Diamonds for Jewelry is so last century. Now They Cool Your GPUs.

2,400. That is diamond's thermal conductivity in watts per meter-kelvin. Copper is 400. Silicon is 150. GaN is 130.

The best heat conductor on the planet spent most of recorded history on engagement rings, drill bits and one very famous laser. Kara wants to bolt it to a graphics card.

I caught a whiff of these people the week their YC launch went up, and it smelled like a desert motel with the air conditioning broken and a man in a bucket hat selling you something you cannot refuse. A fifth-generation diamond dealer. A Berkeley chemical engineer who grew diamond quantum sensors for the Department of War. Two of them. San Francisco. Summer 2026 batch.

The Fifth-Generation Diamantaire

Aarin Jhaveri is a fifth-generation diamond manufacturer. The family trade predates the transistor, the hard drive, and the phrase "data center." Jhaveri sold millions of dollars of natural and lab-grown stones across the supply chain, the kind of volume that teaches you exactly who touches a diamond before it touches a customer. Then went to UC Berkeley for economics and data science. Then decided diamonds should stop being jewelry.

Anuveer Chadha is the other half. Chemical engineer, Berkeley Chemical Engineering dropout, built diamond quantum sensing and computing systems at Lawrence Berkeley National Laboratory working with the Department of Defense and the Department of Energy, and did machine learning for chemical formulations at Regeneron. Somebody who knows what one defect in a crystal lattice costs you at the quantum level.

Aarin Jhaveri, Founder / CEO. Fifth-generation diamond manufacturer. UC Berkeley, economics and data science. @AarinJhaveri | LinkedIn

Anuveer Chadha, Founder. Diamond quantum sensing and computing systems at Lawrence Berkeley National Laboratory with the Department of Defense and Department of Energy. ML for chemical formulations at Regeneron. LinkedIn

Two founders. Their YC partner is Brad Flora. Their line, from the launch post, is a keeper: "Diamond spent centuries as a symbol of extraordinary. We are making it something the world can build with."

The Squeeze: AI Cannot Shed Its Own Heat

Here is the problem nobody wants to talk about between panels at the AI conference. The industry spent a decade making chips faster and is now losing the fight against their heat. Frontier GPUs throttle. You do not buy compute. You buy heat you cannot get rid of.

Diamond moves heat up to 5x better than copper while weighing roughly 60 percent less for the same volume. It does not conduct electricity. It does not flinch in extreme environments. Bandgap 5.47 electron volts against silicon's 1.12. Breakdown field 10 megavolts per centimeter against silicon's 0.3. On the Baliga figure of merit, normalized to silicon at 1, diamond scores around 24,660.

Kara grows the stuff in their own reactors: single-crystal wafers in research, optical and quantum grades, heat spreaders for AI and high-performance computing, diamond-metal composites in copper and aluminium, metallized and bondable plates, isotopically engineered layers, thin diamond membranes, custom microfabrication. They say custom samples are going out now under pilot agreements and purchase orders into thermal, electronic, aerospace and quantum programs.

Now read their open ask, because it tells you who they think is bleeding. Introductions to AI accelerator packaging teams. Advanced packaging and OSAT engineers. Data-center infrastructure companies. High-power RF and GaN teams. Aerospace and defense electronics.

That is the guest list for the machines running the AI boom, and Kara is offering them free diamond samples to test. A two-person company walking up to the hottest silicon on earth with a rock and a business card.

Two People Doing What Used to Take a Division

A materials company with a headcount of two is the real story. Diamond wafer production spent decades in the hands of national labs and a handful of industrial giants, because growing single-crystal diamond with specified purity, orientation, isotope fraction and a sub-nanometer surface finish is a very long, very expensive apprenticeship.

That a two-person team attacks it at all is the fingerprint of this era: cheap simulation, cheap models, and software eating the process engineering that used to require a building full of staff. The tools compress execution. They do not hand you a supply chain your family built over five generations, and they do not hand you a national lab's quantum program. Kara has both, and says so plainly. Diamond is the unfair advantage, and diamond is who they are.

"In 2014 I was Salieri and Sam was the Mozart. He saw AGI coming and went for it. I didn't understand. Instead of saying no, say yes and."

Garry Tan, via TheFoundersPack

Salieri watched the future walk past him and kept composing for the old court. Kara's version of the lesson: do not defend the old material. There is a whole periodic table of things that were good enough until somebody industrialized the better one.

The Pack Angle

Kara is giving away samples. That is not charity. That is pack-building. Get your material designed into somebody else's roadmap while they are still sweating, and you become the default before anybody has written a standard.

Silicon did not win because silicon was the best possible answer. Silicon won because an entire pack of fabs, toolmakers, packaging houses, equipment vendors and standards bodies agreed to build around one material for fifty years. Same argument as Form Your Own Pack: one company cannot build an industry. A pack can. Diamond has no such pack yet. The first company to assemble one owns the next fifty years of it.

The next time your GPU throttles mid-render, remember that a chunk of what you are waiting for was sold as a rock in a shopping mall. Kara is betting the rock stops being a rock. Two people, one reactor, five generations of the trade behind them. Say yes and.

Links: YC Profile | karalabs.ai | @KaraLabsAI | Company LinkedIn | Launch video | YC Launch post | Partner: Brad Flora