Why Capital Pouring into AI is Not a House of Cards for Data Centers

Markets recently are instantly rewarding AI spending. Recent headlines seem to describe the circular nature of chipmakers and tech giants investing in AI platforms. The scenario feels familiar to the heady days of the dot-com bubble. But we don’t see this as a headwind for data centers, rather an opportunity - and here is why:

The numbers tell a chilling story. AI chip demand doubles every nine months, but power infrastructure takes years to build. Power for AI has become its own asset class. In the AI economy, Power = Profit, and all the GPUs in the world mean nothing if you can't plug them in.

Regional Variations:

• Singapore/Hong Kong : Will be considered for proof of concepts due to land and power constraints and leveraging on the robust connectivity to the rest of region. 

• Japan and Korea : May be the preferred destinations for large scale AI if power can be sourced in non-traditional locations.

• Australia : Has a case for large scale AI development given its relatively domestic nature. Parts of Sydney and Melbourne will be attractive for such businesses.

• India : A massive market where cloud computing continues to grow. The large domestic market is still underserved compared to other established markets. Cheaper land and availability of power generation will naturally lend itself to a rapidly growing AI market.

• Emerging markets : (Malaysia, Indonesia, Thailand, and Philippines): SEA markets have been developing its infrastructure and policies towards the data center sector. Availability of renewables will also be critical for large scale developments. 
   

The core trend is real - power availability and quality have become primary value drivers in data center real estate, with AI workloads accelerating this dynamic significantly.

The global data center industry is experiencing unprecedented expansion, with Asia Pacific capacity growing at 13.7% CAGR from 2023-2029 to 23.2 GW and revenue increasing to $65.3 billion. This growth is driven by surging demand for big data, IoT, 5G rollouts, and most significantly, the artificial intelligence revolution expected to account for 50% of data capacity by 2030.

But here's the crisis: At the completion of 2025, an estimated 10 GW will have broken ground globally, with 7 GW reaching completion, equating to roughly $170 billion in asset value needing financing. Each hyperscale data center now requires 300-400 megawatts of electricity, enough to power a small city. Multiply that by dozens of centers planned worldwide, and the existing grid simply wasn't built for this demand.

A new category of operators is reshaping the landscape: neoclouds. These specialized cloud providers offer GPU-as-a-Service for AI workloads, encompassing approximately 190 operators with an extraordinary 82% five-year CAGR in revenue growth. S&P Global reports over $10 billion allocated to the sector in the last year, with NVIDIA directly investing in CoreWeave — one of the world's largest AI infrastructure providers.

The game-changer: neoclouds can locate away from population centers, prioritizing power availability over proximity. This represents a fundamental departure from traditional location strategies and JLL predicts a decoupling of AI training and inference facilities. Neoclouds may catalyze investment in traditionally second-tier locations with abundant power supply—exactly what's needed as tier-1 markets reach capacity constraints.

India represents the most compelling growth story, with the market projected to expand at 27.7% CAGR from 2025 to 2029. Mumbai is seeing increases in GPU-as-a-Service availability as hyperscalers develop AI infrastructure, while Chennai benefits from Tamil Nadu's policies encouraging sustainable data hub creation.

Malaysia showcases even more dramatic expansion: capacity grew 243% from 2023 to 2025, from 0.23 GW to 0.79 GW, with further 25.6% CAGR growth expected through 2029. Johor Bahru's 2,570 MW pipeline could potentially exceed Singapore's total capacity, driven by the Johor-Singapore Special Economic Zone and strategic partnerships like ByteDance.

Major cloud providers and chip manufacturers revenue is growing at astronomical rates with some citing sales of 34% year on year. These aren't normal growth numbers, they're signs of future-proof buying capacity before the lights go out.

The financing landscape reflects this urgency. Hyperscalers commit to 5-10 year leases providing stable cash flows, while neoclouds prefer shorter terms for technological flexibility. This creates obsolescence risks as AI technology advancement means specialized infrastructure investments face rapid outdating. Consequently, AI-focused data center developments are increasingly underwritten at higher development yields compared to traditional cloud facilities, reflecting the accelerated obsolescence risk as AI training technologies evolve rapidly.

Development projects targeting neocloud tenants require stronger pre-leasing commitments but command yield premiums. The majority of a data center's asset value lies in existing power supply and building infrastructure, especially valuable given current challenges securing power for new developments.

Every nine months, a new AI chip generation arrives requiring exponentially more power per rack. Nvidia's Blackwell, AMD's latest GPUs, Amazon's custom chips—the demand is relentless. One megawatt isn't just one megawatt anymore; it's become a strategic asset determining who controls technology and who gets left in the dark.

The sector faces $170 billion in financing needs this year alone, with momentum gaining through 2026 while investment volumes grew at 27.8% 10-year CAGR through 2024. Those who successfully navigate the power constraints will benefit from strong-performing assets in a class expected to grow rapidly over the foreseeable future.

This isn't just an energy shortage, it's a once-in-a-century reset of technological power dynamics. The winners will be those who secure electricity access today for the AI economy of tomorrow.