In its second-quarter 2026 earnings report, Cango disclosed the latest progress in scaling back its bitcoin mining operations and pivoting toward AI computing power.

The company’s total revenue fell about 50% quarter on quarter, from $102 million in the first quarter to $50.8 million in the second. Its net loss from continuing operations was $81.6 million.

Management said the company is shifting its focus from expanding computing power to improving unit economics. Its operating loss narrowed about 68%, from $254.4 million in the first quarter to $80.6 million in the second, while adjusted EBITDA improved from a loss of $154.1 million to a loss of $10.7 million.

Behind the simultaneous decline in revenue and costs, however, was Cango’s decision to retire inefficient mining rigs, reduce its self-mining computing power and move some higher-cost computing power to a revenue-sharing model.

Scaling Back Mining to Cut Operating Costs

The main reason for Cango’s second-quarter revenue decline was its reduction in operating computing power. The company continued to retire less energy-efficient S19-series mining rigs and shifted some computing power to a leasing model.

Cango’s total operating computing power stood at 37.01 EH/s at the end of the first quarter. By the end of the second quarter, it had fallen to 27.58 EH/s.

Self-mining computing power declined from 27.98 EH/s to 19.84 EH/s, with another 7.74 EH/s under leasing arrangements. As a result, second-quarter bitcoin production fell from 1,266 BTC in the first quarter to 656 BTC, while bitcoin mining revenue dropped from $98.4 million to $47.4 million.

Cango began phasing out S19-series mining rigs in late 2025 while deploying more energy-efficient S21-series models. At its first-quarter earnings call, management said S19 and S21 rigs accounted for roughly 80% and 20%, respectively, of operating machines as of the end of May 2026.

At the second-quarter earnings call, management said S21-series mining rigs now accounted for slightly more than one-third of operating machines after excluding leased computing power. The company is still retiring some S19-series rigs at sites with high electricity prices.

The equipment changes also resulted in substantial accounting losses. Cango recognized a $42.9 million impairment loss on mining rigs and an additional $8.5 million loss on equipment disposals in the second quarter. The combined $51.4 million was one of the main drivers of the quarterly loss.

At some sites with higher electricity costs, Cango moved 7.74 EH/s of computing power to revenue-sharing lease arrangements. Management said on the earnings call that, under this model, site operators cover electricity, maintenance and operating expenses, while the bitcoin mined is divided with Cango according to an agreed ratio.

The arrangement reduces the electricity and operating expenses borne directly by Cango, but it also means that some computing power no longer contributes the full amount of mining revenue it would generate under a self-mining model.

The company’s cost metrics improved as a result. Cango’s cash cost per bitcoin fell about 4.7% quarter on quarter, from $76,928 in the first quarter to $73,313 in the second. Its all-in cost, including mining rig depreciation, declined from $99,747 to $98,405.

Cost of revenue excluding depreciation fell about 49%, from $99.6 million to $50.7 million, while depreciation expenses declined about 42.5%, from $29.4 million to $16.9 million.

Cango’s operating loss narrowed from $254.4 million in the first quarter to $80.6 million in the second. Losses from changes in the fair value of crypto assets also fell, from $151.8 million to $4.1 million. Cango also began using bitcoin hedging instruments to reduce the impact of price volatility on operating cash flow.

These improvements reflected a combination of reduced operating scale, changes to the mining rig mix, the leasing model and movements in bitcoin prices. The narrower costs and losses therefore do not mean that mining revenue has returned to growth.

3 MW AI Business Remains at an Early Stage

As it scales back bitcoin mining, Cango is expanding into high-performance computing and AI inference through its subsidiary EcoHash.

Established in 2025, EcoHash is the Cango subsidiary responsible for its high-performance computing and AI inference business. It uses the power supply and infrastructure at existing mining farms to deploy modular AI computing equipment.

Cango’s wholly owned site in Georgia has a total capacity of 50 MW. A dedicated AI computing zone can support up to 3 MW, with room reserved for further expansion.

According to Cango’s latest announcement on September 3, EcoHash has completed infrastructure upgrades for the site’s AI computing zone. High-density computing containers have been delivered, installed and tested, and the first batch of GPU servers is powered on and operational.

EcoHash also said it has begun offering commercial GPU computing services using its first batch of servers, while additional GPU equipment is being deployed in phases. The company said its initial and prospective customers include GPU cloud platforms and AI-native cloud infrastructure providers, but it did not disclose customer names, contract values, service terms or profit margins.

Cango plans to explore both bare-metal GPU hosting and colocation services, and is operating test nodes in Texas and on the U.S. West Coast. It is also evaluating other potential sites and the possibility of building its own facilities.

Compared with Cango’s existing mining operations, however, the 3 MW AI computing zone remains small.

Bitcoin mining generated $47.4 million in the second quarter, accounting for about 93% of Cango’s total revenue. Commercial services at the Georgia AI project began only after the quarter ended and were therefore not included in the second-quarter financial results.

The company had previously expected to begin recognizing AI-related revenue in the third quarter. Although the September 3 announcement confirmed that commercial GPU services had launched, Cango did not disclose the revenue value of its initial contracts or provide full-year guidance for the AI business.

It is therefore too early to determine whether the business can offset the decline in mining revenue in the near term or improve Cango’s overall profitability.

Cash Reserves and Transition Spending Remain Key Concerns

As of June 30, 2026, Cango held $10.1 million in cash and cash equivalents, up from $7.2 million at the end of the first quarter. Long-term related-party debt rose slightly to $31.2 million from $30.6 million.

Cango’s long-term related-party debt reached $557.6 million at the end of 2025 before falling sharply to $30.6 million in the first quarter of 2026 through active deleveraging. Although its debt load at the end of the second quarter was far below its level at the start of the year, cash reserves remained limited.

Cango must simultaneously fund its existing mining operations, mining rig upgrades and the expansion of its AI infrastructure. Its future financial position will depend on bitcoin prices, electricity costs, mining rig efficiency, AI customer acquisition and additional capital expenditure.

Management said on the second-quarter earnings call that the company would not set a hard computing power target for its mining business, instead focusing more closely on margins and cash flow. This means Cango may continue adjusting the balance among self-mining, leasing and equipment retirements based on the economics of each site.

Meanwhile, the AI business remains at an early stage of commercialization. The Georgia site has begun providing GPU computing services, but Cango has not disclosed customer numbers, contract values, equipment utilization or revenue contributions.

If the AI business expands more slowly than expected, or if mining rig upgrades fail to reduce costs further, Cango could continue to face cash flow and financing pressure. Conversely, only if EcoHash can add customers and replicate the Georgia site’s modular deployment model elsewhere will Cango be able to gradually reduce its dependence on bitcoin mining revenue.

For now, Cango has taken its first step in transforming from a bitcoin mining company into an AI computing power infrastructure operator. But its initial 3 MW deployment is not yet large enough to show that the model can become a major new source of revenue. The transition will need to be assessed through third-quarter AI revenue, subsequent customer contracts and further infrastructure expansion.