AI Data Centre Shortage 2026: Demand Outstrips Supply

The AI boom has hit a physical wall. According to a new Jefferies report, global demand for data centres is far outstripping available supply despite record spending by tech giants. The shortage is now “structural” and expected to last years as AI adoption accelerates worldwide.

Demand vs Supply: The Numbers

Hyperscalers like Amazon AWS, Google Cloud, Microsoft Azure, and Meta are pouring billions into AI infrastructure. AWS just reported rising profit as it deepens its AI push with OpenAI. Google Cloud crossed the $20B milestone with 22% revenue growth, driven by AI workloads.

But Jefferies says it’s not enough. AI models require massive compute power, cooling, and energy. Every new ChatGPT, Gemini, or Claude query needs data centre capacity. The result: waitlists for GPU clusters, delayed cloud launches, and soaring rental prices for data centre space.

Why the Shortage is “Structural”

Three bottlenecks are choking supply:

1. Power: Data centres consume massive electricity. In the US, UK, and even South Africa, grid capacity is struggling. Nvidia’s new “AI factories” for robotics and mobility need dedicated power plants.
2. Chips: Nvidia GPUs remain the bottleneck. TSMC and Samsung can’t manufacture enough AI chips fast enough. Vietnam is now beating India in attracting chip plants due to faster execution.
3. Construction: Building a hyperscale data centre takes 2-4 years. Land, permits, cooling systems, and fiber all add delays while AI demand doubles every 6 months.

The South Africa Angle

SA isn’t immune. Amazon Prime just launched locally, and cloud adoption is growing fast. But Eskom’s load shedding + grid constraints make SA a tough market for hyperscale builds. That’s why companies like Teraco and Vantage are expanding, but slowly.

Meanwhile, Adani Ports secured a 10-year LNG contract to create Argentina’s first LNG export corridor to India. Energy access will decide which countries win the AI race.

Who Benefits from the Shortage?

Winners: Nvidia, AMD, data centre REITs, energy companies, and cooling tech firms. Nvidia just launched a 1-million-core chip for AI factories.

Losers: Startups and smaller AI firms stuck in GPU waitlists. Cloud costs are rising as capacity gets rationed.

What Happens Next?

Analysts expect 3 things in 2026-2027:

1. More “AI factories”: Purpose-built facilities like Nvidia + LG’s new plant, designed only for AI workloads.
2. Energy deals: Tech firms signing direct deals with nuclear, solar, and LNG providers to secure power.
3. Geographic shift: Countries with cheap, stable power like Saudi Arabia, Norway, and parts of the US will attract more data centres.

Bottom line: AI isn’t just a software story anymore. It’s an infrastructure story. Until power grids and chip fabs catch up, the data centre shortage will slow AI innovation and raise costs for everyone.

What do you think? Will South Africa attract AI data centres, or will load shedding push them away? Comment below 👇