Woofun AI reports that Cango (NYSE: CANG) has executed a radical strategic pivot, moving away from the industry-standard model of hyperscale AI training to focus on decentralized inference through its new subsidiary, EcoHash. This unconventional trajectory, described by senior communications director Juliet as a "what not to do" strategy, marks the company’s third major transformation since its initial public offering. While competitors rush to lease power to massive cloud providers, Cango is betting on the fragmented, small-scale mining infrastructure that hyperscalers typically ignore. The move has drawn attention from Forbes and analysts like AididiaoJP and Foresight News, who note the stark contrast between Cango’s approach and the prevailing market trend. Juliet emphasized that this counterintuitive path is central to their identity, stating, "From an outside perspective, people would surely think this company is crazy." Yet, this perceived madness is rooted in a calculated rejection of the crowded AI training market in favor of a niche, distributed inference model.

The timeline of Cango’s evolution reveals a deliberate shedding of its original identity. Listed in New York in 2018 as China’s only auto financing platform to go public in the U.S., the company began its metamorphosis in November 2024. At that time, it agreed to acquire approximately 50 EH/s of mining hardware from Bitmain, signaling a shift toward becoming a pure Bitcoin mining entity. This acquisition was significant, as 50 EH/s represents roughly 6–8% of the global Bitcoin network’s total hash rate, which typically hovers around 600–800 EH/s. Such a volume places Cango on par with single acquisitions by major mining firms, capable of generating substantial mining capacity.

However, the transformation accelerated on April 13 of this year, when Cango launched EcoHash, an AI inference subsidiary equipped with its proprietary software layer, EcoLink. This launch marked a decisive turn away from traditional mining operations, positioning the company at the intersection of energy infrastructure and artificial intelligence.

The strategic philosophy driving this pivot is rooted in energy infrastructure rather than cryptocurrency speculation. Ms. Ye, who has worked at Cango for eight years after previously serving at the Wall Street Journal and consulting firm FTI, articulated this vision clearly. She stated that the company never intended to mine Bitcoin from the start; instead, it sought access to energy assets. This perspective emerged from Cango’s early investments in the automotive sector, including a stake in Chinese electric vehicle maker Li Auto before its listing. When Li Auto went public in 2020, Cango recorded a fair value gain of around 3.3 billion yuan (about $508 million), which sparked an interest in the power business behind automotive manufacturing. By 2023, the company began scouting for energy projects in Australia and the Middle East. During a trip to evaluate solar projects in the Middle East, the management team encountered Bitmain, leading to the realization that mining sites are essentially energy infrastructure. "All these mining sites are essentially energy infrastructure," Ms. Ye explained. "The only reason mining farms exist is that they consume energy and convert it into Bitcoin. We can still convert energy into other things."

Financial restructuring was necessary to facilitate this transition. In November 2024, Cango paid $256 million in cash to acquire 32 EH/s of mining hardware from Bitmain. Subsequently, it acquired another 18 EH/s through stock purchases from a company run by former Bitmain finance executives. To shed its 'Chinese concept stock' label, Cango sold off its entire domestic auto business for about $352 million. This divestiture allowed the company to bring in a crypto-native leadership team, including a new CEO and a chairman who founded Antalpha, a financing company related to Bitmain’s ecosystem. By mid-2025, the lending business was entirely gone, replaced by a mining-focused entity. This financial overhaul was critical in positioning Cango to pursue its new strategy, ensuring that resources were allocated toward energy infrastructure rather than legacy automotive finance operations.

The broader industry context highlights the intersection of mining and AI energy needs. Leo Wang, an executive at Canaan Creative, noted on the On The Margin podcast that "the future of high-performance AI computing might be the past of Bitcoin mining." In 2021, miners were often criticized for consuming electricity, but now that same electricity is highly sought after by AI labs. 'It’s all about an energy game,' Wang said. "We believe that in the future, energy will become an even scarcer asset for everyone." Miners possess power outlets that AI labs desire, and building new substations or signing long-term grid contracts can take years. Consequently, hyperscale cloud providers turn to Bitcoin mining companies because they have already invested in and secured power. Crypto investor Michael Terpin also highlighted the alignment of timing and cycles, noting that after each halving, mining profit margins tighten, prompting operators to find alternative revenue streams. This trend has been observed in companies like Core Scientific, which leased capacity to AI cloud provider CoreWeave, and others such as IREN and Bitfarms. An analyst behind the @0xCristal account on X wrote that "crypto mining warehouses are quietly shifting to AI inference, generating about four times more revenue," emphasizing that GPU warehouses serving LLM inference earn more than mining blocks.

Cango’s niche strategy focuses on decentralized inference rather than hyperscale training. The company operates over 30 sites globally, most with 10 to 50 megawatts of capacity. These sites are too small to satisfy hyperscale cloud providers seeking 100-megawatt campuses, but Ms. Ye believes they are ideal for AI inference. "For AI inference, you need distributed deployment. You need to be close to customers to reduce latency," she said. '10 to 50 megawatts is too small for hyperscale cloud providers, but it’s ideal for AI inference.' Ms. Ye cited a key statistic: "Over 70% of the electricity in the mining industry is actually owned by individual miners and small sites. Only 30% is controlled by those listed mining companies." These small operators own land and electricity but lack AI technology, customers, or funding. Cango aims to provide them with these resources, creating a symbiotic relationship. "We offer them a symbiotic relationship. We go to the sites and bring AI; they have the land and electricity," Ms. Ye explained. 'If there’s anything that can help Cango establish itself in the AI space over the next three to five years, it’s this symbiotic relationship among small sites.'

EcoLink serves as the technological glue for this strategy. Since small sites cannot match the 24/7 uptime of hyperscale cloud providers, Cango spreads reliability across multiple locations. "If one side goes down, we can redirect workloads to another site within milliseconds," Ms. Ye said. The target customer base includes 'long-tail customers' such as GPU rental market platforms like Runpod and Vast.ai, distributed inference clouds like Zenlayer, and AI startups too small to qualify for hyperscale cloud provider terms. Price is a significant draw: top providers may charge several dollars per GPU per hour, while the market rents out the same chip for less than a dollar. Ms. Ye noted that there are no exclusive agreements with early test customers, but most have renewed their contracts, indicating that "customer demand is absolutely real."

Cango maintains a hybrid approach, keeping Bitcoin mining as a cash engine while pursuing AI. The company still operates about 31.7 EH/s, generating $98.4 million in mining revenue in the first quarter. This cash flow supports the company’s operations while it raises funds for AI initiatives. To clean up its balance sheet, Cango sold 6,451 Bitcoin, worth about $442 million, and reduced its long-term debt from $557.6 million to $30.6 million in one quarter—a 94.5% drop. Its Bitcoin reserves were reduced to about 1,000 coins. With this financial clarity, Cango raised $75 million to launch EcoHash. The first AI node will be deployed at a 50-megawatt site in Georgia, which Cango acquired for $19.5 million last August. Ms. Ye calls this site a "living showroom," with two to three more nodes expected to come online by the end of this year.

Despite these developments, market skepticism remains. Wang noted that "people are a bit cautious about this," citing concerns about bubbles. Converting warehouses filled with fans into air-cooled AI data centers is costly, and many mining companies see their stock prices soar due to press releases without achieving tangible results. For instance, the company formerly known as Bitfarms saw its stock rise by hundreds of percentage points after rebranding to an AI-focused company, but this was before it earned a single dollar in AI revenue. Analysts warn that the funds required for such transitions amount to billions of dollars. Bitcoin holders also have concerns, arguing that as miners shut down machines, the network’s hash rate declines, and security costs are being overlooked. A widely shared post on X warned that "Bitcoin miners are sacrificing the network for AI funding." Cango’s own buffer is thin, with only $7.2 million in cash at the end of the quarter after debt reduction, leading some media outlets to question its status on the NYSE. Even landmark deals are shaky, as evidenced by CoreWeave’s $9 billion acquisition offer for Core Scientific falling through earlier this year.

Ms. Ye’s response to these challenges reflects the discipline that underpins Cango’s strategy. She acknowledges that giant facilities and landmark training leases will belong to the giants.

However, Cango is betting on the rest: thousands of megawatts of power spread among small independent miners, power that giants cannot easily access. She believes that much of AI inference will operate quietly in this distributed landscape. This bet on small-scale distributed power represents a significant departure from the industry norm, positioning Cango as a unique player in the evolving crypto-AI energy shift. The success of this strategy will depend on Cango’s ability to maintain its symbiotic relationships with small miners and deliver reliable, cost-effective inference services to its long-tail customers. As the market continues to evolve, Cango’s approach may offer a viable alternative to the hyperscale model, leveraging the untapped potential of decentralized energy infrastructure.