Author: Forbes
Translation: AididiaoJP, Foresight News
"From the outside, people must think this company is crazy," said Juliet. "Who are they? They make such bold moves while knowing nothing about the industry." She was referring to the day a Chinese auto loan company spent hundreds of millions of dollars to become a Bitcoin miner.
That was about a year and a half ago. Now, it's doing the opposite. Almost every listed Bitcoin mining company is racing to lease power to hyperscale cloud providers building massive AI training clusters. But Cango Inc. (NYSE: CANG) is bucking the trend.
Cango is currently in its third stage of transformation. It went public in New York in 2018 as China's only US-listed auto financing platform. In November 2024, it agreed to acquire about 50 exahashes of mining machines from Bitmain, becoming a pure-play Bitcoin miner. Then, on April 13 this year, it launched an AI inference subsidiary called EcoHash, equipped with its own software layer EcoLink. No AI training, no building of giant new data centers. It's simply betting that the fragmented, small-scale mining sites that hyperscalers can't utilize are where a significant amount of AI compute will land.
Note: 50 exahashes is an enormous amount of computing power. The global Bitcoin network hash rate typically fluctuates around 600–800 EH/s, so 50 EH/s represents about 6–8% of the total global hash rate. This is a large-scale, single-acquisition level for a mining company, bringing significant mining capacity.
"What not to do is as important as what to do," said Juliet, Cango's Senior Director of Communications. She repeated this phrase. These nine words are the core of the entire strategy.
Energy First, Bitcoin Second
Ms. Ye stated that from the very beginning, the company's intention was not to mine Bitcoin, but to own energy.
She knows this history inside out. She has been with Cango for eight years, previously working at The Wall Street Journal and consulting firm FTI. The story she tells starts with cars. Cango invested early in Chinese EV maker Li Auto, before its IPO. When Li Auto went public in 2020, Cango recorded a fair value gain of about 3.3 billion RMB (approximately $508 million) and became interested in the power business behind the cars. By 2023, it began scouting energy projects in Australia and the Middle East.
"During a trip to the Middle East to look for solar projects, the management team happened to run into Bitmain," said Ms. Ye. This was the meeting of an auto loan company and Bitcoin mining.
What truly impressed them wasn't the coin, but the infrastructure. "All these mining sites are essentially energy infrastructure," Ms. Ye said. "The only reason mining farms exist is that they consume energy and turn it into coins. We can still convert energy into other things." Mining was just an entry point. "From day one, we weren't thinking about doing Bitcoin mining. We were thinking about operating energy infrastructure from day one."
The entry cost was steep. Cango paid $256 million in cash in November 2024 to acquire 32 EH of Bitmain miners, then acquired another 18 EH via a stock deal with a company run by Bitmain's former financial controller. To shed the "China concept stock" label, it sold its entire domestic auto business for about $352 million. It brought in crypto-native leadership, including a new CEO and a chairman who founded the Bitmain-related financing company Antalpha. By mid-2025, the loan business was gone. A mining company stood in its place.
Why Everyone Is Pivoting
Cango isn't the only miner pivoting for AI. The math of mining met the math of AI, both competing for the same thing: electricity.
"The future of AI high-performance computing may be the past of Bitcoin mining," said Canaan Creative executive Leo Wang on the On The Margin podcast. In 2021, miners were the villains, blamed for consuming power. Now, that same power is in high demand. "It's all an energy game," Wang said. "We believe energy will become a scarcer asset for everyone in the future."
What miners hold, and what AI labs crave, isn't chips, but a plug. Building new substations and securing long-term grid contracts can take years. "When hyperscalers look for suppliers who can guarantee power on short notice, they turn to Bitcoin miners, because Bitcoin miners have already put up capital and secured the power," Wang said. He added that miners are "lucky" that AI emerged just as block rewards are decreasing.
The timing matches the cycle. "We've followed the four-year cycle very precisely," crypto investor Michael Terpin said on the On The Margin podcast. After each halving, mining margins tighten, and operators look for a second way to make money.
The market has followed. Core Scientific was an early mover, leasing capacity to AI cloud provider CoreWeave, and others from IREN to the company formerly known as Bitfarms have followed suit. "Crypto mining warehouses are quietly pivoting to AI inference for ~4x the revenue," wrote the analyst behind the @0xCristal account on X. "A GPU warehouse serving LLM inference earns more than mining blocks."
Betting Against the Megasite
This is where Cango differs. The popular move is to convert a few large sites into AI training campuses and sign a long-term lease with one hyperscaler. Cango rejected this.
"We absolutely do not do AI training," Ms. Ye said. "That space is already crowded with hyperscalers. It's not realistic for us to compete." This decision stems from the company's own scale. Cango has over 30 sites globally, mostly 10 to 50 megawatts. Too small to satisfy hyperscalers hunting for 100 MW campuses. But Ms. Ye argues it's perfect for the other half of AI. "For AI inference, you have to deploy it distributed. You have to be near customers to reduce latency," she said. "Ten to 50 megawatts is too small for hyperscalers, but it's perfect for AI inference."
Then she mentioned her favorite data point. "Over 70% of the power in the mining industry is actually owned by individual players, small sites," Ms. Ye said. "Only 30% is controlled by those listed miners." These small operators have land and power. They don't have the AI tech, the customers, or the financing. Cango wants to bring it all to them. "We offer them a symbiotic relationship. We come to the site, bring AI, they have the land and power," she said. "If anything is going to allow Cango to stand on its own feet in AI in the next three to five years, it's this symbiotic relationship among small sites."
EcoLink is the glue. A single small site can't match hyperscaler uptime, so Cango distributes reliability. "If one side goes down, we can reroute workloads to another site within milliseconds," Ms. Ye said. The buyers so far are what she calls long-tail customers. GPU rental marketplaces like Runpod and Vast.ai, distributed inference clouds like Zenlayer, and AI startups too small to negotiate hyperscaler terms. Price is the draw: top providers might charge a few dollars per GPU per hour, while the market rents the same chip for less than a dollar. Ms. Ye says no early test customers signed exclusive deals, and most have renewed. "Customer demand is absolutely real."
The Cash Engine, and the Cost
Cango hasn't abandoned Bitcoin. It's still running about 31.7 EH, which brought in $98.4 million in mining revenue in Q1. That's the cash to keep the company running while it raises funds for AI. "Most mining companies just completely abandon Bitcoin mining," Ms. Ye said. "For us, it's more of a hybrid approach."
The cleanup was brutal. "We basically cleaned the decks," Ms. Ye said. "Investors might want to invest in our AI transformation, but they don't want their money going to repay old debt." So Cango sold 6,451 bitcoins, about $442 million, and cut long-term debt from $557.6 million to $30.6 million in one quarter—a 94.5% reduction. Its bitcoin reserve shrank to about 1,000 coins. It then raised $75 million for the EcoHash launch. The first AI node will deploy at a 50 MW site in Georgia that Cango acquired last August for $19.5 million. Ms. Ye calls it the "living showroom." Two to three more nodes will go live before year-end.
The Skeptics
Not everyone is buying it. "People are a little cautious about it," Wang said of the AI hype, "because people worry about a bubble." The story is years ahead of revenue. Converting a fan-filled warehouse into a liquid-cooled AI data center is expensive. Many miners saw their stock prices surge on press releases but have nothing to show. The company formerly known as Bitfarms saw its stock rise hundreds of percent after its AI rebrand, before earning a single dollar of AI revenue. Analysts tracking these pivots constantly warn that the capital needed to complete them runs into billions.
Bitcoin holders have different concerns. As miners shut down machines, the network's hash rate has slipped, with some arguing the security cost is being ignored. "Bitcoin miners are abandoning the network for AI money," a widely circulated X post warned. Cango's own cushion is thin. After the debt cleanup, quarter-end cash stood at just $7.2 million, and at least one outlet questioned its status on the NYSE. Even marquee deals have wobbled: CoreWeave's $9 billion takeover offer for Core Scientific fell apart earlier this year.
Ms. Ye's answer is the discipline that runs through everything she says. Megasites and marquee training leases will belong to giants. Cango is betting on the rest: the thousands of megawatts scattered among small, independent miners and the power that giants can't easily reach. That, she believes, is where a vast amount of AI inference will quietly run.
