Semiconductors & Advanced Manufacturing
The data center building boom is going truly global — and the industry is starting to sweat the heat
The story in chips right now isn't just about what gets fabricated — it's about where the chips end up living. Data centers, the vast warehouse facilities packed with AI processors, are expanding at a pace that's stress-testing power grids, cooling systems, and capital markets on every continent. This week's signal is almost entirely about infrastructure: who's building, who's backing out, and why keeping these buildings cool is becoming one of the industry's harder engineering problems.
THE GLOBAL BUILDOUT: BIG MONEY, EVERY TIME ZONE
The scale of data center investment is genuinely staggering right now. In Thailand, Digital Edge and B.Grimm — B.Grimm is a Thai industrial conglomerate that's become a significant Southeast Asian data center developer — secured an $880 million green loan (debt financing structured around environmental criteria) to build a 100MW campus outside Bangkok. To put 100MW in context: that's roughly the power draw of 80,000 average American homes, dedicated entirely to running servers.
In Australia, CDC Data Centres signed what it's calling the "largest ever" Australian data center deal — 555MW of contracted capacity for a single unnamed US client, pushing CDC's total contracted capacity above 1 gigawatt. A gigawatt of data center capacity in a single country from a single operator is a milestone that would have seemed implausible five years ago. The "US client" almost certainly means one of the hyperscalers — Amazon, Google, Microsoft, or Meta — which routinely operate at this scale globally.
A Chicago facility changed hands for $750 million when DigiCo Infrastructure REIT (a real estate investment trust specializing in digital infrastructure) sold the property it had acquired from Prologis (an industrial real estate giant) just a year prior. The quick flip suggests demand is liquid enough that well-located data center assets are trading like commodity real estate.
Not every project makes it to completion, though. AWS (Amazon Web Services, Amazon's cloud division and one of the world's largest buyers of semiconductor capacity) quietly absorbed a NZ$45 million hit after walking away from a planned Auckland data center in early 2024. Site selection, permitting, and power availability failures are becoming a real cost of doing business at this pace.
COOLING: THE UNSEXY BOTTLENECK NOBODY CAN IGNORE
Two separate cooling stories in one day's news is a signal worth taking seriously. The reason this is suddenly urgent: AI accelerators — chips like Nvidia's GPUs (graphics processing units, repurposed for AI math) and Google's TPUs (Tensor Processing Units, purpose-built AI chips) — run dramatically hotter than conventional server chips. A single AI server rack can draw 10 to 30 times more power than a standard compute rack, and most of that power becomes heat that has to go somewhere.
Traditional data centers used air cooling — essentially industrial-scale air conditioning. That worked fine for conventional chips. For AI-dense deployments, it increasingly doesn't.
Barocal, a Cambridge-based startup, raised $10 million to develop solid-state cooling — a technology that uses special materials that change temperature when a magnetic field or pressure is applied, rather than relying on refrigerant fluids or fans. The pitch: more efficient, more reliable, no moving parts. $10 million is early-stage, but the fact that investors are funding the basic science of novel cooling at this moment reflects genuine anxiety about whether conventional approaches can scale.
Meanwhile ZutaCore (a sponsored feature, but directionally meaningful) is making the case for direct liquid cooling — running coolant directly over chip surfaces rather than cooling the surrounding air. Their argument is that cooling at the chip source is fundamentally more efficient than trying to chill an entire room. This isn't new technology, but adoption is accelerating as chip thermal density increases.
The cooling problem is a direct semiconductor story: the hotter chips run, the more aggressive the infrastructure required to keep them in spec. It's also an energy story — data centers already consume a substantial percentage of national grids in some countries, and that number grows with every AI chip generation.
LEADERSHIP MOVES IN AI CLOUD
Lambda, one of the leading "GPU cloud" providers — companies that rent AI chip access to developers and researchers who can't afford to build their own clusters — made a significant leadership move, hiring Michel Combes (a veteran telecom executive who previously ran Softbank's Vision Fund portfolio and ran Sprint) as CEO. Co-founder Stephen Balaban shifts to CTO. John Donovan, former AT&T CEO, joins as chairman.
The pattern of bringing in telecom-era executives to run AI infrastructure companies is interesting: it signals that Lambda's investors see the next phase as scaling a capital-intensive, regulated, infrastructure business — more telco than startup. Also notable: Google deployed an AI-powered water sustainability platform in Belgium in partnership with Agua Segura and Agrow Analytics, continuing the trend of hyperscalers applying AI to their own operational challenges (water is a major data center resource, used in cooling towers).
WHAT TO WATCH
The theme to track is the tension between the pace of AI chip demand and the physical limits of the infrastructure that has to contain it. Power and cooling are becoming as strategically important as the chips themselves. The deals happening in Thailand and Australia suggest that Southeast Asia and the Pacific are becoming genuine second and third theaters of the AI buildout — not just appendages of US hyperscaler strategy. And the cooling innovation cycle is only beginning: the chips planned for 2027-2028 will be denser still.
TL;DR - The AI data center buildout has gone fully global — $880M in Thailand, 555MW in Australia, and a $750M Chicago sale in a single day's news - Keeping AI chips cool is becoming a genuine engineering crisis; startups and established players alike are racing to replace traditional air cooling before chip density overwhelms it - AWS took a NZ$45M write-off walking away from a New Zealand project — a reminder that even the largest operators are miscalculating demand at the pace they're building - Lambda's pivot to telecom-style executive leadership signals that the GPU cloud business is maturing from a startup into a capital-intensive infrastructure play
Compiled from 1 source · 8 items
- Data Center Dynamics (8)