Keynote by Wes Cummins, CEO and Co-Founder of Applied Digital | RAISE Summit 2024 | Paris

Wes Cummings, founder and CEO of Applied Digital, presented on the intersection of AI growth and power scarcity, emphasizing the company's pivot to purpose-built infrastructure located at points of power generation.

• Applied Digital originally built 500 megawatts of capacity over 24 months for blockchain/Bitcoin mining (ASICs) starting three years ago.
• In 2022, the company designed and began constructing HPC-style, high-power-density liquid-cooled data centers.
• The company launched a commercial cloud GPU service in May of the previous year, citing Character AI and Together AI as early enterprise customers.
• Current projects aim to deliver 400+ megawatts of power capacity while co-locating with renewable energy sources to minimize carbon footprints and PUE.

Power scarcity has intensified from a significant challenge to a critical bottleneck due to the surge in AI demand, specifically driven by NVIDIA H100 GPUs.

• The "hunt for power" is estimated to be 10 to 20 times larger today than during the blockchain mining boom.
• One utility provider example illustrates the scale: a 5-gigawatt grid recently received 15 gigawatts of new data center power requests within nine months.
• The US data center market has cumulatively built 23 gigawatts over the last 30 years, contrasting sharply with current forecasts requiring massive new capacity.

Forecasted demand for data center power through 2030 presents a supply-side challenge that exceeds historical build-out rates.

• Independent forecasts suggest a need for 37 gigawatts of DC capacity by 2030; some hyperscaler CEOs project 100 gigawatts may be required solely to serve the three largest hyperscalers.
• This anticipated volume would represent approximately four times the total US data center capacity built in the prior three decades.

Strategic infrastructure placement involves moving facilities away from population centers (traditional low-latency zones) to "points of power generation" where grid congestion often limits transport.

• Facilities are co-located with renewable sources, such as wind farms in North Dakota (the 6th largest wind producer in North America but 48th in population).
• Applied Digital utilizes curtailed renewable energy that would otherwise be wasted due to grid overload or negative pricing.
• AI workloads (training and inference) can tolerate higher latency than real-time applications like video streaming, enabling this shift in geographic strategy.

Operational capabilities focus on high power density and liquid cooling to support intensive AI workloads.

• The company currently deploys thousands of H100 GPUs across multiple sites.
• Sites are optimized for both training and inference applications while maintaining ultra-low Power Usage Effectiveness (PUE).
• The company anticipates that power capacity constraints will become a visibly dominant issue in 2024 and beyond.