US AI Infrastructure Gap: Why Speed and Power Decide the Next Edge
Nvidia CEO Jensen Huang is blunt: the US risks losing its AI edge if it can't build fast enough or secure enough power. In a conversation with John Hamre, President of the Centre for Strategic and International Studies, he pointed to a widening gap with China on two fronts that matter most: construction velocity and energy capacity.
For executives, this isn't a tech story-it's a build-and-power story. Strategy now lives or dies on timelines and megawatts.
Construction Speed Is the First Constraint
Huang contrasted US timelines with China's ability to mobilize quickly. "If you want to build a data centre here in the United States from breaking ground to standing up an AI supercomputer is probably about three years. [China] can build a hospital in a weekend."
That lag changes the calculus for any AI roadmap. If you can't stand up compute within the market window, you hand first-mover advantage to someone who can. Plans must assume multi-year lead times-often longer than product cycles and demand spikes.
What to Do Now (Build-Speed Levers)
- Prioritize brownfield over greenfield to compress entitlement and utility timelines.
- Standardize designs (liquid cooling-ready, AI-tuned power densities) to cut redesign loops.
- Pre-negotiate interconnects and permits in parallel with site selection.
- Use modular phases (e.g., 20-40 MW blocks) to ship value earlier while scaling.
- Keep a balanced portfolio: lease near-term capacity, build for scale where unit economics win.
Energy Capacity Is the Hard Wall
Huang's second point: national power. "China has twice as much energy as we have as a nation and our economy is larger than theirs." While China adds capacity, US growth is flatter. For multi-gigawatt AI campuses, firm, predictable power is now as strategic as access to advanced accelerators.
If your power isn't contracted, your AI plan is hypothetical. Interconnection queues and substation upgrades can erase quarters-sometimes years.
Power Moves That De-Risk Execution
- Lock long-term PPAs with firming; evaluate on-site or adjacent generation (nuclear-adjacent, hydro, gas with CCS, or renewables plus storage).
- Pick sites where utility headroom and transmission upgrades are already funded.
- Model cooling and density early; liquid cooling changes power and water math.
- Structure location strategy around power first, latency second-especially for training clusters.
Tech Lead Isn't Enough Without Deployment
Huang stressed that the US still leads in AI chips and systems-"generations ahead" by his framing. But he also cautioned: "Anybody who thinks China can't manufacture is missing a big idea."
Translation: design leadership won't carry you if you can't deploy at scale. Policy shifts under President Donald Trump could fuel domestic production and infrastructure, but execution still depends on sites, power, and capital discipline.
Capital Is Moving-Fast
Raul Martynek, CEO of DataBank, sees 5-7 GW coming online in the US over the next year to feed AI demand, per Fortune. He pegs data centre costs at US$10-15 million per MW; with smaller builds around 40 MW, the projected capacity implies roughly US$50-105 billion in new spend.
Those numbers signal the real constraint: syncing AI adoption with construction windows, power availability, and supply chains-without blowing up your balance sheet.
Executive Checklist: Turn Strategy Into Megawatts
- Set a two-track plan: lease capacity for immediate workloads; build for cost efficiency and control.
- Commit to power first: secure multi-year PPAs and interconnection queues before ordering hardware.
- Adopt modular, repeatable designs to save months and cut change orders.
- Diversify accelerator supply and cooling approaches to avoid single-point delays.
- Phase capital with clear stage gates tied to power, permits, and supply milestones.
- Place big training clusters where power is abundant; put inference near users where latency matters.
- Develop an "if-slips" plan: where do workloads go if GPUs or power arrive late?
- Engage policymakers early to accelerate permits and unlock incentives.
The Takeaway
The US still leads in AI technology. But the edge goes to whoever can build faster and secure more power-consistently. Treat construction velocity and energy capacity as core strategy, not back-office logistics.
If you're upskilling leaders and teams to execute on this, see our practical learning tracks by role: AI courses by job.
Your membership also unlocks:
