Open in interactive viewer → charts, metric popovers & call review

Canaan Stops Building Chips to Become a Texas Power Landlord

A grid milestone, a buyback authorization larger than the market cap, and a quiet surrender of the ASIC road map — all in one quarter
CAN · Earnings Call · 2026-11-10

Canaan Stops Building Chips to Become a Texas Power Landlord

The headline numbers from Canaan's third quarter read like noise: revenue of $14.5 million inside an $11–15 million guide, net income of $8.0 million, adjusted EBITDA of $22.4 million. Almost none of that profit was earned. It came from “$51.9 million in fair value gains on digital assets and related financial derivatives” — James Cheng, Chief Financial Officer · 2026-11-10. Strip out the mark-to-market on a bitcoin treasury and you have a hardware company whose revenue is roughly a tenth of what it was a year ago, guiding the fourth quarter to $18–22 million. But the quarter's real event had nothing to do with the P&L. Canaan used the call to announce it is no longer going to fund the business that made it famous.

The chipmaker that stopped chasing chips

Nine months ago, this was a company racing down the process node. On the February call, the CEO described a roadmap in motion: “we have already moved to the next generation development for the chips” — Nangeng Zhang, Chairman and Chief Executive Officer · 2026-02-10, with a sub-10 joule target and A16 volume ramp planned for 2026. The September call was still promising deployments of new machines into the mining fleet next year. On Tuesday that roadmap was folded up. Several next-gen ASIC projects are complete or substantially complete, engineering samples are in hand, and management now says the portfolio “can support the market demand over the next several years” — Nangeng Zhang, Chairman and Chief Executive Officer · 2026-11-10 — so “we will no longer rely on sustained large-scale investments and successive product generations to drive the company's growth” — Nangeng Zhang, Chairman and Chief Executive Officer · 2026-11-10. The reason is a cost squeeze that cuts both ways. Rising wafer costs — advanced node capacity being absorbed by AI demand — collide with a demand collapse for mining equipment. The CEO did the math out loud: at roughly $0.023 per terahash per day of mining revenue against $0.015 of electricity, the equipment price that a customer could justify is about $6 per terahash — “well below the cost of any existing industry mining equipment inventories” — Nangeng Zhang, Chairman and Chief Executive Officer · 2026-11-10. Building a better machine at higher wafer prices widens the gap rather than closing it. Little wonder the network hash rate hasn't reclaimed its 2025 peak, or that management is blunt that “we do not have yet seen a meaningful recovery in mining equipment demand” — Nangeng Zhang, Chairman and Chief Executive Officer · 2026-11-10. There is a nice irony worth noting: in May, the CEO told investors that “when the hash price growth to like $40 to $45, then you will observe a significant market recovery” — Nangeng Zhang, Chairman and CEO · 2026-05-19. It never got there. Hash price is now a fading keyword for this company, and the broader mining-adjacent tape agrees — cost per Bitcoin cohorts are down roughly 55% over 90 days and the joules per terahash efficiency narrative is down more than 35%. The market has already stopped paying for more hashes per watt.

The disclosure dam breaks in Texas

For four straight quarters, Canaan refused to name a site. In September the CEO said plainly: “it is not appropriate at this stage to disclose the exact number and the location of the sites, the power capacity for each project or a specific stage of each project” — Nangeng Zhang, Chairman and CEO · 2026-09-08. The reason given was that premature disclosure could jeopardize execution. This week it disclosed everything that matters. The 1-gigawatt power resource project in Duval County, Texas received a conditional Studied Load – Firm classification under ERCOT's Batch Zero process, with a second gigawatt in Zapata County earlier in the queue. Energization is framed around 2027. That is a genuine, verifiable step in a grid interconnection — and it is not a company-private narrative. The Batch Zero theme was the single highest-momentum keyword across the entire market in the prior quarter, and fellow bitcoin-miner-turned-datacenter-host MARA used the same phrase on its own call this week. Canaan is now riding a wave rather than imagining one.

Our objective is to generate positive cash flow from these projects while would support the share repurchase and provide funding for subsequent such power resources development, land acquisition and related activities.

Nangeng Zhang, Chairman and Chief Executive Officer · 2026-11-10
What softens the announcement is how deliberately small Canaan is keeping its ambition. The strategic partner is described only as a U.S. institution with an investment-grade credit profile; confidentiality agreements block anything more. And the CEO explicitly disclaimed the obvious read: “currently we are not focusing on like building big AI HPC data centers” — Nangeng Zhang, Chairman and Chief Executive Officer · 2026-11-10. That is an odd thing to say in the same breath as a gigawatt of AI HPC data centers demand — but it is the honest version. Canaan wants to be the land-and-interconnection developer that sells or partners into someone else's capital, not the operator writing the checks.

A buyback bigger than the company

Here is the number that stopped me. Canaan's market capitalization is roughly $81 million. Its board just approved a share repurchase authorization of up to $100 million, launching with an initial $20 million tranche. The authorization exceeds the entire equity value of the business. The funding question answers itself only partly. Unrestricted cash was $22.8 million at quarter-end; another $50 million sits as restricted collateral for a letter of credit tied to the power projects, taking the total cash position to $72.8 million, up 10.3% sequentially. The $20 million initial program is roughly matched by what an analyst correctly flagged as thin liquidity — “with most of the bitcoin pledged against loans and unrestricted cash floating around about $23 million, how do you guys plan to fund that $20 million buyback” — Sky Moore, Analyst · 2026-11-10. Management's answer was directional, not arithmetic. In September, the same CFO was hedging on a much smaller $30 million program; the escalation is a real change in tone. The balance sheet is the real story here, and it is stranger than the income statement. Canaan still holds roughly 1,875 bitcoin after monetizing Ethereum and part of its BTC stack for about $40 million at ~$79,000 per coin, using $5.4 million of proceeds to retire 13.6 million ADSs. Against an $81 million market cap, that treasury plus a Texas interconnection queue is the entire equity thesis — a levered, illiquid option on power, not a hardware multiple. The pivot's discipline claim is also where the quarter gets slippery. Operating expenses fell 15.5% to $34.2 million, and capital efficiency is now the CEO's organizing principle: “we are shifting our focus from capital expansion to generating returns on capital” — Nangeng Zhang, Chairman and Chief Executive Officer · 2026-11-10. Yet the same call offered two irreconcilable inventory figures — the CFO citing $119 million, the CEO describing “approximately USD 8.2 million, down significantly from USD 25.3 million in second quarter” — Nangeng Zhang, Chairman and Chief Executive Officer · 2026-11-10. One of those is wrong, and when a company is asking investors to trust a multi-year land option, small inconsistencies get expensive.

The verdict

Canaan has done something genuinely unusual: it has stopped investing in the only business it has ever had revenue from, in the same quarter it revealed a real grid milestone and a buyback authorization larger than its own float. Project ABC contributed $1.6 million of non-cash equity income and $1.9 million of actual cash; an 8-megawatt Nordic district-heating project serves 2,800 households for $1.1 million of revenue. These are rounding errors today against a $14.5 million quarter. The bet is legible and the ERCOT classification is verifiable. Everything else — the partner, the capital structure, the ultimate owner of the electrons — is still behind a confidentiality agreement. For a company this small, that gap between announcement and economics is the whole investment case, and management knows it.