ChatGPT developer OpenAI has paused its flagship UK data center project, known as “Stargate UK,” citing high energy costs and regulatory uncertainty. The project was part of a broader £31 billion ($40+ billion) investment plan aimed at expanding artificial intelligence (AI) infrastructure in the country.
The initiative was designed to deploy up to 8,000 GPUs initially, with plans to scale to 31,000 GPUs over time. It was aimed to boost the UK’s “sovereign compute” capacity. This means building local infrastructure to support AI development and reduce reliance on foreign systems.
However, the company has now paused development. An OpenAI spokesperson stated that they:
“…support the government’s ambition to be an AI leader. AI compute is foundational to that goal – we continue to explore Stargate UK and will move forward when the right conditions such as regulation and the cost of energy enable long-term infrastructure investment.”
Energy Costs Are Now a Core Constraint
The main issue is energy. AI data centers require large amounts of electricity to run GPUs and cooling systems.
In the UK, industrial electricity prices are among the highest in developed markets. Recent estimates show costs at around £168 per megawatt-hour, compared to £69 in France and £38 in Texas. This gap creates a major disadvantage for large-scale data center investments.
AI workloads are especially power-intensive. A single large data center can consume as much electricity as tens of thousands of homes. As AI adoption grows, this demand is rising quickly.
Globally, the International Energy Agency estimates that data centers could consume over 1,000 terawatt-hours (TWh) of electricity by 2030, up sharply from about 415 TWh in 2024. This growth is largely driven by AI.
The result is clear. Energy is no longer just a cost. It is a key factor in where AI infrastructure gets built.
Regulation Adds Another Layer of Risk
Energy is only part of the challenge. Regulation is also slowing investment. In the UK, uncertainty around AI rules, especially copyright laws for training data, has created hesitation among companies.
Earlier proposals to allow AI firms to use copyrighted content were withdrawn after backlash. This left companies without clear guidance on compliance.
For large infrastructure projects, this uncertainty increases risk. Data centers require billions in upfront investment. Companies need stable rules before committing capital.
Planning delays and grid connection timelines also add friction. These factors increase both cost and project timelines.
Together, energy costs and regulatory uncertainty create a difficult environment for hyperscale AI infrastructure.
OpenAI’s Global Infrastructure Expands, But More Selectively
Despite the pause, ChatGPT-maker is still expanding globally. The company is investing heavily in AI infrastructure through partnerships with Microsoft, NVIDIA, and Oracle. It is also linked to a much larger $500 billion “Stargate” initiative in the United States, focused on building next-generation AI data centers.
At the same time, the company faces rising costs. Reports suggest OpenAI could lose billions of dollars annually as it scales infrastructure to meet demand.
This reflects a broader industry shift. AI is becoming more like energy or telecom infrastructure. It requires large capital investment, long timelines, and stable operating conditions.
The pause also highlights a deeper issue. AI growth is increasing pressure on energy systems and the environment.
The Hidden Carbon Cost Behind Every AI Query
ChatGPT and similar tools rely on large data centers. These facilities already account for about 1% to 1.5% of global electricity use. Projections for their energy use vary widely due to various factors.
Each individual query may seem small. A typical ChatGPT request can use about 0.3 watt-hours of electricity, which is relatively low. However, usage at scale changes the picture.
ChatGPT now serves hundreds of millions of users. Even small energy use per query adds up quickly. Training models is even more energy-intensive. For example, training GPT-3 required about 1,287 megawatt-hours of electricity and produced roughly 550 metric tons of CO₂.
Newer models are even larger. Some estimates suggest training advanced models like GPT-4 could emit up to 15,000 metric tons of CO₂, depending on the energy source.
At the system level, the impact is growing fast. AI systems could generate between 32.6 and 79.7 million tons of CO₂ emissions in 2025 alone. By 2030, AI-driven data centers could add 24 to 44 million tons of CO₂ annually.
Looking further ahead, global generative AI emissions could reach up to 245 million tons per year by 2035 if growth continues. These numbers show a clear pattern. Efficiency is improving, but total demand is rising faster.
Big Tech Scrambles to Balance AI Growth and Emissions
OpenAI has not published a detailed standalone net-zero target. However, its operations rely heavily on partners such as Microsoft, which has committed to becoming carbon negative by 2030.
The company has acknowledged that energy use is a real concern. Leadership has pointed to the need for more renewable energy, including nuclear and clean power, to support AI growth.
Across the industry, companies are responding in several ways:
- Improving model efficiency to reduce energy per query
- Investing in renewable energy and long-term power contracts
- Exploring new cooling systems to reduce water and energy use
Efficiency gains are already visible. Some AI systems have reduced energy per query by more than 30 times within a year, showing how quickly technology can improve. Still, total emissions continue to rise because demand is scaling faster than efficiency gains.
The Global AI Infrastructure Race
The pause in the UK highlights a larger trend. AI infrastructure is becoming a global competition shaped by energy, policy, and cost.
Regions with lower energy prices and faster permitting processes have an advantage. The United States and parts of the Middle East are attracting large-scale AI investments due to cheaper power and supportive policies.
At the same time, governments are trying to attract these projects. The UK has pledged billions to support AI growth and improve compute capacity. But this case shows that policy ambition alone is not enough. Companies need reliable energy, clear rules, and predictable costs.
AI’s Next Phase Will Be Decided by Energy, Not Code
The decision by OpenAI does not signal a retreat from AI investment. Instead, it reflects a shift in priorities.
Companies are becoming more selective about where they build infrastructure. They are focusing on locations that offer the right mix of energy access, cost stability, and regulatory clarity.
The UK project may still move forward, but only if conditions improve. For now, the message is clear. The future of AI will not be shaped by technology alone. It will also depend on energy systems, policy frameworks, and long-term investment conditions.