Cango Inc., once a leader in auto finance, has officially launched a 50 MW commercial AI inference operation in Winterville, Georgia. Run by its wholly owned subsidiary EcoHash, the company has fully abandoned auto lending and pivoted to computing power services in an attempt to escape a delisting crisis.

The newly formed subsidiary EcoHash uses standardized “plug-and-play” modules and integrates a proprietary software layer called the EcoLink orchestration platform.

The system is designed to aggregate geographically dispersed computing capacity and provide failover capabilities for artificial intelligence (AI) developers and data center operators.

Cango said the Georgia facility will serve as a showcase center to demonstrate performance under different cooling and power configurations, while also offering commercial services.

The move is part of Cango’s effort to diversify its energy-intensive infrastructure.

Citing Goldman Sachs research, the company said U.S. data center power demand could reach 700 terawatt-hours (TWh) by 2030, implying a potential 400 TWh gap between demand and existing infrastructure.

By retrofitting existing mining farms, Cango is offering low-latency computing capacity for AI inference workloads closer to data sources.

The Pain and Compulsion Behind Three Cross-Industry Pivots

Looking back at Cango’s twists and turns over the past few years, the story reads like a bruising account of a company repeatedly battered by the times.

In 2018, Cango peaked as soon as it went public, with its market value reaching $1.9 billion.

But few expected new-energy vehicle makers to embrace direct sales, cutting out intermediaries like Cango that lived off spreads.

In 2025, Cango exited its China operations, leaving behind a $450 million loss and walking away in disarray.

At its later peak after switching businesses, Cango mined 5% of the world’s Bitcoin, but that business depended too heavily on forces beyond its control. When coin prices swung and mining rigs depreciated, its share price fell straight to $0.46.

What the New York Stock Exchange handed it was not a lifeline, but a six-month “delisting notice.”

For Cango today, the AI bet looks less like conviction than a lack of alternatives.

The machine rooms, power systems and cooling infrastructure it still holds have become the last scrap metal it can gamble with.

Commercial Logic

Cango’s underlying logic for moving into AI is what it calls “infrastructure reuse.”

It is not building models or chasing high-end R&D. It simply wants to retrofit old mining farms and become a kind of landlord in the computing power era.

By retrofitting old mining farms and using “plug-and-play” modules, it can target gaps in the U.S. market, where computing power remains in severe shortage.

One term needs explaining here: AI inference. If “training” is making AI read thousands of books, then “inference” is sending it into the exam room.

When you ask an AI a question on your phone, the backend computation that runs is inference.

Right now, the world is short of “exam rooms.”

Goldman Sachs forecasts a 400 TWh U.S. power shortfall by 2030. Cango’s 50 MW site in Georgia is like a bag of biscuits stockpiled during a famine.

Fifty megawatts may sound large to most people, but next to the thousands of megawatts deployed by the “computing power armies” of Amazon and Google, Cango is little more than a guerrilla unit.

With no ecosystem and only raw computing power to sell, this business is ultimately a contest of brute capacity, not intelligence.

The Brutal Truth Behind the Logic

To survive in AI inference, Cango has three mountains to climb:

1. The “price war” in electricity: Power accounts for 60% of AI inference costs. Cango’s electricity price in Georgia is roughly 5 to 7 cents per kWh, while giants can secure ultra-low rates of 3 cents in Northern Europe and Iceland. In a long-term fight, Cango’s price advantage is a soap bubble.

2. The “life-or-death delivery race” for chips: A single NVIDIA H100 sells for tens of thousands of dollars. With Cango’s limited resources, can it keep up with chip iteration over time?

Once its hardware falls behind, this 50 MW center could instantly turn into a pile of worthless industrial scrap.

3. The “cash-flow hemorrhage”: As of the first quarter of this year, Cango had just over $40 million in cash on its books, while carrying $550 million in debt.

Although selling coins has temporarily kept it alive by paying down debt, an AI computing power center is a capital sink.

Without a steady stream of ammunition, its machines will soon stop roaring.

Cango’s machines in Georgia are already running, their green indicator lights blinking late at night like shifting odds on a casino floor.

For Cango now, this is no longer just a business expansion. It is one last frantic “all-in” on the edge of delisting.