Teleconferência de Resultados da Cango (CANG) do 2T26: Reorganização da Mineração e Receita de IA do 3T

Principais Destaques

• A Cango Inc. reportou receita de US$ 50,8 milhões no 2º trimestre de 2026, queda de aproximadamente 50% em relação ao 1º trimestre, à medida que a empresa reduziu a taxa de hash operacional e migrou parte da capacidade de mineração para um modelo de leasing.
• A mineração de Bitcoin gerou US$ 47,4 milhões em receita. A Cango minerou 656 Bitcoins a um custo em caixa médio de US$ 73.313 por moeda, uma redução sequencial de cerca de 5%, e um custo total (all-in) de US$ 98.405 por moeda.
• O prejuízo líquido das operações continuadas foi de US$ 81,6 milhões. O resultado incluiu US$ 42,9 milhões em encargos de redução ao valor recuperável (impairment) de máquinas de mineração e uma perda por alienação de US$ 8,5 milhões, ambos vinculados à reestruturação da base de ativos de mineração.
• A taxa de hash operacional ficou em 27,58 EH/s em 30 de junho, composta por 19,84 EH/s de capacidade de automineração e 7,74 EH/s de capacidade arrendada.
• Após o encerramento do trimestre, a Cango concluiu a infraestrutura em sua unidade Georgia LN com capacidade para suportar até 3 megawatts para operações de IA e assinou seu primeiro contrato com cliente de IA. A administração espera que uma receita modesta relacionada à IA comece no 3º trimestre de 2026.
• A Cango começou a implementar um programa de hedge de Bitcoin durante o 2º trimestre. A administração o caracterizou como uma ferramenta de gestão de risco destinada a reduzir a sensibilidade do fluxo de caixa às variações de preço do Bitcoin, e não como posicionamento especulativo.

Principais Dados Financeiros

Desempenho Operacional e dos Negócios

A Cango continuou a retirar máquinas de mineração legadas de menor eficiência e a priorizar a rentabilidade unitária em detrimento da escala. A estrutura de leasing transfere os custos operacionais diretos associados à taxa de hash arrendada para o arrendatário, reduzindo a exposição da Cango a despesas variáveis.

A menor capacidade de automineração e a transferência de parte da capacidade para leasing reduziram a produção de Bitcoin sequencialmente. No entanto, despesas menores com eletricidade e hospedagem reduziram o custo da receita para US$ 50,7 milhões, ante US$ 99,6 milhões no 1º trimestre.

A administração afirmou que a maioria das máquinas de mineração está hospedada em locais de terceiros. Alguns contratos de hospedagem incluem mecanismos que reduzem os preços de energia quando os preços do Bitcoin caem, oferecendo uma margem de proteção de custos na baixa. Os custos em caixa de mineração também caíram mês a mês durante o 2º trimestre.

Entre as máquinas operacionais de automineração em rack, excluindo a capacidade arrendada, um pouco mais de um terço eram unidades da série S21. A Cango planeja continuar descontinuando a capacidade legada menos eficiente.

O negócio de infraestrutura de IA da Cango avançou após 30 de junho e, portanto, não contribuiu para a receita do 2º trimestre. A infraestrutura de IA da unidade Georgia LN foi concluída no início de julho com capacidade para suportar até 3 megawatts. A instalação de contêineres e entregas graduais de GPUs estavam em andamento no momento da teleconferência.

A empresa assinou seu primeiro contrato com cliente de IA após o encerramento do trimestre. Seu modelo planejado inclui hospedagem bare-metal de GPU e colocation, embora nenhum contrato formal de colocation tenha sido assinado. A Cango também possui nós de teste no Texas e na Costa Oeste e está avaliando locais adicionais.

Perspectivas da Administração

A administração espera que o reconhecimento de receita relacionada à IA comece no 3º trimestre de 2026. A expectativa é que a contribuição inicial seja modesta, enquanto as conversas com clientes continuam em andamento.

A Cango espera que a taxa de hash operacional e o saldo de máquinas de mineração permaneçam amplamente estáveis durante o 3º trimestre. No entanto, a administração observou que o racionamento regional de energia em julho e agosto pode afetar as operações.

Para o segundo semestre de 2026, as prioridades da administração são otimizar o mix de taxa de hash de automineração e arrendada, executar a implantação de IA, adicionar clientes e avaliar uma expansão adicional de locais, mantendo a disciplina de capital.

Riscos e Pontos de Atenção

• A volatilidade do preço do Bitcoin continua sendo um risco direto para os fluxos de caixa de mineração. O programa de hedge da Cango visa reduzir, e não eliminar, essa exposição.
• O custo total (all-in) de mineração no 2º trimestre, de US$ 98.405 por Bitcoin, permaneceu acima do custo em caixa informado porque incluiu a depreciação das máquinas de mineração.
• A desativação contínua de máquinas mais antigas pode levar a efeitos adicionais de reestruturação, embora as perdas por impairment e alienação no 2º trimestre já tenham totalizado aproximadamente US$ 51 milhões.
• Os racionamentos de energia no verão podem afetar a atividade de mineração no 3º trimestre, mesmo se a taxa de hash operacional instalada permanecer estável.
• A comercialização de IA está em estágio inicial. A receita inicial contratada é pequena e nenhum acordo formal de colocation havia sido concluído no momento da teleconferência.

Destaques das Perguntas e Respostas dos Analistas

A administração explicou que o hedge de Bitcoin está estruturado como um empréstimo de curto prazo denominado em BTC. O saldo no encerramento do trimestre era de aproximadamente US$ 8 milhões, com um ativo circulante aproximadamente equivalente registrado. A Cango vende o Bitcoin tomado por empréstimo ao preço à vista no primeiro dia e pode quitar o empréstimo usando o Bitcoin minerado posteriormente se os preços caírem. O dimensionamento da posição é geralmente baseado em um ou dois meses de produção.

Sobre as operações de mineração no 3º trimestre, a administração disse que a taxa de hash e o estoque de máquinas de mineração não devem mudar significativamente, sujeitos a possíveis racionamentos de energia no verão. As negociações de contratos de hospedagem e os mecanismos de ajuste do preço de energia atrelados ao Bitcoin podem proporcionar maior flexibilidade de custos.

Em relação à capacidade de IA, a administração afirmou que o foco no curto prazo é a unidade própria de 50 megawatts Georgia LN. Pequenos nós de teste também foram instalados em locais de parceiros, mas a administração não quantificou quanto da infraestrutura de mineração poderia ser convertida para uso em IA nos próximos três anos.

Transcrição Completa da Teleconferência de Resultados

Transcrição completa da teleconferência de resultados

Comentários da administração

Operator

Good day, and welcome to the Cango Inc. Second Quarter 2026 Earnings Conference Call.

[Operator Instructions]

Please note today's event is being recorded. I'd now like to turn the conference over to Paul Yu, CEO. Please go ahead.

Peng Yu

Thank you. Hello, everyone, and thank you for joining Cango's Second Quarter 2026 Earnings Call. Let me start with a quick overview of the quarter. On the mining side, we deliberately scaled back operations as planned. That's reflected our second quarter results. On the AI side, since the end of the second quarter, we have made real progress on infrastructure and signed our first customer contract, moving that business from build-out into commercialization. I will be clear that these AI developments occurred after June 30 and are not reflected in this quarter's reported results.

In terms of the numbers, total revenue for the quarter was approximately $50.8 million with about $47.4 million coming from Bitcoin mining. Net loss was approximately $81.6 million, mainly driven by noncash impairment and disposal losses on our mining machines, a direct result of the deliberate restructuring of our asset base. As of June 30, we held 1,056 Bitcoins. In addition, our cash, cash equivalents and cryptocurrencies totaled approximately $23 million, while long-term debt was approximately $31.2 million.

Now let me walk through the mining business and AI infrastructure business in more detail. This quarter, we continued to actively rightsize our mining operations, disposing of machines with lower marginal efficiency and introduced a leasing model to shift our focus from scale to economics. As of June 30, our self-mining hashrate was 19.84 exahashes per second, and our lease hashrate was 7.74 exahashes per second for a combined operating hashrate of 27.58 exahashes per second. Under the leasing arrangement, the lessee bears the direct operating costs associated with the hashrate, which also reduces our exposure to variable costs. We mined 656 Bitcoins this quarter.

Production was down sequentially, largely reflecting the deliberate reduction in self-mining capacity and shift of some capacity into leasing. We will continue to evaluate the mix between self-mining and leasing based on economics rather than scale. We will keep phasing out less efficient legacy capacity. This quarter, we also began implementing a hedging arrangement to manage our exposure to Bitcoin price volatility, thus enhancing the predictability of our operating cash flows. Our average cash mining cost in Q2 was $73,313 per coin, down about 5% from Q1.

Now let's turn into AI infrastructure. A quick on timing, everything I'm about to cover took place after June 30 since the start of the third quarter. So it isn't reflected in the quarter's financial results, but we want to share it with you. On infrastructure, construction at our Georgia LN site was completed in early July with the site infrastructure able to support up to 3 megawatts, leaving room for future expansion. Container units have arrived on site and are being installed and GPUs are arriving on site in batches. On the customer side, since the start of the third quarter, we've signed a customer contract and discussion with prospective customers are ongoing. That takes our AI business from technical validation into commercial monetization. This is a development since quarter end. Contracted revenue is still small, and we expect to begin recognizing related revenue in the third quarter.

On the business model, we expected to pursue both bare-metal GPU hosting using our existing site and power infrastructure to offer a standardized deployment environment and colocation intended to improve overall infrastructure utilization. We haven't signed a formal colocation contract yet and terms are still being worked out. We also have test nodes in Texas and on the West Coast, mainly to support customers who need deployment closer to their location in the future. We are evaluating several potential new sites as well, and we haven't ruled out building our own. We will continue to run mining and AI as parallel businesses.

Looking into the second half, our priority are managing the mix of self-mining and lease hashrate prudently, executing our AI deployment and continuing to sign new customers and building on the operating experience from Georgia as we evaluate further site expansion. Capital discipline and operating efficiency remain our priorities.

That concludes my remarks. I will now turn it over to our CFO, Simon, for a detailed review of the financials. Thank you.

Ming Yeung Tang

Thanks, Paul. Hi, good morning. Hi, everyone, and welcome to our second quarter 2026 earnings call. Before I start to review our financials, please note that unless otherwise stated, all amounts discussed are in U.S. dollars.

Total revenues were $50.8 million. Revenue during the quarter from the Bitcoin mining business was $47.4 million with a total of 656 Bitcoins mined during the period. The average cost to mine Bitcoin, excluding depreciation of mining machines, was $73,313 per Bitcoin and all-in cost of $98,405 per Bitcoin. Compared to the first quarter of 2026, total revenue decreased by approximately 50%. This decline primarily reflects our proactive reduction in operational hashrate as we continued to selectively phase out older, less efficient S19 series mining machines and temporarily transitioned some capacity to a hosted leasing model. While this adjustment has reduced our top line mining revenue, it has also significantly lowered our operating costs and improved our cash flow profile. And some of these efforts continued throughout the second quarter.

Now let's move on to our cost and expenses. Cost of revenue, exclusive of depreciation was $50.7 million, down from $99.6 million in the first quarter, driven by lower electricity and hosting expenses following the hashrate reduction. Depreciation was $16.9 million, down from $29.4 million in the first quarter. General and administrative expenses, including related party fees, totaled $8.4 million. Impairment loss from mining machines in the second quarter was $42.9 million and loss on disposal of mining machines in the second quarter was $8.5 million. Loss from changes in the fair value of crypto assets was $4.1 million compared with a loss of $151.8 million in the first quarter.

The change was primarily driven by 2 factors: the decrease in Bitcoin prices as of June 30, and this was partially offset by the implementation of our hedging program. As Paul mentioned earlier, we began implementing a Bitcoin hedging program during the second quarter. The purpose of this program is to manage our exposure to Bitcoin price volatility and provide greater predictability to our operating cash flow. We intend to selectively continue to use hedging as a risk management tool, and this is not for speculative purposes. The related short-term positions are reflected on our balance sheet and will be reflected as we continue to execute this program in a disciplined manner.

Operating loss for the quarter was $80.6 million with a net loss from continuing operations of $81.6 million in the second quarter. The net loss was primarily driven by the noncash impairment and disposal losses I just mentioned, which together totaled approximately $51 million. On a non-GAAP basis, adjusted EBITDA was a loss of $10.7 million, including a $4.1 million loss from the changes in the fair value of the receivables for the Bitcoin collateral.

Lastly, moving on to our balance sheet. As of June 30, we had cash and cash equivalents of $10.1 million compared with $7.2 million as of March 31. At the same time, our balance sheet also has Bitcoins in the number of 1,056 Bitcoins held in treasury. In terms of operational assets, we carry our mining machines at a net value of $58.7 million after depreciation. On the liability side, we had $31.2 million in long-term debt compared with $30.6 million as of March 31.

And this concludes our prepared remarks. Operator, we are now ready to take questions.

Operator

[Operator Instructions]

And today's first question comes from Pingyue Wu with Citic Securities.

Perguntas e respostas

Pingyue Wu

I have 3 questions. First, can management provide more color on the Bitcoin hedging program in terms of overall notional size, instrument structure and duration? And additionally, could you clarify whether this is risk mitigating or it involves any directional positioning?

And my second question is regarding the AI infrastructure progress you highlighted such as the Georgia site completion and container deployment. We think it is a milestone occurred towards the second quarter? And what is the rationale for including them now? And more importantly, could we incorporate this development as material increase in our third quarter financial models?

And my third question is regarding the newly signed customer contracts. Could you provide some visibility into the anticipated revenue contribution and time line for top line recognition?

Ming Yeung Tang

Thanks, Pingyue. It's Simon here. Why don't I take the first question and then Paul can address your second and third questions with regards to the AI progress. In terms of the hedging program, it's structured as a short-term loan denominated in BTC. So that is reflected in our balance sheet under short-term debt, which as of quarter end was around USD 8 million. And at the same time, there is a roughly equivalent amount recorded under current asset as well. So this short-term loan in BTC is led to us on day 1 and then which we typically size based on the scale of our Bitcoin mining production. For example, we might want to think about, okay, we'll do 1 month of production or 2 months of production. So that's the way we think about this.

And then this loan in BTC is sold at spot price on day 1. So if in the coming months, if Bitcoin prices fall below that, then we'll choose to repay in the BTC that is mined out of our mining operations. So I hope that illustratively addresses your question with regards to the -- how we think about the sizing and the structure. And again, I would like to emphasize that we purely think of this as a risk management tool and the purpose is really just to reduce the sensitivity of our cash flow to the Bitcoin price ranges. And then with that, I'll pass it to Paul for the second and third question.

Peng Yu

Sure, sure. Thank you. We wanted to give you the most current picture of where the AI business stands. Even though this development fall after June 30 cutoff, we are not reflected in this quarter's revenue and only a small amount of property-related costs have been capitalized in Q2. The amount is immaterial. We expect the related revenue to start showing up in our third quarter numbers, which we will report in the normal course. And that means we expect to begin recognizing AI-related revenue in the third quarter. The initial contribution will be modest, but it provides initial validation of the commercial viability of our AI infrastructure strategy and establishes an operating track record we can build on. Thank you.

Operator

[Operator Instructions]

Our next question today comes from Sid Rajeev with Fundamental Research Corp.

Siddharth Rajeev

Should we expect Q3 mining revenue to stabilize at current levels or anticipate further hashrate reductions?

Ming Yeung Tang

Sid, thank you for your question. In terms of the operational hashrate and the mining machines that we have on our balance sheet, in the third quarter, it would not change significantly -- it will not change significantly. But again, given the third quarter includes the summer months of July and August, whereby we may experience some regional power curtailment.

Siddharth Rajeev

Got it. Maybe you could provide some color on roughly how much of the current hashrate is from S19 versus newer generation machines?

Ming Yeung Tang

This percentage is increasing. In terms of the mix between the 19s and the 21s, I would say -- and this is purely the amount that is operational that is on rack and excluding -- let me think about how to address this. Excluding the part that is leased, the split is roughly a little bit above 1/3 in the 21 series.

Siddharth Rajeev

Got it. Are you able to talk about your cash costs? Can you further cut costs? Because I see you did have cost reductions in the quarter. So how about Q3, how should we look at it?

Ming Yeung Tang

Yes. Sid, and I think that is a great question. And the reason that in the second quarter, the cost continued to optimize. There were 2 reasons. One reason was that we were -- we continue to negotiate with our hosted sites because as you remember, most of our sites are externally hosted instead of our self-owned mining sites. Our own self-owned mining site is just a 50-megawatt site in the state of Georgia in LN. And the rest of our mining machines are hosted externally with third parties. So we continue to negotiate contracts with them.

And a lot of these contracts have a power price reduction mechanism, whereby the power prices would decrease in an environment where Bitcoin prices are decreasing as well. So if we were to look at the cash cost on a month-by-month basis between each month of the second quarter, the cash cost was on a downward trend. So this is, in a way, is a price reduction mechanism to give us a little bit more downside protection.

Siddharth Rajeev

Got it. If I may, one more question. This is slightly more long term. How much of your existing, say, mining infrastructure or power capacity could realistically be converted to AI infrastructure over the next 3 years?

Ming Yeung Tang

We're starting in the U.S. at the moment. We're still more focused on our own 50-megawatt site right now, but we have started to install small test nodes in other sites. But these are sites that are not necessarily our own, but they could be with partner sites.

Operator

And that does conclude our question-and-answer session. I'd like to turn the conference back over to the management team for any closing remarks.

Ming Yeung Tang

Any other remarks? Thank you very much for dialing for our conference call. Thank you.

Operator

Thank you, sir. That does conclude our conference for today. We thank you all for attending today's presentation. You may now disconnect your lines, and have a wonderful day.