Bitmain vs MicroBT: The Real Cost of Owning a Miner

Bitmain’s newest flagship, the Antminer S23 XP Hyd, is rated at 8.9 joules per terahash and is shipping from November. That makes it the first Antminer to break below nine joules, nudging the company back ahead of Bitdeer’s SEALMINER A4 Ultra (9.45 J/TH) and its own S23 Hydro (9.5 J/TH) at the top of the efficiency table, according to the launch coverage at Million Miner and the live spec sheet on ASIC Miner Value. Headlines will frame the Bitmain versus MicroBT contest the way they always do: as a race to the lowest joule-per-terahash.

For the operator actually writing the check, the number on the box is the cheapest part of the decision. Two machines with similar efficiency can cost very different amounts to own across three years once you add street price, cooling plumbing, firmware, downtime, repairs, and what the rig fetches on the secondary market when you are done with it. In a market where Bitcoin trades near $83,000 and is down about 4% on the week, hashprice sits flat around $39.61 per petahash per day, and difficulty hovers near 132.7 trillion with a small uptick expected at the mid-October retarget, margins are thin enough that total cost of ownership, not the spec sheet, is where the Antminer versus Whatsminer question is settled.

This is the fourth piece in our Bitmain versus MicroBT series, and the one about the bill rather than the brochure. We have already covered who these two companies are, the sub-10 J/TH efficiency race, and where the machines are built. Here we open the box and add up what it really costs to run one.

The duopoly in sixty seconds

Two companies build almost every Bitcoin miner on earth. Bitmain, founded in Beijing in 2013, makes the Antminer line and runs Antpool. MicroBT, founded in Shenzhen in 2016 by Yang Zuoxing, a former Bitmain chip-design lead, makes the Whatsminer line. The University of Cambridge’s 2025 digital mining industry report put Bitmain at roughly 82% of ASIC production, MicroBT at about 15%, and Canaan at around 2%, which leaves the top three controlling about 99% of the market, per the Cambridge Digital Mining report.

There is a third name worth holding in mind. Bitdeer, the Nasdaq-listed vehicle tied to Bitmain co-founder Jihan Wu, now designs its own SEAL chips and ships SEALMINER rigs, so the founder feud that split Bitmain a few years ago has produced a genuine third contender. Both Bitmain and MicroBT fabricate their chips at TSMC, which means the entire network’s hardware pipeline runs through one Taiwanese foundry. Those are the players. The rest of this piece assumes you know them and focuses on what happens after you decide to buy.

The spec sheet is the cheapest part of the bill

Total cost of ownership for a mining rig has three buckets. There is capital expenditure: the purchase price, the power supply, the cooling infrastructure, shipping and any import duty. There is operating expenditure, which is dominated by electricity but also includes hosting fees, maintenance, replacement parts and pool fees. And there is terminal value, meaning whatever you can sell the machine for when a newer model or a difficulty spike pushes it off your floor. The joule-per-terahash figure on the marketing page touches only one of those buckets, the power slice, and even then only the part of your bill that scales with efficiency.

Fred Thiel, chief executive of MARA Holdings, put the underlying economics plainly to CoinDesk: “Bitcoin mining is a zero-sum game. As more people add capacity, it gets harder for everybody else. Margins compress, and the floor is your energy cost.” Energy is the floor. But above that floor, the gap between two operators running similar-efficiency machines comes down to everything the spec sheet leaves out: what they paid per terahash, how much infrastructure the cooling demanded, how often a board failed, how long the rig stayed profitable, and what it was worth at the end. Those are the differentiators this piece works through, one at a time.

Sticker price versus street price: what a terahash actually costs

Miners do not price machines in dollars; they price them in dollars per terahash, because that is the unit that lets you compare a 258 TH/s air-cooled box against a petahash-class hydro chassis. On that basis the efficiency premium is stark. The S23 XP Hyd lists at roughly $25.5 per terahash, and the S23 Hydro came to market at a similar number after coupons, per Forbes. MicroBT’s air-cooled Whatsminer M70S, by contrast, works out to about $18.80 per terahash in Luxor’s Hashrate Index hardware review. The most efficient Antminers cost roughly a third more per terahash up front than a 13.5 J/TH Whatsminer. You are prepaying for a lower power bill, and that trade only works out if your electricity is expensive enough and your runtime long enough for the saved joules to clear the price gap.

List price is also not street price. Mining journalist Colin Harper, writing for Forbes when the S23 was unveiled, noted that compressed post-halving margins and higher import costs had “flipped a seller’s market into a buyer’s market,” with Bitmain, long used to holding the upper hand on price, now offering payment installments and cutting deals with large buyers such as CleanSpark and Hut 8 to stimulate demand. That matters because the number you negotiate, not the number on the web shop, is the one that lands in your TCO model. Whatsminers have historically launched 10% to 20% cheaper than equivalent Antminers, according to a 2026 brand comparison by D-Central, though as we will see that discount narrows once resale is factored in.

There is also the question of new versus used, which is really a question of counterparty risk. The secondary market is deep, but it is also where stolen, overclocked-to-death and misrepresented machines change hands, so a used rig at a tempting price can carry hidden board damage or a voided warranty. Buying new from the manufacturer or an authorized reseller costs more but comes with firmware provenance and a warranty; buying used demands escrow, testing and a trusted broker. For a first fleet, the safety of new hardware is often worth the premium; for an operator who can test and repair in house, the used market is where the cheapest terahash lives.

The late-2026 lineup you are actually choosing between

Before the economics, the hardware. Here is the flagship field as it stands heading into the end of 2026, drawn from ASIC Miner Value and the manufacturers’ own specifications. Treat every figure as nameplate: real-world hashrate and efficiency vary with firmware, ambient temperature and how hard you push the machine.

The pattern is clear. Bitmain and Bitdeer own the top of the efficiency table, clustered around 9 J/TH, while MicroBT’s best hydro machine sits at 12.5 J/TH and its volume sellers at 13.5. That is a gap of roughly three to three and a half joules per terahash at the frontier. It is also a smaller gap than it looks on a profit model, because efficiency gains have become incremental: the network keeps getting more efficient in aggregate even as its total power draw stays roughly flat, a dynamic we unpacked in our look at how Bitcoin hashrate grew in 2026 while the power bill did not. For the buyer, the lineup is the menu. The price of each dish, and the cost of the kitchen you need to serve it, is where the real decision lives.

Firmware: the hidden performance layer, and the lock-in

A mining rig is not a sealed appliance. Custom firmware, installed in place of the manufacturer’s stock software, can reshape the economics of a machine you already own. Packages such as Braiins OS+, LuxOS and VNish add per-chip voltage tuning, dynamic power targeting and autotuning that can cut effective joules-per-terahash by anywhere from 10% to 25%, and just as importantly can keep an older machine profitable longer by letting you dial power up or down as hashprice moves. This is where Bitmain and MicroBT diverge in a way the spec sheet never shows.

Antminer has the richer aftermarket ecosystem by a wide margin. Braiins OS+ and LuxOS are built primarily for Antminers, and VNish is the main cross-brand option, supporting Antminer models and a subset of Whatsminers such as the M30S line. MicroBT’s newer hardware has thinner third-party support, and the company leans instead on power modes built into its stock firmware, which reduces the need for aftermarket software but also caps how far you can push a machine, according to the D-Central brand comparison. If your operating edge is squeezing extra efficiency out of rigs with tuned firmware, that capability is worth real money, and it tilts toward Bitmain.

The catch is lock-in. Bitmain has steadily tightened control over its boards, and the timeline matters if you plan to run custom firmware. A 2022 security update disabled SSH access; from March 2024, Amlogic control boards blocked direct custom-firmware installs and had to be opened with a roughly $85 unlock kit; and a September 2025 stock update blocked the micro-USB port that kit relies on, so boards shipped after that must first be downgraded via SD card to the August 2025 build before they can be unlocked, a process that runs about two hours per machine, as tracked by D-Central’s firmware lock tracker. On both brands, installing third-party firmware voids the manufacturer warranty. There is a quieter governance question buried in all of this, too: running tuned firmware and a remote fleet-management stack means trusting whoever controls the update and autotune keys to thousands of machines, a control-plane trust problem not unlike the one playing out in DAO security councils. The TCO takeaway is a fork: if you want to tune aggressively, budget for Antminer plus the unlock friction; if you want set-and-forget operation with the warranty intact, MicroBT’s stock power modes are the lower-effort path.

Cooling: the capex you do not see on the box

Here is the trap that catches first-time buyers. Every flagship in the table above, the sub-9 and sub-10 machines from all three makers, is hydro-cooled. You cannot bolt an S23 XP Hyd, an A4 Ultra or a Whatsminer M73S+ into an air-cooled shed and switch it on. Hydro machines need dry coolers, manifolds, pumps, plumbing and water treatment, which in practice means a purpose-built facility and often hundreds of dollars per kilowatt of extra capital before a single terahash is produced. Immersion machines need tanks and dielectric fluid. The most efficient rig on the market can quietly carry the most expensive infrastructure around it.

That is why the air-cooled models still matter. The air S23 at roughly 318 TH/s and 11 J/TH, or the Whatsminer M70S at 258 TH/s and 13.5 J/TH, are less efficient, but they plug into existing sites, retrofits and small or home setups without a cooling retrofit. Both manufacturers ship air, hydro and immersion variants, so the cooling decision is really about your site rather than the brand. But it belongs in the TCO model at the top, not the bottom: on a greenfield build, the facility capex for liquid cooling can dwarf the per-rig price difference between an Antminer and a Whatsminer. The top end of the market, from Bitmain’s 1.16-petahash 3U chassis to MicroBT’s roughly one-petahash M79S, is now rack-scale hydro aimed squarely at operators who have already made that infrastructure commitment.

Power and hosting: the opex that dwarfs the rig

Over a machine’s life, electricity usually swamps every other line item, often the majority of the total cost of running it. This is where the efficiency number finally earns its keep, but it is also where the choice between self-hosting and colocation reshapes the whole model. Run rigs yourself and you carry the power contract, the substation, the staff and the curtailment risk directly. Pay a hosting provider and you typically hand over an all-in rate per kilowatt-hour that bundles power, cooling, maintenance and uptime guarantees, trading a lower headline power price for a predictable one. Either way, the less efficient Whatsminer burns more joules for the same hashrate, so its lower purchase price is slowly repaid to the utility, month after month, for as long as it runs.

The efficiency number translates into opex through a single figure every operator should know: the breakeven power price, the cost of electricity at which a machine stops making money at a given hashprice. A more efficient rig tolerates a higher power price before it goes cash-negative, which is exactly why large miners chasing the lowest cost per Bitcoin pay up for sub-10 J/TH hydro machines. At a hashprice near $40 per petahash per day, the frontier Antminers stay profitable at power prices that would already have idled a 13.5 J/TH machine. If your power is genuinely cheap, that headroom matters less and the cheaper, less efficient box can be the smarter buy; if you are exposed to grid rates, efficiency is survival.

Two subtler opex lines decide close calls. Pool fees skim a small percentage off every block of revenue, and they apply regardless of brand, so they rarely tip the decision. Curtailment does. Operators on demand-response programs or volatile wholesale power switch machines off when prices spike, and a rig with flexible firmware power modes, or MicroBT’s built-in tiers, can throttle down instead of going fully dark, preserving some revenue. An uptime guarantee in a hosting contract, the difference between 95% and 99% availability, can swing annual revenue by more than the efficiency gap between two machines. None of this appears on a spec sheet, and all of it belongs in the model before you compare two miners on joules alone.

Uptime, failures, and the RMA you hope never to use

Reliability is a cost line, not a nice-to-have. A rig that is down earns nothing while difficulty keeps climbing, so every hour offline is lost revenue you never recover. This is the dimension where MicroBT tends to win. Field data compiled by D-Central points to lower failure rates over long runs for Whatsminers, with M30S++ units documented past 30,000 hours without a hashboard failure and more consistent batch-to-batch quality. Antminers, by the same account, see comparatively more hashboard-chip and fan failures.

The twist is in the repair, not just the failure. Antminer uses an external power supply, the APW series, so a dead PSU is a part swap rather than a whole-machine return, which cuts repair cost by an estimated 20% to 30% and lets you upgrade the supply independently, according to the same D-Central comparison. MicroBT integrates the PSU into the chassis, which simplifies the enclosure but means a supply failure can take the whole unit out of service. Both brands offer standard warranties in the six to twelve month range, and both now run US repair hubs, MicroBT through its Luxor-backed US operation with repair centers on both coasts, Bitmain through its domestic line, which matters more than it used to now that cross-border RMA runs into the tariff wall we covered in the reshoring piece.

Luxor research associate Kaan Farahani, reviewing the M70S for Hashrate Index, framed MicroBT’s pitch around exactly this: at roughly $18.80 per terahash, the machine is sold less on record efficiency than on operational stability and fleet-level reliability. For a large operator, a machine that runs quietly for years with predictable failure rates can beat a slightly more efficient box that needs more hands-on attention. For a small operator with no spare rigs, downtime hurts even more. Either way, uptime belongs in the model with a dollar figure attached.

Spares strategy is the practical expression of all this. Serious operators keep a share of their fleet as cold spares and stock replacement hashboards, fans and power supplies, because mean-time-to-repair, not just mean-time-to-failure, determines how much revenue a fault actually costs. Here Antminer’s external PSU and larger parts market help, while MicroBT’s lower failure rate means fewer events to repair in the first place. The right answer again depends on scale: a single home miner cannot justify a shelf of spares, so reliability out of the box matters most, whereas a large farm can absorb failures with inventory and staff and will weigh efficiency and resale more heavily.

Resale value and the depreciation cliff

ASICs rarely depreciate because they wear out. They depreciate because the network’s hashrate and difficulty climb and because newer, more efficient models arrive, pushing older machines toward the margin. That makes terminal value one of the least predictable and most important lines in the whole model. As a rule of thumb drawn from secondary-market data at ValueHash, a current-generation machine can lose 30% to 50% of its value in the first year, and the launch of a next-generation flagship can knock 20% to 40% off current-gen resale almost overnight. The arrival of the S23 XP Hyd is precisely that kind of event for older S21 and even S23 Hydro units. On the other hand, a twelve-month-old rig in good condition can still fetch 50% to 70% of its purchase price if Bitcoin has risen and new supply stays constrained.

Brand matters here in a way it does not on the spec sheet. Antminers command a secondary-market premium of roughly 10% to 20% over equivalent-spec Whatsminers, according to D-Central, thanks to stronger brand recognition, a larger buyer pool, the aftermarket firmware that extends their useful life, and the modular external PSU that makes them easier to resell in parts. That premium quietly claws back much of MicroBT’s lower launch price. The table below is illustrative, not a quote: it models a $15,000 hydro rig under a flat-Bitcoin, rising-difficulty scenario, applying the depreciation ranges above, to show how the curves can diverge.

Read it as a shape, not a forecast. The point is that if you flip rigs or run short cycles, the resale premium leans toward Antminer; if you intend to run a machine into the ground on genuinely cheap power, the lower entry price and the reliability record lean toward Whatsminer. Your holding period changes the answer.

One more resale wrinkle rewards attention: timing. Because a next-generation launch resets the whole curve, the worst moment to sell is right after a new flagship ships, and the best is into a rising Bitcoin price when new supply is backordered and buyers cannot get current-gen machines from the factory. Operators who treat rigs as assets to be rotated, selling before the next efficiency step lands and redeploying into newer hardware, capture more of the curve than those who hold until a machine is worthless. That discipline, more than the badge on the box, often separates a profitable fleet from a stranded one.

Financing a fleet: the rig as collateral

Few operators pay cash for a fleet. Most use equipment financing or ASIC-backed loans, and that turns the choice of machine into a financing decision as much as a technical one. Lenders value rigs on dollars-per-terahash, efficiency tier, age and, crucially, how liquid the machine is on the secondary market. A rig with a deeper resale market and a better efficiency position makes better collateral, which can translate into a higher loan-to-value ratio and a lower cost of capital. That is another channel through which Antminer’s resale premium pays for itself, and another reason MicroBT’s cheaper sticker is not the whole story.

Because ASIC prices track hashprice and Bitcoin, lenders keep loan-to-value conservative; a sharp difficulty jump or a drop in the Bitcoin price can erode collateral value fast and trigger margin calls, which is how fleets end up liquidated into weak markets. Capital structure also separates the players. Bitdeer is publicly listed and can tap equity and debt markets directly, while Bitmain and MicroBT remain private. The large public miners, from MARA to Riot to CleanSpark and Hut 8, increasingly fund rig purchases through convertible debt and equity raises, so their true cost of ownership includes a cost-of-capital layer that a self-funded home miner never sees. The price on the invoice is only the beginning of the number.

Depreciation also has a tax dimension that feeds back into TCO, at least for US operators. Mining hardware is a capital asset that can be written down over time, and the schedule a miner uses changes the after-tax cost of the machine, sometimes materially. That is an accounting decision rather than a hardware one, but it means two operators buying the identical rig can carry very different real costs depending on how they are structured and financed. The lesson repeats: the sticker price is an input, not the answer.

The third force and the AI overhang

The duopoly is not quite a duopoly anymore, and that is good news for buyers. Bitdeer’s SEALMINER program climbed from the A2 at around 16.5 J/TH to the A4 Ultra at 886 TH/s and 9.45 J/TH, launched in April 2026, and because Bitdeer is US-listed and building domestic capacity, it offers a way to diversify both single-vendor risk and tariff exposure. A third credible supplier is the simplest force keeping Bitmain and MicroBT honest on price, and the numbers in the lineup table suggest it is working.

The larger shadow over all three is artificial intelligence. Bitmain’s AI-chip affiliate Sophgo was added to the US Entity List in January 2025 after a TSMC-made chip matching a Sophgo design was found in a Huawei AI processor, which means Bitmain’s ambitions in AI silicon are export-control-exposed and could compete internally for the engineering and foundry capacity that goes into miners. At the same time, the miners themselves are chasing AI and high-performance computing revenue, as Thiel’s argument that miners must become power companies implies, with the likes of MARA and Riot leasing capacity to AI tenants. The same TSMC wafers and the same cheap megawatts that mine Bitcoin are now courted by the broader AI economy, from centralized hyperscalers to decentralized compute networks such as Bittensor and verifiable-inference experiments like opML. For a buyer, that means vendor roadmaps and chip supply are now partly a function of the AI cycle. Choosing a machine in 2026 is an implicit bet that Bitmain keeps shipping leading miners even as it chases AI, and that a third vendor stays strong enough to keep the market competitive.

One more name rounds out the field. Canaan, the third-largest maker at roughly 2% of the market, ships the Avalon line, which rarely tops the efficiency charts but offers another publicly traded, non-Bitmain source of supply and is running early US production trials of its own. For a buyer wary of concentration, a second challenger alongside Bitdeer adds another axis of price competition. For most operators the real decision is still Antminer versus Whatsminer, but the presence of two credible outsiders is part of why the buyer’s market has teeth, and why keeping an eye on all four names, not just the big two, belongs in a serious procurement process.

Where the machine is made still moves the price

Geography bleeds straight into TCO through the landed price. The 2026 US tariff stack pushed duties on Southeast-Asia-assembled rigs to roughly a fifth of the machine’s value and on China-origin machines to more than half, up from low single digits before. Both makers responded by reshoring assembly: Bitmain stood up its first US production line in late 2024 and MicroBT runs a US online shop and warehouse through its Luxor partnership. US-assembled units dodge much of the duty, but assembly is not fabrication, and the chips still come from TSMC, so the single-point-of-failure risk does not disappear. The full treatment, including the Entity List exposure and the politics, lives in our dedicated reshoring piece. For the purposes of this model, the lesson is narrow: the same machine can carry very different landed costs depending on where it ships from, so always price the rig delivered and cleared, not FOB the factory.

A buyer’s decision framework

Put the differentiators together and there is no single winner, only a machine that fits your situation better than the others. The matrix below maps common operator profiles to the brand that tends to fit, and why. It is a starting point for your own model, not a substitute for running the numbers on your actual power price and holding period.

Notice that Antminer shows up most often, which reflects its efficiency lead and resale depth, but Whatsminer wins outright the moment cheap power and operational simplicity dominate your model, which for many operators they do. The right answer is the one your own spreadsheet produces, not the one with the lowest joule count.

What to watch into the end of 2026

A few things will move the calculus over the next quarter. First, whether the S23 XP Hyd’s 8.9 J/TH holds up in volume once machines are in the field from November, rather than on a launch slide. Second, whether MicroBT answers with anything below its current 12.5 J/TH best, and whether Bitdeer’s next SEAL chip keeps the top end contested. Third, the network itself: difficulty is set to tick up at the mid-October retarget per CoinWarz, and total hashrate is pressing against 1 zettahash per second on Hashrate Index, both of which erode the terminal value of every rig already running.

Fourth, price and hashprice around the late-October Federal Reserve meeting, since a stronger Bitcoin lifts resale and loosens the buyer’s-market pressure that is currently working in purchasers’ favor. And fifth, the AI bid for TSMC capacity and cheap power, which will shape how much attention Bitmain and its rivals can spare for the next generation of miners. None of these change the core lesson. The spec sheet sells the machine; the total cost of ownership decides whether it makes you money.

Frequently Asked Questions

Is Bitmain or MicroBT better in 2026?

Neither wins outright. Bitmain leads on top-end efficiency and holds a 10% to 20% resale premium with a richer custom-firmware ecosystem, while MicroBT is cheaper per terahash and tends to show lower failure rates. The better machine depends on your power price, holding period and how hands-on you want to be.

What is the most efficient Bitcoin miner right now?

The Antminer S23 XP Hyd, rated at 8.9 J/TH, is the first sub-9 flagship and sits ahead of Bitdeer’s SEALMINER A4 Ultra at 9.45 J/TH and Bitmain’s own S23 Hydro at 9.5 J/TH. It is shipping from November 2026.

Why do efficient miners cost more per terahash?

Because you are prepaying for a lower power bill. The most efficient Antminers run around $25 per terahash versus roughly $19 for a 13.5 J/TH Whatsminer, and that premium only pays back if your electricity is expensive enough and the rig runs long enough to recover the difference.

Do Antminers hold resale value better than Whatsminers?

Yes. Antminers typically command a 10% to 20% premium over equivalent Whatsminers on the secondary market, helped by brand recognition, aftermarket firmware and a modular external power supply. All ASICs still lose roughly 30% to 50% of their value in the first year as difficulty climbs.

Does custom firmware void the warranty?

Yes, on both brands. Packages like Braiins OS+, LuxOS and VNish can cut effective efficiency by 10% to 25%, but they void manufacturer coverage, and newer Antminer control boards must be unlocked first, which can take about two hours per machine.

Marcus Okafor is a senior mining correspondent at HOGE Wire, covering ASIC hardware, hashrate and the economics of proof-of-work.