No PriorsNo Priors Ep. 93 | With Akash Systems' Felix Ejeckam and Ty Mitchell
Sarah Guo and Felix Ejeckam on diamond-cooled AI servers promise cooler chips, faster compute, safer grids.
In this episode of No Priors, featuring Sarah Guo and Felix Ejeckam, No Priors Ep. 93 | With Akash Systems' Felix Ejeckam and Ty Mitchell explores diamond-cooled AI servers promise cooler chips, faster compute, safer grids Akash Systems’ founders Felix Ejeckam and Ty Mitchell explain how lab-grown diamond, the most thermally conductive material, is being integrated directly with semiconductors to cool everything from space radios to AI GPUs. They argue that current data center cooling—fans, liquids, rack-level tricks—only treats symptoms and that physics- and chemistry-driven materials solutions are essential to avoid power crises and performance limits. By bonding synthetic diamond to chips, Akash claims significant temperature drops, better reliability, and headroom to ‘hyper-accelerate Moore’s law’ for AI workloads. The conversation also covers sovereign cloud deployments, U.S. manufacturing and CHIPS Act support, and how AI itself will accelerate materials research and advanced manufacturing.
At a glance
WHAT IT’S REALLY ABOUT
Diamond-cooled AI servers promise cooler chips, faster compute, safer grids
- Akash Systems’ founders Felix Ejeckam and Ty Mitchell explain how lab-grown diamond, the most thermally conductive material, is being integrated directly with semiconductors to cool everything from space radios to AI GPUs. They argue that current data center cooling—fans, liquids, rack-level tricks—only treats symptoms and that physics- and chemistry-driven materials solutions are essential to avoid power crises and performance limits. By bonding synthetic diamond to chips, Akash claims significant temperature drops, better reliability, and headroom to ‘hyper-accelerate Moore’s law’ for AI workloads. The conversation also covers sovereign cloud deployments, U.S. manufacturing and CHIPS Act support, and how AI itself will accelerate materials research and advanced manufacturing.
IDEAS WORTH REMEMBERING
7 ideasAttack heat at the materials level, not just with more fans and fluid.
Akash’s thesis is that air and liquid cooling mostly operate far away from the heat source; directly integrating ultra-conductive diamond with chips shortens the thermal path, meaning lower temperatures, less throttling, and higher sustained performance.
Diamond cooling stacks with existing data-center cooling, it doesn’t replace it.
Their servers ship with both NVIDIA’s liquid cooling and diamond cooling, showing that materials-level advances can be layered on top of rack- and facility-level solutions to gain extra thermal and performance margin.
Thermal constraints are becoming a primary blocker for AI hardware roadmaps.
Mitchell points to delays and heat issues in recent high-end GPUs as an early sign that keeping performance doubling every few years will be limited less by transistor counts than by the ability to remove heat from increasingly dense chips.
Diamond-enabled cooling could effectively “hyper-accelerate” Moore’s law for AI.
By reducing thermal crosstalk, the company believes transistor density and GPU operating points can rise faster, potentially shrinking timelines for compute-heavy tasks (like generating feature-length films) from days to seconds.
Sovereign cloud and localized AI infrastructure are major growth vectors.
The NexGen India deal illustrates how countries want AI capacity and data to remain within their borders, creating many regional opportunities for differentiated, high-efficiency infrastructure vendors.
Rebuilding U.S. manufacturing across the AI stack is a strategic imperative.
They argue that relying on overseas fabs and assembly is risky for both economic resilience and national security, and that CHIPS Act support should foster an end-to-end domestic ecosystem—from materials and chips to racks and data centers.
AI will dramatically accelerate materials research and manufacturing innovation.
Both founders expect AI models to mine decades of literature, explore chemical space, and run vast design iterations (e.g., DFT, inverse design), enabling faster breakthroughs in materials like diamond and boosting automated, AI-powered manufacturing capacity.
WORDS WORTH SAVING
5 quotesAt Akash, as material scientists, we come in at the periodic table level.
— Felix Ejeckam
If a physics or chemistry approach to solving the heat problem is not used today, the needs of AI data centers will crash the grid as we know it.
— Felix Ejeckam
Everything you mentioned is heat… unless you go to the source of the problem, you’re really just playing whack-a-mole.
— Felix Ejeckam
We think that with our diamond technology, we will be able to hyper‑accelerate Moore’s law.
— Felix Ejeckam
We are limited by our ability to frame the questions… only by our own imagination really.
— Ty Mitchell
QUESTIONS ANSWERED IN THIS EPISODE
5 questionsWhat technical and manufacturing hurdles remain to scaling diamond–semiconductor integration for all major AI chips, not just select server SKUs?
Akash Systems’ founders Felix Ejeckam and Ty Mitchell explain how lab-grown diamond, the most thermally conductive material, is being integrated directly with semiconductors to cool everything from space radios to AI GPUs. They argue that current data center cooling—fans, liquids, rack-level tricks—only treats symptoms and that physics- and chemistry-driven materials solutions are essential to avoid power crises and performance limits. By bonding synthetic diamond to chips, Akash claims significant temperature drops, better reliability, and headroom to ‘hyper-accelerate Moore’s law’ for AI workloads. The conversation also covers sovereign cloud deployments, U.S. manufacturing and CHIPS Act support, and how AI itself will accelerate materials research and advanced manufacturing.
How do the economics of diamond cooling (cost per watt saved or performance gained) compare to aggressive liquid cooling or facility-level retrofits?
What kinds of new AI workloads or model architectures become feasible if GPUs can reliably run much hotter and denser without throttling?
How should policymakers balance short-term cost pressures with long-term national security when deciding how much AI manufacturing to localize in the U.S.?
In practice, how can AI tools be embedded into materials R&D workflows to meaningfully shorten the path from novel idea to deployable device?
EVERY SPOKEN WORD
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome