Superconducting chip claim: don't change the process node, change the temperature zone — 28nm logic can hit 5nm performance
Move logic into the liquid helium temperature zone and replace transistors with superconducting switches. That's Snowcap's bet: tape out RISC-V first, then claim 28nm-class logic can deliver 5nm-class performance.
The argument · tap a timestamp to hear it
The ceiling on extreme overclocking is heat removal, not the chip
This episode opens up from the overclocking industry: everyday chips rely on air cooling, higher-power parts use water cooling, and beyond that you need sub-zero media like dry ice and liquid nitrogen. Extreme overclockers add voltage to chase the highest frequencies, the chip gets hotter and hotter, and the only way to deal with it is ever more extreme cooling. The host says he was an extreme overclocker himself and once held the world number two spot — though it only lasted a day and a half. The key reference point he gives: this kind of overclocking exists just to run a benchmark for a few minutes at 8.5GHz, whereas real superconducting computing isn't chasing an instantaneous record but logic at close to zero power.
With liquid helium overclocking, the expensive part isn't the helium, it's the transfer line
Going from liquid nitrogen to liquid helium, physics starts to interfere. Liquid helium costs about 20 times as much as liquid nitrogen (liquid nitrogen is about one dollar a litre), but that isn't the main expense: getting liquid helium from a large storage tank into a 3-4 kilogram copper overclocking pot through the transfer line costs 5000 dollars for a single use. Both Intel and Asus have done liquid helium overclocking, and records have broken past 9GHz. The host warns that when temperatures drop this far, material behaviour changes drastically — some platforms have a ‘cold bug’ where the system simply won't boot below a certain temperature; only platforms without a cold bug are suitable for liquid helium play.
The lure of superconductivity isn't faster, it's zero resistance
Overclocking is about squeezing frequencies out of existing materials; superconductivity is a different story: electrons flowing through a superconducting material meet no resistance, which amounts to free transport. The vast majority of superconductors today require extremely low temperatures, and room-temperature superconductivity is still stuck in a state of research back-and-forth. The key inference here: if an entire chip's switching logic could operate in the superconducting state, energy per operation would be ‘several orders of magnitude’ better than today's laptops. The host deliberately states it fully and then pulls back — superconductivity is still the holy grail, not yet in hand. This sets up Snowcap's entrance later.
What the former Intel CEO bet on is a temperature zone, not a process node
Snowcap Compute has just closed a seed round, with former Intel CEO Pat Gelsinger on its board. Their bet is that they can build superconducting chips that run at roughly 4-15K, the liquid helium temperature zone. Structurally it's no longer billions of transistors but a new kind of switch that works at superconducting temperatures — the Josephson junction. At this stage Snowcap's goal is to first build a RISC-V-class CPU at low temperature, getting power consumption close to free; the price is that the whole machine must stay in the liquid helium temperature zone, and the cooling itself consumes energy.
The new variable in superconducting logic is fitting into a standard EDA flow
There is no shortage of long-term researchers: a US defence contractor built chips integrating tens of millions of Josephson junctions, IMEC reached hundreds of millions, and built accompanying logic and EDA tools. Snowcap's differentiator is compatibility: they say this kind of superconducting logic can go into a standard EDA flow and can be taped out on ordinary CMOS processes, with niobium titanium nitride as the material. The core conclusion the host takes away: the energy-efficiency advantage superconductivity brings is equivalent to achieving near-5nm chip performance with 28nm logic — note this is not a smaller process node but the equivalent gain from lower power consumption.
Cryo CMOS is only an intermediate state; the real hurdle is scale
Snowcap plans to deliver its first chip in 2026, followed by a deployable RISC-V design. For the intermediate state you can look at cryo CMOS: it runs in the roughly 72-77K liquid nitrogen temperature zone, and although there's no superconductivity, the low temperature itself pushes power consumption down. The real hurdle to cross is going from cryo CMOS into superconductivity; the host's doubt about this is ‘can you scale it up at high performance’ — being able to run and being able to commercialise at scale are two different things.
Cold booting is nothing new; keeping the whole machine in liquid helium is the hard part
This section provides a reality check on low-temperature tape-outs: when AMD was most short of money, the first Zen 1 A0 sample it taped out could only boot at dry-ice sub-zero temperatures because of a design flaw, and was later fixed to run at room temperature. So low temperatures in the test and boot stages are nothing unusual; but keeping an entire processor in the liquid helium temperature zone is an infrastructure problem, including helium supply — liquid helium is also the lifeblood of the superconducting magnets in MRI machines, and there was a helium shortage a few years ago.
Snowcap isn't selling IP, it's selling a closed all-in-one machine
Snowcap positions itself not as selling IP but as delivering a complete system. The host's judgement is that precisely because of the infrastructure barrier of the liquid helium temperature zone, the product form will be a near-closed all-in-one machine, with the software stack rebuilt from scratch. A seed-stage company still has A/B/C rounds ahead of it; however, they have already brought in advisors and former employees with Apple, Nvidia and Google backgrounds. This topic is worth being sceptical about, and worth following.
In their own words · checked verbatim
The cost of liquid nitrogen is about, you know, a dollar a litre. And liquid helium is about 20 times that. That's not the main cost with liquid helium. The main cost is actually the tube you need to get it from this massive compressed gas sewer into your massive 3-4 kilogram copper pot. That tube, single use, is five grand.
When you go down to those cold temperatures, the way the materials work is incredibly different. In some chips and in some platforms, you might experience what's called a cold bug. It gets too cold for the system to boot or to work.
There has been a strong effort to make room temperature semiconductors. We're not there yet. Research, if you follow that area, sometimes stalls and sometimes there's a breakthrough. If that happens, one day we may have free energy. Right now, we don't. Right now, superconducting is the holy grail.
But what if you could make a chip that was superconducting? Suddenly, you now have a processor that can do compute for next to no energy. And we're talking orders of magnitude better energy consumption per operation than this laptop, for example.
defense contractors that have designed chips with tens of millions of these Josephson junctions. And then other companies like IMEC doing hundreds of millions of Josephson junctions.
And the headline I got here is that you can essentially build chips with such superior efficiency that you're essentially building five nanometer chips on what is equivalent of 28 nanometer logic.
out comes their first A0 piece of silicon, their first Zen 1 chip. And it would only boot at subzero temperatures using dry ice because of some of the issues that they had in the design of the chip. But they fixed that and it works at room temperature.
Figures
| Liquid nitrogen price | About 1 dollar per litre | 2:01 |
| Liquid helium price relative to liquid nitrogen | About 20 times | 2:01 |
| Liquid helium overclocking record | Past 9 GHz | 3:03 |
| Snowcap operating temperature target | About 4-15 Kelvin (liquid helium temperature zone) | 5:07 |
| Scale of Josephson junctions integrated by the US defence contractor | Tens of millions | 7:08 |
| Scale of Josephson junctions integrated by IMEC | Hundreds of millions | 7:08 |
| Snowcap first chip delivery date | 2026 (planned) | 8:12 |
| Cryo CMOS operating temperature zone | About 72-77 Kelvin (liquid nitrogen temperature zone) | 8:12 |
| Boot conditions for AMD's first Zen 1 A0 chip | Could only boot at sub-zero temperatures using dry ice | 9:12 |
Glossary
- cold bug
- The phenomenon where a chip cannot boot or cannot work properly after being cooled too far.
- Josephson junction
- A new type of switching device made of superconducting materials that works at extremely low temperatures.
- cryo CMOS
- Low-power CMOS that runs in the liquid nitrogen temperature zone (about 77K), not yet at the superconducting level.
How to listen
Hard-tech investors trying to verify whether superconducting computing is a route or a story, plus CPU and supercomputer architects studying the chip power wall.
Anyone already familiar with overclocking and liquid nitrogen cooling can skip the 0:00-2:01 overclocking setup.