DDN CEO Alex Bouzari: Nvidia uses us internally, xAI's data center was built in 4 months, and China builds them 3-5x cheaper

Summary

Alex Bouzari, co-founder and CEO of DDN, makes the case that the bottleneck in AI infrastructure is not GPUs or energy in isolation — it is the cost and speed of putting all of it together productively. DDN sits at that intersection, supplying the data layer that sits alongside compute in large-scale AI deployments. Nvidia uses DDN internally, and xAI's Colossus cluster — 200,000 GPUs — runs on DDN infrastructure.

The xAI data center benchmark

The most concrete claim Bouzari makes is about construction speed. xAI built its Memphis data center in four to four and a half months. Bouzari says DDN was involved throughout and that his initial reaction was that it was "completely mad" — his firm had completed more than 100 large data center deployments and had never seen one finished in under three years. The method was deliberate: Musk bypassed domain experts who carried the mental block that it couldn't be done quickly, and instead hired generalist problem-solvers. The team slept on-site over Christmas and New Year's to hit the timeline. Bouzari says xAI then applied lessons from that first build to three more, and eventually to 32.

China's cost advantage

The sharper competitive concern Bouzari raises is China. U.S. data center construction costs run $10–$15 per kilowatt. China builds them for one-third to one-fifth of that — roughly $2–$5 per kilowatt. He attributes this to a modular, iterative approach similar to what xAI used: treat each build as a learning cycle, optimize, replicate. Bouzari says a Middle Eastern sovereign customer is already reviewing Chinese design architectures with DDN to replicate that cost structure in the Gulf. His read is that the gap is not incremental — a three-to-five times cost differential is structurally significant and the U.S. needs to close it.

Where enterprise AI spending is actually landing

Bouzari identifies financial services and life sciences as the two clearest areas of real enterprise AI adoption, distinct from the hyperscaler and frontier lab conversation. In financial services — hedge funds, high-frequency traders — the ROI is direct and the technical appetite is already there. In life sciences, the value proposition is compressing drug development timelines and improving FDA approval odds. He points to digital twin simulation with synthetic data as an emerging tool: run eight candidate drug pathways in simulation, identify the optimal combination before committing to trials. Autonomous vehicle manufacturers, factory automation, and sovereign AI programs round out his deployment picture.

The token cost problem

Bouzari flags a structural issue for agentic AI companies that doesn't get enough attention: many are "upside down" — what they charge customers is less than what it costs them to run the underlying models. Token consumption is too high relative to the price they can sustain in the market. His argument is that the real software play right now is compressing the compute required to complete a given task, not just adding more GPUs. SSD costs have tripled in recent months, further squeezing margins. DDN's position is that its data plane can deliver equivalent or better performance with less hardware spend, which is increasingly the pitch to enterprises trying to make AI ROI work.

The incumbent risk

On the SaaS disruption question, Bouzari draws a clear line between forward-moving and static incumbents. IBM and Intel are his examples of companies that had the resources to lead but lacked execution velocity. Accenture, with 750,000 employees, faces the same test — how many of those roles remain relevant if it fails to transform fast enough. His view is that leadership posture matters more than current revenue health: companies that treat today's top line as a reason to move slowly will be displaced.