AI and the Future of Work: Artificial Intelligence in the Workplace, Business, Ethics, HR, and IT for AI Enthusiasts, Leaders and Academics

364: Inside the AI Infrastructure Race: TensorWave CEO Darrick Horton on Power, GPUs and AMD vs NVIDIA.

Dec 1, 2025

Guest

Darrick Horton

Darrick Horton, CEO and co-founder of TensorWave, shares insights from his journey in AI infrastructure. He discusses the strategic decision to choose AMD over Nvidia, highlighting how this choice created a competitive edge in a GPU-constrained market. Horton explains why training clusters offer greater versatility than inference clusters and the advantages of NeoClouds like TensorWave in innovation speed. He warns about potential power limits impacting AI scaling and reflects on the importance of optimizing for green energy and the future of hardware architecture.

Ask episode

AI Snips

Chapters

Transcript

Episode notes

INSIGHT

Power Is The Real Bottleneck

Power, not GPUs, is the primary constraint for scaling AI infrastructure today.
Darrick Horton predicts we'll run out of power before we run out of AI demand by 2027–2029.

ANECDOTE

Skunk Works Sparked An Infrastructure Career

Darrick Horton left university early and started at Lockheed Martin's Skunk Works working on nuclear fusion.
That experience sparked his long-term interest in infrastructure, energy, and computing.

ADVICE

Bet On Practical Alternatives

Evaluate alternatives to dominant vendors when customers value cost, access, and simplicity more than brand.
Choose platforms that offer production capacity and a strong software investment to avoid supply constraints.

Get the Snipd Podcast app to discover more snips from this episode

Get the app

Darrick Horton is the CEO and co-founder of TensorWave, the company making waves in AI infrastructure by building high-performance compute on AMD chips. In 2023, he and his team took the unconventional path of bypassing Nvidia, a bold bet that has since paid off with nearly $150 million raised from Magnetar, AMD Ventures, Prosperity7, and others. TensorWave is now operating a dedicated training cluster of around 8,000 AMD Instinct MI325X GPUs and has already hit a $100 million revenue run rate.

Darrick is a serial entrepreneur with a track record of building infrastructure companies. Before TensorWave, he co-founded VMAccel, sold Lets Rolo to LifeKey, and co-founded the crypto mining company VaultMiner.

He began his career as a mechanical engineer and plasma physicist at Lockheed Martin’s Skunk Works, where he worked on nuclear fusion energy. While he studied physics and mechanical engineering at Andrews University, he left early to pursue entrepreneurship and hasn’t looked back since.

In this conversation we discussed:

Why Darrick chose AMD over Nvidia to build TensorWave’s AI infrastructure, and how that decision created a competitive advantage in a GPU-constrained market
What makes training clusters more versatile than inference clusters, and why TensorWave focused on the former to meet broader customer needs
How Neocloud providers like TensorWave can move faster and innovate more effectively than legacy hyperscalers in deploying next-generation AI infrastructure
Why power, not GPUs, is becoming the biggest constraint in scaling AI workloads, and how data center architecture must evolve to address it
Why Darrick predicts AI architectures will continue to evolve beyond transformers, creating constant shifts in compute demand
How massive increases in model complexity are accelerating the need for green energy, tighter feedback loops, and seamless integration of compute into AI workflows

Resources: