Howard Hughes could feel with his hands that the rivets on an aircraft were disturbing the airflow, Henry Quan told his audience, and that kind of intuition made Hughes a billionaire. Then Quan reminded the room how the story ended: the Spruce Goose never flew. For Bitcoin miners who choose hardware and cooling on gut feeling, his conclusion was the same one he gave Hughes. Intuition is not good enough; the work needs science.
Quan introduced himself on stage at Mining Disrupt Miami 2024 (June 24–26, 2024) as vice president of mining equipment at Foundry Digital. He described 30 years in the semiconductor business, including senior roles at AMD and ATI, where he was responsible for Radeon GPUs and console chips. His subject was what he called Mining 3.0: how to stay profitable as hashrate rises and returns fall, by adding up many small gains rather than relying on one big hardware purchase. He called it the sum of parts.
He opened with Sears, Kodak, Amazon and Apple. Sears was the leading catalog company and Kodak the largest camera ecosystem supplier; today Amazon and Apple hold those positions. "If we fail to innovate, we die," he said, and he pointed to the consolidation and compressing margins the mining industry was going through. But innovation, in his view, no longer means simply buying the newest rigs. For the previous eight years that had been a surefire way to mine more Bitcoin and gain efficiency. He argued that it no longer works as well.
New and more doesn't work as well as it used to.
Henry Quan · 2:25
By his account, each new generation now delivers far less: the S9 brought a tenfold efficiency gain eight years earlier, the S21 only about 20%, and new ASICs cost much more. Rising rig prices, network hashrate growth and the halving had left miners facing a payback of about two years on new machines, and the extra capex of immersion stretched that payback by 30%.
Intuition versus science
Quan placed much of the industry's cooling work in the Hughes camp. "Like Howard, a lot of crypto developments such as immersion are built on intuition rather than science," he said. The alternative he offered was the way aircraft have improved: jets have flown for decades, yet engineers keep finding efficiency through hundreds of small improvements that add up to large ones.
Hey, Howard, intuition isn't good enough.
Henry Quan · 4:57
Why older 7nm chips can compete
One slide compared a miner built on older 7-nanometer chips with the 5-nanometer chips in the S19 XP. Both sat in the range of 22 joules per terahash. The older design matched the XP because it used three times as many chips, each running at lower power, which Quan said brings two advantages.
The benefit of using more ASICs is lower cost and superior tuning because each chip is using less power.
Henry Quan · 3:21
He made the same point about choosing configurations. The S19 XP uses the same ASIC as the S19 k Pro, he said, but runs its chips 20% slower and at lower power, so an XP run at k Pro settings gives 20% more hashrate at the same efficiency. He took that as evidence that high-chip-count rigs are better for tuning, and he warned against buying first-generation hardware just to be early. "Pioneers get arrows in their backs," he said; version two of a product is usually better than version one.
Small gains in power and wiring
Beyond firmware tuning, Quan listed changes around the rig itself: high-efficiency power supplies in place of factory units, three-phase power or 240-volt single-phase power to remove step-down transformers and their losses, and computational fluid dynamics (CFD) modelling of airflow in air-cooled sites. He noted that the S21 and S19 shut down when ambient temperatures go over 100 degrees, and that a high-efficiency power supply also helps keep rigs running on hot days.
His example used the S21. A high-efficiency power supply saves about 100 watts, or $44 a year in profit, roughly 3% of power. Replacing the power supply, eliminating step-down transformers and choosing high-chip-count rigs together came to about $200 more profit per miner per year, or $55,000 a year for one megawatt.
Buying immersion on science
Quan called immersion the latest shiny toy and said many immersion products had been designed by intuition rather than science. He showed CFD images of two tanks. In the first, fluid flow was uneven and little heat came out of the miners; in a $250,000 container, he said, uneven flow can cause miners to overheat. In the second, heat rose evenly from the inlet to the outlet in the centre of the tank.
Unfortunately, many of the immersion products have been designed by intuition rather than science.
Henry Quan · 9:30
He told buyers to ask vendors for their CFD models and to measure the temperature difference between inlet and outlet fluid. By his figures, an efficient system gives two to four times better heat extraction, at a delta T of 15 to 20 degrees, against 5 to 10 degrees for an average tank. He also pointed to miner orientation: fluid should flow from the cold side of the machine to the hot side, while many operators install miners backwards so they can see the status lights.
The $260,000 payback
Quan said the right immersion system and setup can add 20% to a site's results, about $800 more profit per miner each year, or $200,000 per megawatt of immersed miners. Added to the power savings, he put the total at over $260,000, or 30% more profit, and said that is enough to pay for the extra hardware.
$260,000 of extra profits pays for additional hardware costs and drops your ROI to 1.6 years.
Henry Quan · 12:47
The talk closed as a product presentation. Quan introduced Foundry products built around the same idea: an immersion orientation kit that he said gives 5% better efficiency through better cooling, with handles for slippery miners and relocated control boards so their lights stay visible; a 4,000-watt power supply he described as the industry's only 80-plus platinum-rated unit, with a 3% power reduction; and a control board for the S19 and S21 with Foundry-designed firmware and no dev fee. He ended by asking operators to look for the weakest link in their own setup and invited them to the Foundry booth.
When this session was recorded, Bitcoin traded at $61,272; on October 17, 2026 it was $81,444.