Late-night chaos! This Chinese mining firm counter-trades AI infrastructure by going short, betting on 70% of retail users’ computing power—an epic strategic masterstroke or a path straight to its own downfall?

Bro, let me tell you an interesting story. About a year and a half ago, a Chinese auto-loan company spent several hundred million dollars to buy 50 EH/s of mining rigs and became a Bitcoin miner. Now it’s doing the opposite thing—no AI training, but instead distributed inference.

The company is Cango (NYSE ticker: CANG). It listed on the NYSE in 2018, when it was the only Chinese auto finance platform listed in the US. In November 2024, it agreed to acquire Bitmain’s roughly 50 exahash worth of mining rigs, becoming a pure-play Bitcoin mining company. Then this April 13, it launched an AI inference subsidiary called EcoHash, equipped with its own software layer, EcoLink. No AI training, and no building massive new data centers. It’s simply betting that those patchwork small mining companies that hyperscale cloud providers can’t use are precisely where a large amount of AI computing power will land.

Note: 50 exahash is enormous. The global Bitcoin network’s total hash rate typically fluctuates around 600–800 EH/s. 50 EH/s is about 6–8% of the global total hash rate—on the scale of a single acquisition by a large mining firm, and it can deliver significant mining capacity.

Cango’s senior director of communications, Juliet Zhu, said: “What you shouldn’t do is as important as what you should do.” She repeatedly emphasized this line—those nine characters are the whole strategy’s core.

Energy first, Bitcoin second. Ms. Ye said the company never planned to mine Bitcoin from the start—it wanted to own energy. She has worked at Cango for eight years; previously, she worked at The Wall Street Journal and consulting firm FTI. Her story began with cars. Cango early invested in China’s electric vehicle maker Li Auto before it listed. When Li Auto went public in 2020, Cango recorded about RMB 3.3 billion (about $508 million) in fair value gains, and it became interested in the electricity business behind the vehicles. By 2023, it began looking for energy projects in Australia and the Middle East.

“In a trip looking for solar projects in the Middle East, management ran into Bitmain,” Ms. Ye said. That’s how the auto-loan company met Bitcoin mining. What truly impressed them wasn’t the coin, but the grid. “All those mining sites are basically energy infrastructure,” Ms. Ye said. “The only reason mining farms exist is that they consume energy and turn it into coins. We can still turn 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 is not cheap. In November 2024, Cango paid $256 million in cash to acquire Bitmain’s 32 EH/s mining rigs, then used stock to acquire another 18 EH/s—shares that went to companies run by former Bitmain’s finance director. To shake off 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 a financing company tied to Bitmain and the world, Antalpha. By mid-2025, the lending business was gone—replaced by a mining company.

Why is everyone shifting? Cango isn’t the only mining firm pivoting for AI. The math of mining meets the math of AI—both are competing for the same thing: electricity. “The future of high-performance computing for AI might be the past of Bitcoin mining,” Leo Wang, an executive at Ganaan/Yang? (Ka Nan Yu Zhi) said on the On The Margin podcast. In 2021, miners were the villains, blamed for consuming power. Now the same power has become sought after. “It’s all an energy game,” Wang said. “We believe future energy will become a scarcer asset for everyone.”

What miners have isn’t chips that AI labs want—it’s a plug. Building new substations and signing long-term power grid contracts can take years. “When hyperscale cloud providers look for suppliers that can guarantee power in the short term, they turn to Bitcoin miners, because Bitcoin miners have already invested and secured power,” Wang said. He added that what’s “lucky” for miners is that AI emerged at the moment block rewards were shrinking. Timing matches the cycle. “We’ve been following the four-year cycle very precisely,” Michael Terpin, a crypto investor, said on the On The Margin podcast. After each halving, mining profit margins tighten, and operators look for a second way to make money.

The market has moved too. Core Scientific, an early mover, leases capacity to AI cloud providers like CoreWeave; other miners—from IREN to companies once known as Bitfarms—have followed suit. “The crypto mining warehouse is quietly turning to AI inference and generating about four times the revenue,” @0xCristal账户背后的分析师在X平台上写道. “A warehouse of GPUs serves large language model inference and earns more than blocks mined.”

Betting against giant sites is what makes Cango different. The popular approach is to turn a few large sites into AI training parks and sign long-term leases with hyperscale cloud providers. Cango rejected that approach. “We absolutely don’t do AI training,” Ms. Ye said. “That field is already crowded with hyperscale cloud providers. Competing with them isn’t realistic.” The decision is rooted in the company’s scale. Cango has more than 30 sites worldwide, mostly 10 to 50 MW each. They’re too small to satisfy hyperscale cloud providers that want 100 MW parks. But Ms. Ye believes that size is perfect for the other half of AI. “For AI inference, you need distributed deployment. You need to be close to customers to reduce latency,” she said. “10 to 50 MW is too small for hyperscalers, but perfect for AI inference.”

Then she mentioned her favorite data point. “In the mining industry, more than 70% of electricity is actually controlled by individual players and small sites,” Ms. Ye said. “Only 30% is controlled by those listed mining companies.” These small operators own land and power. They don’t own AI technology, customers, or financing. Cango wants to bring all of that to them. “We provide a symbiotic relationship. We come to the sites, bring AI, and they have the land and power,” she said. “If there’s anything that helps Cango stand firm in AI over the next three to five years, it’s the symbiotic relationship between these small sites.” EcoLink is the glue. A small site can’t match hyperscalers’ always-on uptime, so Cango spreads reliability across nodes. “If one side goes down, we can route workloads to another site within milliseconds,” Ms. Ye said.

So far, the buyers are what she calls the tail customers. Platforms for GPU rental like Runpod and Vast.ai, distributed inference cloud like Zenlayer, and AI startups too small to sign hyperscale cloud provider terms. Pricing is the draw: top-tier providers may charge several dollars per GPU per hour, while the market rents the same chip for under $1. Ms. Ye said there are no exclusive agreements signed with early test customers, and most renewed. “Customer demand is absolutely real.”

Cash engine, and costs. Cango didn’t give up on Bitcoin. It’s still running about 31.7 EH/s, generating $98.4 million in mining revenue in the first quarter. This is the cash that keeps operations going while raising funds for AI. “Most miners just completely abandon Bitcoin mining,” Ms. Ye said. “For us, it’s more of a hybrid approach.” The cleanup is brutal. “We’re basically clearing the deck,” Ms. Ye said. “Investors may want to invest in our AI transformation, but they don’t want their money used to pay off old debts.” So Cango sold 6,451 Bitcoins (about $442 million) and cut long-term debt from $557.6 million to $30.6 million within a quarter—a drop of 94.5%. Its Bitcoin reserves fell to about 1,000 BTC. Then it raised $75 million to launch EcoHash. The first AI node will be deployed at a 50 MW site in Georgia, acquired by Cango last August for $19.5 million. Ms. Ye calls it a “living showroom.” Two to three more nodes are expected to go live before the end of this year.

Skeptics aren’t buying it—at least not everyone. “People are a bit cautious about it,” Wang said when discussing the AI boom, “because people worry about a bubble.” The story is ahead of the revenues by several years. Turning a warehouse full of fans into a liquid-cooled AI data center is expensive. Many miners’ stocks spiked on press releases but didn’t actually make much. The company once known as Bitfarms saw its stock rise by hundreds of percentage points after renaming itself for AI, but it didn’t make $1 of AI revenue yet. Analysts tracking these pivots keep warning that the funds needed to complete these transformations run into billions. Bitcoin holders have different concerns. As miners shut down rigs, the network’s hash rate has declined, and some people believe security costs are being ignored. “Bitcoin miners are giving up the network to fund AI,” a widely shared post on the X platform warned.

Cango’s buffer is thin. After the debt cleanup, cash at quarter-end was only $7.2 million, and at least one media outlet questioned its standing on the NYSE. Even flagship deals are shaky: CoreWeave’s $9.0 billion acquisition offer for Core Scientific earlier this year reportedly fell through. Ms. Ye’s answer is the discipline that runs through everything she described. Giant sites and flagship training leases will be owned by the giants. Cango is betting on the rest: thousands of megawatts of power distributed among small independent miners, and power the giants can’t easily reach. She believes a large amount of AI inference will quietly run there.

Follow me: Get more real-time crypto market analysis and insights! $BTC $ETH $SOL

#预测世界杯挪威VS英格兰 # GateUS compliance expansion in Florida #US stock AI concept stocks rise broadly