Jensen Huang Says Connectivity Is the AI Bottleneck; The Next One Is Power.

On 2 June 2026 at Computex in Taipei, Nvidia's Jensen Huang called Marvell "the next trillion-dollar company" sending its stock up more than 30% in its biggest single-day gain ever. The keynote's title said everything: The Future of AI Depends on Connectivity. Huang's argument was that AI is now a distributed-systems problem, and the bottleneck has moved from how fast a single chip computes to how fast thousands of chips can talk to each other.
He's right. But the bottleneck keeps moving down the stack and the layer beneath connectivity is the one that gets least airtime: electrons.

Why the constraint keeps shifting

First the scarce resource was compute. Then it became interconnect — the optical and networking layer Marvell sells. The next hard limit is simply getting enough reliable power to the rack. You can wire a thousand chips together perfectly and still be stalled by a grid connection that won't arrive for years.

Why power, not chips, is becoming the real bottleneck

The numbers are blunt. The International Energy Agency projects data-centre electricity demand will more than double by 2030 to around 945 TWh roughly the entire electricity consumption of Japan with demand from AI-focused facilities tripling. Meanwhile, grid interconnection queues in many markets run two to three years. As the IEA put it, grid capacity, not chips, may be the real constraint on AI at scale. The AI build-out is no longer a digital overlay on the grid; it's becoming a primary driver of where and how power gets built.

What that means on the ground

The operators who scale will be the ones who treat power as a first-class engineering problem, not a procurement afterthought pairing fast-deployable on-site generation with storage that can absorb the violent, second-by-second load swings these workloads create, on-grid or off-grid. The bottleneck no one priced into the AI trade is the one measured in megawatts.

247 Energy is a Belgian energy company building distributed, on- and off-grid power generation and storage for energy-intensive operations patented LNG/hydrogen power units deployable in under four hours, and supercapacitor storage with zero thermal runaway and 15.000-cycle life. Power, anywhere, without compromise.

Sources: CNBC (Computex 2026); HPCwire (keynote details); IEA, "Energy and AI" / "Key Questions on Energy and AI" (2025–2026).