Akash Systems puts a diamond under the chip.

In a 40,000 sq ft cleanroom being readied in West Oakland, a startup is preparing to manufacture something that didn't exist a generation ago: lab-grown diamond, bonded under gallium nitride, packaged into chips that fly above Earth and run inside AI datacenters.

Walk past the chain-link fence on a quiet block of West Oakland and you wouldn't guess. The building looks like any other low-slung industrial shell - until you notice the federal hard-hats and the trade-union signage. Inside, Akash Systems is building one of the only new semiconductor facilities the Bay Area has seen in years. It will manufacture diamond. Not the kind you put on a finger. The kind you put under a transistor.

This is not, despite the headlines, a jewelry company. Akash makes semiconductors. The diamond is the substrate. It is there because every chip ever made has the same enemy, and that enemy is heat.

The problem they saw

Modern chips are limited less by how fast they can switch and more by how quickly the heat they make can leave. Cram more transistors into a millimeter and the math gets worse, not better. Satellite radios fry. AI accelerators throttle. Datacenters install absurd quantities of water-cooling plumbing to chase a problem the chip itself created. The whole industry is, in a sense, fighting its own thermodynamics.

For a long time the answer was: build a bigger heat sink. Spread the chip out. Slow it down. Add fans, then add liquid, then add immersion tanks. Each fix bought a year or two. None of them addressed the source.

Felix Ejeckam, who would later co-found Akash, had a different instinct. If heat is the problem, put the chip on top of the best thermal conductor on Earth. Diamond's thermal conductivity is roughly five times that of copper. It is, as materials go, ridiculous. The catch was that nobody had figured out how to do it manufacturably.

The founders' bet

Ejeckam invented GaN-on-Diamond in 2003, while at a small materials outfit called Group4 Labs. The process - in plain terms - lifts a layer of gallium nitride off its growth substrate and re-bonds it onto synthetic, chemical-vapor-deposited diamond. The hot transistor ends up tens of nanometers from a perfect heat sink. The bet was that this would be worth the effort.

It took thirteen years. In 2016, Ejeckam and Ty Mitchell founded Akash Systems in San Francisco. In 2017 Khosla Ventures led a $3.1M seed alongside Social Capital, Data Collective and Backstage Capital. The pitch was unfashionable for the moment - this was the era of food delivery and group chat apps - but it was specific: we will sell GaN-on-Diamond power amplifiers to satellite operators first, because they pay the most for thermal performance.

The product

Akash now ships a small family of products that all start from the same wafer.

First, the RF power amplifiers - MMICs built on diamond substrates, sold to satellite and ground-station integrators. Second, integrated CubeSat radios, which is where Akash first broke through. In 2019 the FCC granted Akash experimental approval to fly a 12U CubeSat carrying a GaN-on-Diamond transmitter. The system demonstrated downlinks above 5 Gbps - an order of magnitude beyond what comparable CubeSats were doing at the time. Third, and more recently, Diamond Cooling AI servers, where the company is applying the same logic to GPU-class chips that have begun to throttle on heat at densities every datacenter operator wants.

The proof

A deep-tech company can spend a decade on the unloved side of the hype cycle. What changes the conversation is customers, dollars, and government willing to write a check. Akash now has all three.

The customer side: in 2024, Akash signed a $27 million contract with NxtGen, a Bengaluru-based datacenter operator, to supply diamond-cooled AI servers. NxtGen runs AI compute, disaster recovery and managed security for a long list of Indian enterprises - not the most glamorous announcement in datacenter media, but a deeply unsentimental one. People who run servers for a living are not buying diamond cooling for the romance of it.

The government side: in November 2024, the U.S. Department of Commerce, under the CHIPS and Science Act, signed a non-binding preliminary memorandum of terms with Akash for up to $68 million in support - $18.2M in direct federal funding plus up to $50M in combined federal and state tax credits. The money is earmarked for the 40,000 sq ft cleanroom in West Oakland, and the total project envelope, including private capital, is roughly $121 million. The estimate is more than 400 manufacturing and construction jobs. There is a project labor agreement with the Alameda Building Trades and a workforce-development pipeline with Berkeley City College, Laney College and Northeastern University Oakland.

The mission

If you ask Akash people what the company is for, they will give you a thermal-management answer and a community answer, and the two are linked in a way that is unusual for a chip company. The technical mission is to remove heat as the limiting variable in high-power electronics. The civic one is to do that manufacturing in a part of the Bay Area that has not had a serious industrial employer in a long time, with union-option hiring and a local training pipeline.

It is fair to be skeptical. Plenty of deep-tech companies talk a beautiful game and then quietly move the fab to a tax-friendlier state. Akash made the call public, attached labor terms to it, and tied the federal money to those terms. That is, at minimum, a real constraint.

Why it matters tomorrow

The two markets Akash points at - satellite communications and AI infrastructure - are both running into the same wall, and both at the same time. Constellations want higher throughput per kilogram of payload. Datacenters want higher compute per square foot of rack. Both are gated by how fast heat can leave silicon.

Diamond is not the only answer. There is silicon carbide. There is direct-to-chip liquid. There are immersion tanks the size of bathtubs. But all of them, at some point, run into the same nanometer-scale problem that Ejeckam started looking at twenty years ago: the heat is generated at a tiny spot, and the closer the heat sink can get to that spot, the better everything works. Diamond gets closest.

Back to that chain-link fence in West Oakland. A few years from now it won't be a fence. It will be a cleanroom shipping wafers to satellite primes and datacenter operators. The chip on your phone won't sit on diamond. The radio above your phone, in low Earth orbit, might. The GPU training the model that answers your question almost certainly will. The substrate will be invisible. That is roughly how good infrastructure works.