Upcoming ASIC miners: what's actually shipping in 2026–2027, and whether to wait

Every few weeks someone asks us the same question in a slightly different form: there's a new machine coming, should I hold off? It is a fair question and the honest answer is usually no — but not for the reason people expect.

The 2026 generation is real. Bitmain, MicroBT, Canaan and Bitdeer have all launched new silicon in the last twelve months, and the best of it is genuinely a step change: 9.45 J/TH against the 13.5 J/TH that was flagship two years ago. But almost none of it is in a warehouse waiting for you. Much of what you can see on manufacturer sites is a futures contract with a delivery window, not stock.

This page is a straight inventory of what has actually been announced, what is actually shipping, what is priced, and what is vapour. Everything below was checked against manufacturer shops and filings in the first week of September 2026. Where a spec is a manufacturer claim with no independent test behind it, we say so.

The short version

• Nothing shipping today is below about 9.45 J/TH. If a listing claims sub-8 J/TH on a machine you can buy now, it is false.
• No Bitcoin ASIC maker has announced a 2nm product. TSMC's N2 capacity is effectively booked out through the end of 2026, and mining is not first in that queue.
• There is no Antminer S25, no S23 Pro, no Avalon A17 and no WhatsMiner M80. All four appear in reseller listings and forum posts. None exists.
• Auradine has left the market. It rebranded to Velaura AI in March 2026 and stopped selling miners. More on that below, because people are still being sold Teraflux units.
• Latest-generation hardware has been getting cheaper at roughly 5% a month through 2026 — but the index rose over the summer, so this is not a trend you can time.

Bitmain: the S23 family

The S23 line launched at Bitmain's own World Digital Mining Summit in Las Vegas in May 2025, not 2026 — a detail worth knowing because it means the platform is now over a year old and the interesting parts are the variants added since. At WDMS Hong Kong on 27–28 August 2026 Bitmain added two more: the S23 XP Hyd and the S23e U2H.

These are Bitmain's own listed prices and delivery windows, read from shop.bitmain.com in the first week of September 2026.

Two things stand out. First, the $25.50 per terahash on the S23 XP Hyd against $11 per terahash on the S23e U2H. You are paying more than double per unit of hashrate for the 8.9 J/TH machine. Whether that is worth it depends entirely on your power price — which is the whole argument of this article and we come back to it at the end.

Second, "sold out" at Bitmain does not mean unavailable. It means Bitmain has allocated its own batch. Distributors hold stock. The S23 Hyd 3U and the air S23 are both on our catalogue now.

The S23 Immersion is not on Bitmain's shop at all. It appears widely in reseller listings. Treat any immersion S23 spec you see with caution and ask the seller where the figures come from.

MicroBT: the M7x series

MicroBT launched the M7x line at Bitcoin MENA in Abu Dhabi on 8 December 2025 — fifteen SKUs across three efficiency bins (12.5, 13.5 and 14.5 J/TH) and three cooling types, plus a 200 kW, 800 V DC off-grid solar mining system that got less attention than it deserved.

Only the M70, M70S and M73 are sold directly by MicroBT. Everything else reaches buyers through distributors, which matters because there is no independent bench test of any M7x machine in existence. Every efficiency figure above is the manufacturer's. That is not an accusation — it is normal at this stage of a product cycle — but it is different from the S19 and S21 generations, which were measured to death by third parties before most people bought one.

MicroBT publishes no roadmap and has never stated a process node for the M7x. The only confirmed MicroBT–3nm datapoint anywhere is a teardown that found a Samsung 3nm gate-all-around die in the older M56S++. Anything you read about "MicroBT 3nm 2026" is analyst inference.

If you are weighing MicroBT against Bitmain more generally, our WhatsMiner guide works through the range as it stands.

Canaan: the A16 series

Canaan announced the A16 at Blockchain Life Dubai on 28 October 2025. Prices below are from Canaan's own shop.

An independent review in April 2026 put the A16's payback at about 17 months at $0.04/kWh, against 21 months for the A15 Pro. That is the clearest like-for-like generational comparison published by anyone this year, and it is worth internalising: a full generation of efficiency improvement bought roughly four months of payback at industrial power. Not nothing. Not transformative either.

On Canaan's Q1 2026 earnings call management said next-generation R&D "has entered the final stage" and that "some projects have recently completed tape-out for technical validation." No node, no hashrate, no date. There is no A16 Pro and no A17. Our Canaan range guide covers the current line-up in more detail.

Bitdeer: the most efficient machine anyone has announced

Bitdeer's SealMiner A4 Ultra Hydro is the headline number of this generation — 886 Th/s at 8,373 W, a claimed 9.45 J/TH ±5%. It is also, at the time of writing, a pre-order.

The SEAL04 chip behind the A4 was released in Q4 2025 with a claimed 6–7 J/TH at chip level. That is not wall efficiency, and it has not been independently verified. The programme has also had a difficult year: Bitdeer disclosed on 10 November 2025 that SEAL04-2 was "significantly delayed," the stock fell around 15%, and a securities class action was filed in the Southern District of New York on 4 December 2025.

The more important caveat is commercial. Bitdeer's SealMiner sales revenue in Q2 2026 was $0.4 million, against $69.5 million a year earlier. It has stopped publishing external-sales figures and, on the evidence, has essentially stopped selling to third parties in favour of deploying into its own fleet. If you want an A4, plan on a long wait and confirm allocation before you pay anything. Our SealMiner guide goes through the chip roadmap properly.

Auradine is gone, and people are still being sold it

This is the finding from our research that will matter most to some readers, so we are putting it in its own section rather than a footnote.

Auradine rebranded to Velaura AI in March 2026 and exited merchant Bitcoin ASIC sales. auradine.com now redirects to velaura.ai, including deep links — the Teraflux AT2880 product page resolves to an archived datasheet with no price and no buy button. The company's stated plan is to work through existing inventory and increasingly deploy unsold units for its own proprietary mining rather than sell them. It raised a $110 million Series A at over a $1 billion valuation, announced 18 August 2026, to build ultra-low-power silicon for AI accelerators. That is now the business.

What this means practically:

• The company has said it will continue to honour warranties and post-sale support. That is a genuine commitment and we have no reason to doubt it.
• But there is no product roadmap and no new firmware pipeline. Whatever your machine does today is roughly what it will do for the rest of its life.
• ASIC Miner Value still lists the AT2880, AH3880 and AI3680 but shows no live seller price for any of them.
• All three are 18–19 J/TH machines. At today's hashprice they show negative daily profit at most retail power rates.

We still list the Teraflux AT2880 and we are not pulling it, because the hardware works and a buyer with very cheap power can make sense of it. But you should buy it knowing it is remnant inventory from a vendor that has left the market, not a current product with a support roadmap behind it. Our Auradine page has the full spec detail.

What about coins other than Bitcoin?

Much less is happening, and one thing is happening that people said could not.

The Antminer X9 deserves a note. RandomX was designed to be ASIC-resistant, and for years that held. Bitmain now sells a commercial RandomX machine with a 180-day warranty. There is no independent bench test of it and no visible evidence of X9 units in Monero's network hashrate, so treat the performance claim as unproven — but the machine exists and is on Bitmain's own shop, which is not something you could have said two years ago.

For Scrypt specifically, the DL1 at 0.149 J/Mh is a real jump on the 0.165 J/Mh that led the field a year ago. Our Scrypt guide covers where that leaves the existing machines, and there is no ElphaPex DG3 or DG Home successor — the newest is the DG2 Mini from October 2025.

Vapourware watch: there is no KS8, no Nexa or Dynex ASIC, and no new Kadena, Handshake or Nervos hardware. If a listing offers you one, it does not exist.

So — buy now or wait?

Here is the data that actually answers it, from the Hashrate Index ASIC price series.

Bitmain listed the 9.5 J/TH S23 Hyd 3U at about $30 per terahash on 1 January 2026. The sub-14 J/TH index sat at $18.38 on 1 September 2026 — roughly a 39% fall in eight months, or about 5% a month. That is the number people usually want.

But look at the two columns again. Both tiers went up over the summer as Bitcoin recovered toward $80,000 and hashprice climbed from a June low near $27/PH/day to about $39/PH/day at the end of August. New-generation ASIC prices do not decay on a schedule; they track hashprice, and hashprice tracks the Bitcoin price. In 2022 they fell 81% in a year. Through mid-2026 they rose.

Which means "wait for the price to drop" is a bet on Bitcoin falling. If that is your view, the consistent action is not to buy a cheaper miner later — it is not to buy a miner. Waiting also costs you the hashrate you did not run, and network difficulty rises underneath you: the same machine mines less every month it sits in a box.

The one case where waiting is clearly right is if you are choosing between a 2024-generation machine at a discount and a 2026-generation machine at full price, and your power is expensive. Efficiency only pays for itself when electricity is a large share of your revenue. The energy hashprice split makes this concrete: a sub-14 J/TH fleet earns about $136 per MWh, while a 25–38 J/TH fleet earns about $51 per MWh. At industrial power around $40–50/MWh the old fleet is roughly break-even and the new fleet is comfortably profitable. At household prices neither works — which is the whole argument of our apartment mining guide.

What we would actually do

Three positions, depending on where you sit.

If you have hosted or industrial power under about $0.05/kWh: buy shipping current-generation hardware now, not futures. The A3 Pro Hydro at 12.5 J/TH and the M70S at 13.5 J/TH are in stock and priced sensibly per terahash. The 8.9–9.45 J/TH machines carry a premium of roughly double per terahash that you will not recover at those power prices.

If you have industrial power between $0.05 and $0.09/kWh: this is where the efficiency premium starts to earn its keep, and where a November S23 XP Hyd allocation or an A4 Ultra Hydro pre-order is defensible — provided you are comfortable paying ahead of delivery for a machine with no independent test data behind it.

If you are buying on domestic power: none of this generation changes your arithmetic. Efficiency improvements shift the break-even electricity price by a couple of cents. Domestic tariffs are two to four times over the line. The fix is not a better machine, it is cheaper power — which is what hosting is for, and what our home-versus-hosted comparison works through with numbers.

Before you commit to anything, run your own figures through the profitability model with your real electricity rate. Every number on this page is current as of early September 2026 and every one of them moves.

Sources: Bitmain, MicroBT, Canaan and Bitdeer manufacturer shops and specification sheets; Bitdeer SEC filings and investor releases; Canaan Q1 2026 earnings call transcript; Velaura AI and Data Center Dynamics on the Auradine rebrand; Hashrate Index ASIC Price Index and hashprice series; ASIC Miner Value. Checked 3–7 September 2026. Manufacturer efficiency claims for the M7x, A16, A4 and X9 series have no published independent verification at the time of writing. Mining income is never guaranteed.