In recent years, AI demand has pushed data center infrastructure to a front-page constraint. Below, MetaProp’s Antonio Chidiac and I examine the wave of innovation emerging to meet the moment — and what it signals for the next phase of digital infrastructure.

How We Got Here

• Global data center electricity consumption is set to double by 2030, growing 4x faster than all other sectors combined
• Data centers now gulp 449 million gallons of water per day in the U.S. alone, triggering $64B in blocked or delayed projects due to local opposition (i.e. data center NIMBYism)
• Liquid cooling has gone mainstream - it’s no longer optional for AI workloads, with the immersion cooling market projected to hit $18B by 2034, growing at a 25% CAGR
• Bezos, Musk, and Pichai all believe orbital data centers will be “normal” within a decade and Starcloud just trained the first AI model in space
• For venture investors, this creates opportunities across cooling tech, energy systems, and space infrastructure

The Infrastructure Crunch

Data centers were once the invisible plumbing of the internet. Not anymore. The AI revolution has turned them into one of the most capital-intensive and resource-constrained sectors in tech.

The numbers are wild. Global data center electricity consumption is projected to grow ~15% annually through 2030 - that’s four times faster than every other sector combined (IEA, 2025). U.S. data centers consumed 183 TWh in 2024, roughly equivalent to Pakistan’s entire electricity demand. And AI is the main culprit: training GPT-4 alone consumed enough energy to power San Francisco for three days.

The grid wasn’t built for this. U.S. data center power demand is forecast to more than double by 2035 (BloombergNEF, 2025). Wholesale electricity prices have risen as much as 267% since 2020 near data center hot spots (Bloomberg, 2025). One study estimates that data centers could increase your electricity bill 8-25% by 2030 depending on where you live (Carnegie Mellon University, 2025).

And it’s not just energy. Data centers are thirsty. Large facilities can gulp 5 million gallons of water per day - the needs of a town of 50,000 people. The result? Over $64 billion in data center projects were blocked or delayed between H2 2024 and H1 2025 due to local opposition (Data Center Watch, 2025). In Warrenton, Virginia, residents voted out every single town council member who supported Amazon’s proposed data center. Data centers have become the new NIMBY battleground.

Cooling Innovation to the Rescue

2025 is the year liquid cooling went fully mainstream. It had to. AI chips generate so much heat that traditional air cooling simply can’t keep up. Rack power densities - power densities of the metal frames holding the servers together - have jumped more than 5x from 15 kW for traditional data centers to 80-120 kW for AI clusters. 

Furthermore, liquid cooling delivers PUE scores below 1.2, reducing energy waste by 20-40% compared to air-cooled facilities (Data Centers, 2025). Still, standard Evaporative cooling demands continuous water consumption, exacerbating a core sustainability issue stemming from data center demand growth.

Direct-to-chip (D2C) / Closed loop cooling has emerged as the go-to for hyperscalers. Cold plates mounted directly on GPUs remove 70-80% of heat at the source. Microsoft’s latest supercomputer uses exclusively liquid-cooled racks. NVIDIA now designs flagship chips with liquid cooling as the preferred thermal solution. And, because water stays inside the pipe in a closed loop, it is recycled over and over, reducing additional consumption in the process.

Immersion cooling takes it further. It submerges entire servers in dielectric fluid, a sort of liquid deep fryer that doesn’t conduct electricity. The more advanced two-phase system - where liquid is boiled into vapor inside the server to absorb heat - can handle heat flux up to 1,500 W/cm². That’s nearly 100x more efficient than standard water cooling mechanisms. The problem is that it’s still CAPEX intensive - a 1 MW system can cost up to $1 million to install. Still, the market is valued at $4.7B today and is projected to hit $18.1B by 2032, growing 24.9% YoY over a mere 6-year period (Yahoo Finance, 2026).

M&A transactions highlight momentum in the space: Eaton acquired Boyd Thermal for $9.5B, Trane bought Stellar Energy Digital, and Carrier scooped up ZutaCore for its waterless HyperCool platform. More deals should be expected.

Notable Startups:

Corintis — Bio-inspired cooling with channels etched into chips. Removes heat 3x better than conventional cold plates. Raised $24M Series A + $25M from Applied Digital. Partnered with Microsoft.

Firmus Technologies — Vertically integrated AI infrastructure with synthetic cooling fluids. Raised $327M Series B to scale to 1.6 GW by 2028.

Crusoe — Closed-loop D2C cooling for AI infrastructure. Raised $1.4B Series E in October 2025.

ZutaCore — HyperCool waterless D2C platform. Acquired by Carrier.

AirJoule — Uses metal-organic frameworks (2025 Nobel Prize in Chemistry!) to extract water from waste heat. Backed by GE Vernova, deploying at a 600 MW Texas facility.

Phasic Energy — Combining AI models with next-gen hardware to capture and transform low-grade waste heat into power and cooling across commercial, industrial, and defense applications.

The Orbital Moonshot

Here’s where it gets interesting. If terrestrial constraints are becoming insurmountable… what about space?

This isn’t sci-fi anymore. Jeff Bezos believes gigawatt-scale data centers will operate in space within 10-15 years. Elon Musk confirmed SpaceX “will be doing” data centers in space. Sundar Pichai said we’ll view orbital data centers as “normal” within a decade. Let the hype cycle begin.

The thesis: orbital data centers could tap unlimited, uninterrupted solar energy (30-40% higher irradiance than Earth’s surface, no night cycles, no weather). Cooling leverages space’s vacuum as an infinite heat sink - no water required. Starcloud estimates energy costs could be 10x cheaper than terrestrial ones. And, lastly, there’s no permitting restrictions required (at least for now). Those certainly sound like strong value propositions.

Besides, things are already moving. In November 2025, Nvidia-backed Starcloud launched a 60 kg satellite with an H100 GPU — the first AI chip 100x more powerful than anything previously in orbit (Nvidia, 2025). It successfully trained NanoGPT on Shakespeare’s complete works and ran Google’s Gemma model from space. “Anything you can do in a terrestrial data center, I’m expecting to be able to be done in space,” CEO Philip Johnston told CNBC.

Google’s Project Suncatcher will launch two prototype satellites by early 2027 to test TPU chips in orbit. They’re envisioning 81-satellite clusters with optical links separated by less than a kilometer to maintain efficient communication. Radiation testing showed their Trillium TPU could handle 3x the expected five-year dose without permanent failures (Google, 2025). In other words, these chips proved to be a lot more resistant than anticipated. 

The catch? While all this excitement and sponsorship indicates there is potential, it is imperative that we navigate through the noise and avoid blindly jumping on the bandwagon. Afterall, launch costs need to drop from ~$1,500-3,000/kg today to under $200/kg for the economics to work (Google, 2025). Radiation hardening adds 30-50% to hardware costs. Thermal management in vacuum is genuinely hard physics that challenges the foundations of thermodynamics and sound economics (Chaotropy, 2025). And carbon emissions from rocket launches could offset sustainability gains if launch tech doesn’t improve and energy requirements don’t become efficient, sustainable, and scalable. 

Nevertheless, the opportunity is captivating and certainly worth our attention. Afterall, these are all major problems worth tackling, so long as capital is allocated with focus, care, and diligence. If the right founders are out there and they’re going about navigating the challenging tasks ahead with rigor, clarity, and poise, smart capital should back them.

Startups to watch:

Starcloud — Backed by Y Combinator, Nvidia, Google, NfX, and In-Q-Tel. ~$24M raised. First to train AI in orbit. Partnering with Crusoe to launch a public cloud in space by 2027.

Lonestar Data Holdings — $120M deal with Sidus for six data-storage satellites. First 15-petabyte system targeting 2027 launch at the Earth-Moon Lagrange point. Lonestar has raised $10 million at a $30 million valuation (Reuters, 2025)

Aetherflux — “Galactic Brain” orbital data center nodes planned for 2027, part of broader space-based solar development.

Star Catcher — The energy grid infrastructure play for orbit. Uses optical power beaming to transmit concentrated solar energy directly to satellites. Plans to take its proven tech stack to orbit this year.

Parsimoni - The cyber security-focused OS for Satellite Payloads. Backed by TechStars and partnered with the likes of Airbus and ThalesAlenia, Parsimoni is taking a software-first approach to building the middleware for space systems.

The Opportunity

The data center sector is at an inflection point. The AI revolution has exposed fundamental constraints in how we build and operate digital infrastructure. For venture investors focused on the built environment, this creates a multi-decade opportunity:

Near-term : Liquid cooling is here and scaling fast. D2C and immersion startups are raising serious capital, with major M&A validating the opportunity.

Medium-term: As terrestrial constraints intensify and launch costs decline, orbital computing shifts from experimental to early commercial.

Long-term: If the tech titans are right, orbital data centers become a standard architecture for the most energy-intensive AI workloads.

Specialized VC funds including Space Capital, Lux, and Seraphim as well as traditional VCs such as a16z, Khosla, Founders Fund, and General Catalyst and corporate funds such as Airbus Ventures and NEOM Investment Fund have all redirected capital flows into space tech over the past 5 years. 

In 2025, private investment in the wider sector grew 48% to $12.4 billion (Reuters, 2026). The global space economy is expected to grow further, with estimates indicating it could hit $1.01 trillion by 2034, implying a 12% CAGR (SpaceNews, 2026). We expect technology that either advances or builds the tech stack and infrastructure for the orbital data center ecosystem to make up a significant portion of future investment growth.

From 2017 to 2025, venture capital invested in space technologies has nearly quadrupled (Pitchbook, 2025). With strong market signaling, massive tailwinds - particularly from data center and energy demand growth - IPOs and M&A deals expected in the coming year, this trend is expected to accelerate. A brief look at the past 8 years gives us some indication into what the future holds:

The boundaries are blending. Cooling tech meets materials science. Energy infrastructure merges with aerospace. Data center ops converge with satellite communications. The investors who spot these convergences early will be positioned to capture the full breadth of this transformation.

The AI age demands infrastructure that doesn’t yet exist at scale. Building it is one of the major opportunities of this era.