No PriorsNo Priors Ep. 93 | With Akash Systems' Felix Ejeckam and Ty Mitchell
CHAPTERS
- 0:05 – 2:29
Diamond-based cooling: Akash Systems’ core idea and why it matters
Sarah introduces Akash Systems and the founders’ thesis: using lab-grown diamond as an ultra-high thermal conductor to cool computing platforms. Felix frames Akash as a materials-first company that starts at the “periodic table level” and builds up to chips, boards, and full systems for space and AI data centers.
- •Akash builds diamond-based thermal solutions to make computers run cooler, faster, and more energy-efficient
- •Materials-science-first approach rather than inserting into the middle of an existing supply chain
- •Initial focus on space hardware, expanding to AI data centers
- •Value proposition: unlock performance and reduce energy use by improving heat extraction at the source
- 2:29 – 4:49
Felix’s path: combining dissimilar materials to create step-change systems
Felix explains his background in electrical engineering and materials science and how his PhD work shaped Akash’s philosophy. The central motif is engineering “1+1=10” outcomes by coupling materials (diamond with silicon/GaN) to unlock new performance regimes.
- •PhD at Cornell focused on integrating dissimilar materials for outsized system gains
- •Diamond’s thermal conductivity paired with mainstream semiconductors enables new capabilities
- •Proved the concept in space by building and launching exceptionally fast satellite radios
- •Now applying the same materials coupling approach to AI compute platforms
- 4:49 – 6:19
Ty’s commercialization lesson: sell the system, not the material
Ty describes his silicon carbide background and what he learned bringing a materials breakthrough to market. The key go-to-market insight is that customers believe system-level proof, so Akash builds complete products to demonstrate the advantage rather than pitching raw materials alone.
- •Materials innovations often fail commercially without system-level validation
- •Customers optimize for price unless they can see end-to-end performance benefits
- •Akash’s strategy: build the module/system to prove outcomes and reduce adoption friction
- •Why AI now: space thermal constraints were harder; success there implies strong leverage for data centers
- 6:19 – 8:04
From space to AI: power density comparisons and confidence in the approach
Felix and Ty quantify why their space heritage translates to AI infrastructure. Cooling requirements in space reach far higher power densities, making data-center GPUs a comparatively easier target for their diamond-coupled thermal path.
- •Space cooling targets ~4,000–5,000 W/cm² vs. a few hundred W/cm² in typical servers
- •Space constraints: no airflow, limited area, high reliability requirements
- •Demonstrated capability in extreme environments builds confidence for terrestrial scaling
- •AI’s heat problem is growing and lacks a sufficiently “at-the-source” solution
- 8:04 – 10:23
Live conductivity demo: diamond wafer melting ice and what that implies for servers
Felix demonstrates diamond’s thermal conductivity by using a diamond wafer to rapidly melt/cut into an ice cube using body heat. He translates the demo into near-term and longer-term temperature reduction goals, highlighting the economic value for data center operators.
- •Diamond wafer conducts heat so well it quickly melts ice when touched
- •Near-term target: ~10°C reduction (already significant economically)
- •Longer-term roadmap: 20–40°C reductions; space applications already achieved 80–90°C
- •Cooling translates into performance headroom, reliability, and energy/cost savings
- 10:23 – 13:51
Where Akash fits in the cooling stack: compatible with air, liquid, and packaging tricks
Ty outlines the multi-layered cooling landscape—from facility airflow management to liquid cooling at the rack and advanced packaging at the chip level. Akash’s diamond approach is positioned as additive across these methods, addressing heat closer to the device where it originates.
- •Cooling exists at multiple levels: data center airflow, rack liquid cooling, chip/package design
- •Diamond cooling can be used standalone or layered with existing cooling methods
- •Goal: increase operating and performance margin to avoid throttling and unlock more throughput
- •Industry signals: heating issues impacting cutting-edge GPU platforms and rollout timelines
- 13:51 – 17:44
Heat as the root cause: throttling, reliability, cost, and “grid crash” risk
Felix argues that many data center pain points—reliability, performance instability, and operating cost—are fundamentally heat-driven and become “whack-a-mole” without source-level intervention. The conversation ties heat to grid-scale sustainability, warning that current trajectories are economically and energetically untenable.
- •Thermal barriers inside servers limit the impact of external cooling alone
- •Heat drives thermal throttling, erratic behavior, and early-life failures
- •Scaling AI without better physics-level cooling risks massive power demand and high electricity costs
- •Thesis: fix heat at the chip/source with materials/chemistry, not just system-level band-aids
- 17:44 – 20:49
Inside-the-server bottleneck: interface layers and thermal barriers in the “sandwich”
Ty details the physical stack-up inside a server—metals, epoxies, solders, boards—and explains how each interface adds thermal resistance. This layered “sandwich” becomes the practical limiter to higher performance, turning heat into a true blocker for future scaling.
- •Multiple material interfaces create cumulative thermal barriers (each interface adds resistance)
- •As chips get denser and hotter, conventional approaches struggle to extract heat effectively
- •Heat constraints can slow product cycles and threaten vendor execution (rollout delays, missed quarters)
- •Need for solutions that reduce internal thermal resistance close to the die/package
- 20:49 – 22:50
Go-to-market positioning: server maker that expands GPU performance envelopes
Felix clarifies that Akash is not building GPUs; they buy the best chips and integrate diamond cooling to expand the system architect’s options. He connects improved thermals to higher density and faster compute, even suggesting it could accelerate effective “Moore’s law” by easing thermal crosstalk constraints.
- •Akash’s entry point: deliver servers/systems that run top GPUs cooler and more predictably
- •Target users: inference/training operators, data centers, cloud providers at the performance edge
- •Thermal crosstalk limits transistor density and sustained performance; better cooling relaxes constraints
- •Vision: materially faster progress in achievable performance timelines via thermal unblocking
- 22:50 – 25:23
How lab-grown diamond wafers are made—and Akash’s “secret sauce” in bonding
Sarah asks how diamond growth works beyond jewelry contexts. Ty describes seed-crystal growth and plasma/CVD-style processes, while Felix emphasizes that Akash’s differentiation is not only growth but also the atomic-level coupling of diamond to semiconductors.
- •Diamond growth parallels other semiconductor substrates: seed crystal + controlled deposition/growth
- •Diamond is carbon; grown in reactors with plasma processes under extreme conditions
- •Key differentiation: intimate diamond–semiconductor coupling using physics/chemistry
- •Non-trivial materials engineering enables deployment at scale in shipped systems
- 25:23 – 28:10
Early customer spotlight: India’s NexGen sovereign cloud and why sovereignty drives demand
Felix explains why NexGen—India’s largest sovereign cloud provider—is an important early customer and proof point. The discussion links sovereign data residency requirements to both cloud and space contexts, and highlights scaling execution as a key reason the partnership matters.
- •Sovereign cloud requires data to remain within national borders; demand is globalizing
- •NexGen as a marquee reference: validates performance, scaling, and delivery capability
- •Akash sees opportunity to replicate country-by-country deployments respecting sovereignty
- •Partnership positioned as the start of larger follow-on expansion
- 28:10 – 31:17
CHIPS Act and U.S. manufacturing: building the AI stack domestically
Sarah pivots to American manufacturing capacity and Akash’s role as a CHIPS Act recipient. Ty argues the U.S. must invest more across the full value chain, while Felix frames CHIPS alignment around supply chain resiliency, domestic jobs, and national security partnerships.
- •Argument: U.S. must lead in AI-critical technologies end-to-end, not just chip design
- •CHIPS Act as a lever to support smaller innovators solving hard bottlenecks (like thermal limits)
- •Domestic manufacturing reduces strategic dependence exposed by COVID-era shortages
- •Akash points to defense collaborations (e.g., Raytheon) and U.S. job creation (CA/TX)
- 31:17 – 36:29
National security and RF/radar: why space and spectrum dominance matter
Sarah and Ty discuss why satellite communications, radar, and RF spectrum capabilities are decisive in modern conflict. Ty uses a WWII radar anecdote to illustrate how foundational sensing/communications advantages can reshape outcomes—paralleling AI’s strategic importance today.
- •Modern conflict increasingly depends on RF spectrum awareness and resilient space systems
- •Radar history shows early adoption can provide asymmetric advantage
- •Akash’s space work ties into defense requirements beyond commercial AI data centers
- •RF and AI are framed as parallel strategic technologies with outsized national impact
- 36:29 – 42:21
AI for materials discovery and manufacturing: faster research, inverse design, expanded imagination
The conversation closes on AI’s role in accelerating materials research and industrial capacity. Ty emphasizes AI as a research accelerator (finding prior art, iterating faster), while Felix broadens the vision: AI plus improved compute/cooling could expand what society can design and manufacture.
- •AI as a “research assistant” enabling rapid literature retrieval and hypothesis iteration
- •Excitement for inverse design, exploring chemical space, and accelerating simulation workflows (e.g., DFT)
- •Core limiter becomes question-framing and imagination, not compute availability
- •Vision: AI-enhanced manufacturing productivity and breakthroughs in biotech, batteries, and beyond
