Every investor pricing artificial intelligence as a chip story is missing the real bottleneck.
A chip can be designed and shipped within roughly two years. A power grid cannot be rebuilt on anything close to that timeline. This gap, not silicon supply, is what will determine how fast the AI industry can actually grow, and Malaysia is currently proving the point in real time.
Malaysia positioned itself as Southeast Asia’s AI computing hub. It’s now on course to show exactly how much that ambition can strain a national power system.
Data center electricity demand in the country is projected to surge eightfold by 2030, from 8.5 terawatt-hours in 2024 to 68 terawatt-hours, potentially consuming 30% of Malaysia’s entire national power supply.
This is roughly equivalent to bolting Singapore’s entire current electricity demand onto Malaysia’s existing grid, inside six years. Malaysia is not an outlier. In my opinion, it’s a preview of a mismatch every major AI investment market will eventually confront.
Across Southeast Asia, the data center market is projected to reach $30.47 billion by 2030, while regional power generation is expanding at less than 7% annually, and roughly 70% of ASEAN’s grid still runs on coal and gas. CBRE projects a regional shortfall of 15 to 25 gigawatts by 2028.
Nuclear capacity, frequently cited as the eventual fix, remains years away in every market discussing it seriously. Investors treating this as a distant planning problem are underestimating how quickly it is compounding.
The scale extends well beyond one region. Globally, data center capacity is projected to nearly double by 2030, reaching around 200 gigawatts, close to the size of Germany’s entire electricity system. Up to $3 trillion is expected to fund that build-out.
A single AI-related computing task can consume up to a thousand times more electricity than a traditional web search, which explains why a handful of AI facilities can strain a regional grid in ways hundreds of conventional data centers never could.
A data center can be built and switched on within roughly two years. A new transmission line takes 10 to 15 years to permit and build in most jurisdictions. Electricity demand grew at roughly 1% annually for most of this century. AI-driven demand is now growing more than four times faster than the electricity market as a whole.
Grids engineered for gradual, predictable growth are being asked to absorb a step change, on a construction timeline that has not moved at all.
Most AI-themed portfolios remain badly underweighted in the infrastructure this actually requires. Roughly $70 trillion in global energy infrastructure investment is needed through 2060, with $1.5 trillion to $2 trillion required annually just for transmission and distribution, the physical wires moving power from generator to consumer.
Almost none of that capital currently shows up in the AI investment products sold to retail and institutional investors, which remain heavily concentrated in semiconductors, cloud platforms, and model developers.
Demand and the capacity to meet it remain two largely disconnected bets, and this disconnect is where real opportunity and real risk both sit.
Some of the most overlooked opportunities sit precisely in that gap. Dynamic line rating, which monitors real-time conditions such as temperature, load, and wind to extract additional capacity from existing power lines, can unlock 10% to 20% more throughput without a single new line being built.
Given how long new transmission construction takes, technology that meaningfully expands what already exists deserves considerably more investor attention than it currently receives, particularly across Asian grids already approaching their limits.
Gas generation’s quiet return to the mix tells a related story. Renewables paired with batteries have absorbed much of the recent demand growth, but they cannot yet guarantee the firm, dispatchable capacity that a data center running continuously requires.
Long-term power agreements between tech companies and energy majors, including large-scale solar deals paired with dedicated grid capacity, show where serious capital is already committing, even where public equity markets have not caught up.
A further shift deserves close tracking. Data center operators are increasingly investing directly in on-site generation, battery storage, and demand-response systems that let them shift computing loads during peak periods, moving from passive electricity consumers to active participants in grid stability.
Companies enabling that flexibility represent a genuinely distinct investment category from either chips or conventional utilities, and one still largely absent from mainstream AI portfolios.
None of this diminishes AI as a global investment theme. Rather, it reframes where the theme’s real constraints, and its genuine opportunities, currently sit.
Malaysia’s grid is simply the clearest, most immediate example of a mismatch playing out across every region racing to build AI infrastructure. Investors who have built exposure entirely around chips and cloud platforms have priced in demand.
Very few have priced in the physical limit now standing directly in front of it, and this gap will define which portfolios are positioned correctly when it finally bites.
Nigel Green is founder and CEO of the deVere Group.
