The Hyperscale Mirage: Can Data Centres Resurrect Carbon Capture’s Broken Economics?

By June 2026, the collision between artificial intelligence and industrial decarbonisation has reached a boiling point. As tech behemoths wage a frantic land grab to lock down gigawatts of power for sprawling, next-generation data halls—such as OpenAI’s freshly minted Stargate facility (“The Barn”) in Saline Township, Michigan—they are doubling down on a historically fraught panacea: Carbon Capture, Utilisation, and Storage (CCUS).

Advocates promise that by injecting emissions deep into subterranean saline aquifers, up to 90% of data centre CO2 emissions can be safely locked away forever. But this grand ambition runs headfirst into a cold, stubborn reality. Three decades of CCUS history are littered with broken promises, eye-watering price tags, and abandoned projects.

The Hyperscale Pivot to Onsite CCUS

Spurred by uncompromising net-zero deadlines and a staggering spike in electricity consumption, hyperscalers are moving past cheap, passive carbon offsets. They are betting instead on active, onsite mitigation. This shift is not a voluntary green PR stunt; it is a desperate response to an AI-driven energy bottleneck. Unlike standard cloud architectures, training and operating state-of-the-art Large Language Models (LLMs) demands relentless, around-the-clock “baseload” power. Wind and solar are far too fickle, and grid-scale battery systems remain prohibitively expensive to keep multi-gigawatt campuses humming through the night. Consequently, tech giants are drifting back toward natural gas. To square this fossil-fuel dependence with their public climate oaths, onsite CCUS has transformed from an exotic R&D experiment into a harsh operational necessity.

The current pipeline of projects reveals a frantic, capital-heavy dash, even if the execution remains messy:

• Prometheus Hyperscale’s ambitious 1.5 GW Carbon Negative Digital Infrastructure Campus in Casper, Wyoming, has officially booted up its inaugural phase. Yet, as mid-2026 sets in, the project has already run into heavy-industrial headwinds, plagued by supply chain logjams for bespoke compressors and a painful 15% budget blowout. Tech money, it seems, cannot entirely escape the laws of physical engineering.
• Google’s Broadwing Energy Center in Decatur, Illinois (400 MW), alongside a massive proposed 1-to-3 GW facility in Southeast Nebraska, are both racing to begin operations with integrated carbon storage before the decade is out.
• Fresh survey data from 2026 indicates that 31% of US data centre sites running on onsite power expect to deploy CCUS by 2030, a metric slated to climb to 41% by 2035.

The Grim Reality of Historical Failure Rates

Despite the sparkling corporate optimism, independent data paints a far bleaker picture. Historically, the mortality rate of large-scale, policy-backed CCS projects is a jaw-dropping 88%. Even with today’s advanced engineering, subsector-adjusted forecasting models in 2026 still project a discouraging 76% failure rate for currently planned facilities. This high-risk trajectory looks remarkably similar to other capital-intensive clean-tech ventures, like floating offshore wind, which suffers from a 90% failure rate during its early market penetration phase.

Four structural bottlenecks continue to choke the commercial life out of onsite data centre carbon capture:

1. The Parasitic Energy Load: Running heavy carbon capture machinery requires an immense amount of auxiliary power, triggering severe efficiency penalties. Operators have to burn up to 30% more fuel simply to run the capture loop itself. That is power that never reaches the AI servers.
2. Prohibitive Costs: Point-source capture remains brutally expensive, hovering between $40 to $120+ per ton of CO2 depending on how concentrated the flue gas is. Without aggressive carbon pricing or heavy state subsidies, these costs simply cannot be absorbed by market revenues.
3. The Water-Energy Nexus: Onsite CCUS setups are notoriously thirsty, typically inflating a facility’s water footprint by 20% to 50%. In 2026, with data centres already facing fierce local backlash and strict municipal limits over their massive cooling water consumption, adding a water-guzzling carbon capture loop is a recipe for a localized resource crisis.
4. Permitting Purgatory and Pipeline NIMBYism: In the United States, the real roadblock is bureaucratic and social, not just technological. The Environmental Protection Agency’s (EPA) Class VI injection well-permitting pipeline is backed up by years of red tape. Meanwhile, fierce local resistance to CO2 pipeline corridors has paralyzed transport infrastructure. The result? Captured carbon has nowhere to go.

Key Takeaway: The “auxiliary power paradox” combined with compounding water strain means data centres must consume more energy and local resources simply to capture the carbon generated by their computing infrastructure, threatening to trap operators in an inefficient loop of diminishing returns.

The Geological Bottleneck: Saline Aquifers vs. Depleted Fields

While saline aquifers are geographically abundant and easy for data centres to access on paper, corporate sustainability brochures conveniently gloss over their physical constraints.

• Low Storage Efficiency: The actual storage efficiency of saline aquifers is incredibly poor, typically hovering between 2% and 20%—and plummeting to a mere 0.5% in closed, incompressible water compartments.
• Massive Footprints: Because up to 98% of the underground pore space is already packed with stubborn residual brine, squeezing a set volume of CO2 into an aquifer requires up to an order of magnitude more rock volume than locking it away in a depleted gas field, where storage efficiency can reach 80%.

Comparing the Global Playbooks: US, Europe, and India

As these geological, energetic, and logistical realities set in, the global landscape has split. The US, Europe, and India are deploying vastly different regulatory playbooks to try and de-risk the technology.

Technological Incrementalism in 2026

To bridge this yawning economic gap, developers are moving away from massive, bespoke legacy setups in favour of modular, highly integrated systems. Current R&D is intensely focused on:

• Advanced Solvent Formulations: Amine-based systems engineered specifically for lower thermal regeneration requirements, helping to curb that painful parasitic energy drain.
• Process Integration: Channelling the waste heat directly from data centre liquid cooling systems to offset the thermal energy required for solvent stripping.
• Membrane Breakthroughs: Recent 2026 research from the Institute of Multidisciplinary Research for Advanced Materials has demonstrated heteroatom-engineered covalent organic frameworks (COFs) that successfully bypass traditional CO2 separation trade-offs, offering a pathway to energy-efficient gas separation.

Ultimately, whilst the technical feasibility of capturing data centre emissions is proven, its economic survival remains tethered to heavy policy support. Until India, Europe, and the US can establish unified carbon pricing and resolve the infrastructure bottlenecks of transport and storage, CCUS will remain an onerous, uphill climb.

1️⃣ The Hyperscale Bet: Tech giants are marrying natural gas with CCUS to secure 24/7 AI baseload power. 2️⃣ The Structural Hurdles: Exorbitant costs, permitting backlogs, and a 20-50% water consumption spike threaten viability. 3️⃣ The Geological Limit: Poor saline aquifer efficiency demands massive rock volumes, complicating localised sequestration.