market, with many AI tokens losing their speculative value. However, fundamentally, DeepSeek is not eliminating Crypto AI but rather forcing the market to evolve more rapidly. Post-DeepSeek R1, the future of Crypto AI will no longer rely solely on speculation but will be restructured around decentralized AI computing, economic incentive mechanisms for model training, fair distribution of AI resources, practical products, and other directions. The real challenge is whether Crypto can leverage the technological revolution brought by DeepSeek to establish a truly valuable AI ecosystem, rather than just creating concepts and hype.

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