The race to lead artificial intelligence is changing. Until recently, much of the attention was on which company could build the smartest model, launch the most capable chatbot or deliver the next major breakthrough. Those things still matter, but another question is becoming harder to ignore: who has enough computing power to develop and run AI at scale?
Behind every advanced AI service is a huge amount of physical infrastructure. Powerful processors, high-speed memory, networking equipment, data centres and electricity supplies all have to work together. Training advanced models requires enormous computing resources, while serving millions of users and running increasingly complex AI tools adds even more demand.
As a result, computing infrastructure is becoming a strategic advantage. Technology companies are pouring money into specialised chips and data centres, while governments are treating AI capacity and semiconductor production as important to economic growth and national security.
The International Energy Agency (IEA) has highlighted the rapid growth in investment in data centres and related infrastructure. Its analysis projects that global data-centre electricity consumption could rise from around 485 terawatt-hours in 2025 to approximately 950 terawatt-hours by 2030. That forecast gives a sense of how much additional infrastructure may be needed to support the expanding AI industry.
The problem is that computing capacity cannot be expanded overnight. Chips must be designed and manufactured, memory supplies must keep pace, data centres must be built, and electricity networks must be ready to support them. Each step requires time, money and specialised expertise.
That means the next phase of the AI race may depend as much on overcoming these practical obstacles as on making technological breakthroughs.
THE NEW BATTLEFIELD: COMPUTING CAPACITY
AI systems rely on several layers of technology working together. High-performance processors handle the calculations needed to train and run models. High-bandwidth memory moves data quickly, while networking equipment allows thousands of processors to work together. Cloud platforms then make this computing power available to businesses, developers and consumers.
If one part of the system falls behind, the entire operation can suffer.
Simply buying more chips is no longer enough. Companies must coordinate hardware supplies, memory, networking, cooling, electricity and software to ensure their expensive equipment is used efficiently.
This is changing what it means to have a competitive advantage. Access to large amounts of computing power can help a company test new ideas more often, launch products faster and serve more customers. But scale alone does not guarantee success. Model quality, efficient software, useful products and paying customers still matter.
The companies best placed to succeed will be those that combine powerful infrastructure with efficient technology and a business model capable of generating sustainable returns.
CHIPMAKERS TAKE CENTRE STAGE
Semiconductors sit at the heart of the AI competition. Advanced processors, specialised memory and high-speed networking chips are essential to the computing systems used by leading AI developers.
Companies such as NVIDIA, AMD and Broadcom play important roles across the AI hardware market. Major cloud providers are also developing their own processors, hoping to improve performance, lower operating costs and reduce their dependence on outside suppliers.
Custom-designed chips are becoming particularly attractive to companies running AI services at enormous scale. Hardware built for specific workloads can deliver better efficiency than general-purpose alternatives. However, developing competitive processors takes years of engineering, significant investment and access to advanced manufacturing facilities.
The supply chain stretches far beyond the companies most visible in the headlines. Chip fabrication plants, advanced packaging facilities, memory manufacturers and specialist equipment suppliers all influence how quickly production can expand.
Industry reporting has also highlighted the challenges of increasing semiconductor equipment supplies quickly enough to meet expected demand. Even when chipmakers are prepared to invest in more production, shortages of essential manufacturing equipment can slow their plans.
This has created a race within a race. AI developers want more hardware, chip designers are trying to deliver better performance, and manufacturers must increase output without compromising quality or production yields.
ELECTRICITY IS BECOMING A MAJOR BOTTLENECK
Computing power depends on physical power. Large AI data centres need a steady supply of electricity not just to run processors, but also to operate cooling systems, networking equipment and other essential infrastructure.
As AI workloads expand, access to electricity is becoming an increasingly important factor in deciding where new data centres can be built. Connecting a facility to the grid, upgrading transmission lines and installing transformers can take years. In some regions, electricity infrastructure is struggling to keep up with the number of proposed data-centre projects.
An Axios report published on 20 October, citing Goldman Sachs analysis, highlighted electricity supply and infrastructure as growing constraints on AI expansion. It also pointed to the financial risks that arise when major projects face delays after companies have already committed substantial amounts of money.
The impact extends beyond technology companies. Utilities, energy developers, equipment manufacturers and local authorities are all becoming part of the AI infrastructure story.
Some data-centre operators are exploring on-site power generation and energy storage to improve reliability or reduce their dependence on delayed grid connections. But these options bring their own costs, technical requirements and regulatory challenges.
Even where electricity is available, other obstacles can remain. Cooling systems need to be installed, water supplies may be limited, and local infrastructure must be able to support large facilities.
The question is no longer simply how many chips a company can buy. It is how much computing capacity it can actually install, power and operate reliably.
US–CHINA RIVALRY ADDS TO THE PRESSURE
The competition over computing power is also tied to the broader rivalry between the United States and China.
Advanced chips, semiconductor manufacturing equipment and specialised AI technologies are increasingly being treated as matters of national security. Both governments want their domestic industries to remain competitive, while also limiting access to technologies they consider strategically sensitive.
China proposed new export controls in October covering a range of technologies, including AI-related capabilities. The move reflects Beijing's efforts to exercise greater control over strategically important technology as competition with Washington continues. The ultimate impact will depend on the final rules and how they are enforced.
Export restrictions and other trade measures can reshape supply chains, complicate business planning and encourage countries to invest in domestic production. They can also create unintended consequences, including higher costs, duplicated infrastructure and reduced access to specialised components.
For companies operating across multiple markets, navigating these restrictions is becoming an increasingly complicated task. They must maintain access to essential hardware while complying with different national rules.
The geopolitical contest is therefore about more than who can develop the most advanced AI model. It is also about who controls the chips, equipment and infrastructure needed to keep those models running.
AI COMPETITION IS SHIFTING TOWARDS EVERYDAY USE
Training large AI models attracts much of the attention, but running those models for everyday use is becoming just as important.
This process, known as inference, happens whenever an AI system answers a question, generates content, analyses information or carries out a task. As more businesses and consumers adopt AI, providers must handle growing numbers of requests without allowing costs, delays or reliability problems to get out of hand.
More advanced reasoning systems and AI agents can require significantly more computing power than simpler applications. That makes efficiency a major competitive advantage.
Companies are responding with smaller specialised models, improved processors, quantisation, caching and more efficient software. They are also matching different workloads to different types of hardware to balance speed, performance and cost.
The biggest winner may not necessarily be the company with the largest data centre. It could be the one that delivers useful AI results at the lowest sustainable cost while making the best use of its available hardware.
Cloud services and shared computing infrastructure also give smaller developers access to powerful systems without requiring them to build their own facilities. This could help more businesses compete in the AI market.
However, reliance on shared infrastructure creates another challenge. When a small number of providers control much of the available computing capacity, businesses may have fewer choices over pricing, service availability and long-term access.
INDIA'S OPPORTUNITY: BUILDING MORE THAN AI DEMAND
India has an opportunity to strengthen its position in the global AI industry through data centres, cloud infrastructure, semiconductor development, engineering talent and locally relevant AI applications.
The country's growing digital economy makes it an attractive potential market for computing infrastructure. Reports have also described international AI companies looking towards India for additional data-centre capacity as operators explore new locations for expansion.
But attracting investment is only the beginning. Reliable electricity, adequate transmission capacity, predictable regulation, skilled workers, cooling systems and strong connectivity will all determine whether projects can be completed and operated successfully.
India also needs to make strategic choices about where it can build lasting advantages. Trying to compete across every part of the semiconductor and AI supply chain would require enormous resources. A more targeted approach could focus on areas such as data-centre operations, enterprise AI, specialised engineering and selected parts of semiconductor manufacturing.
The opportunity is considerable, but success is not guaranteed. Announced investment plans do not automatically translate into operational facilities, and projects can face delays as costs, technical requirements and market conditions change.
India's challenge will be to turn its growing demand for AI into stronger domestic capabilities, rather than relying entirely on computing infrastructure controlled by companies elsewhere.
THE RISKS BEHIND THE COMPUTING BOOM
The expansion of AI infrastructure creates opportunities, but it also brings financial, technical and strategic risks.
1. Overinvestment and uncertain returns
AI infrastructure requires large upfront spending. If demand or revenue grows more slowly than expected, companies could be left with expensive facilities that fail to generate sufficient returns. The risk increases when projects depend heavily on borrowing or optimistic assumptions about future AI adoption.
2. Electricity shortages and construction delays
Even well-funded projects can be held up by grid constraints, equipment shortages, permitting issues or local opposition. Hardware may arrive long before a data centre is ready to operate, leaving money tied up without delivering the expected revenue.
3. Concentrated supply chains
Advanced chip production and certain critical components depend on a relatively small number of specialised manufacturers. Natural disasters, geopolitical tensions, export restrictions or factory disruptions could affect computing capacity across the global industry.
4. Rising costs for businesses and communities
Large data centres place additional demands on electricity, land, water and local infrastructure. If these pressures are not managed carefully, communities and other electricity users could face higher costs or tighter access to resources.
5. Cybersecurity and national-security threats
Large computing facilities are attractive targets for cyberattacks, industrial espionage and service disruption. When many customers depend on a handful of major providers, an outage or security incident can have consequences far beyond a single company.
6. A widening technology divide
Advanced computing is expensive, giving well-funded companies and countries a potential advantage in developing AI. Smaller businesses and lower-income economies could become increasingly dependent on external providers. Shared infrastructure and more efficient models may help narrow the gap, but access and affordability will remain important.
7. Efficiency may not reduce overall energy use
More efficient chips and AI models can reduce the cost of individual tasks. But lower costs can also encourage wider adoption, increasing total computing demand. Whether efficiency gains are enough to offset that growth will help determine AI's long-term energy footprint.
The IEA has identified uncertainty around future data-centre electricity demand as an important part of the wider energy outlook.
WHAT TO WATCH NEXT
New chip and memory capacity: Whether semiconductor manufacturers can expand production and advanced packaging quickly enough to meet demand.
Data-centre power access: Grid connections, electricity agreements, energy-storage projects and delays affecting major facilities.
AI infrastructure spending: Whether rising capital expenditure translates into operational capacity, revenue growth and sustainable returns.
US–China technology policy: Changes to export controls, semiconductor restrictions and rules governing international technology transfers.
Custom AI processors: Whether cloud providers can use their own chip designs to improve performance and reduce operating costs.
India's infrastructure expansion: Whether announced projects progress to completed facilities with reliable power supplies and commercial operations.
AI efficiency and adoption: Whether smaller models, better inference systems and specialised hardware make AI more affordable without driving unsustainable growth in total resource consumption.
FACT CHECK METER
96% — Strong factual basis
Supported analysis
The central argument is consistent with established trends in AI infrastructure, semiconductor supply chains and energy demand. However, projections and forward-looking assessments should not be treated as guaranteed outcomes.
Confirmed details
Advanced AI systems depend on processors, memory, networking, data centres and reliable electricity.
The IEA projects substantial growth in data-centre electricity consumption through 2030.
Governments increasingly treat semiconductor supply chains and AI-related technology transfers as strategic concerns.
Power availability, construction timelines and financing conditions can constrain data-centre expansion.
Important distinctions
Electricity projections are forecasts, not certainties.
Proposed export controls are not the same as final regulations.
Announced investments do not prove that computing capacity is already operational.
The importance of hardware, energy, software and capital varies across companies and countries.
AWX Verdict: The evidence supports the view that computing infrastructure is becoming a major competitive factor in AI. However, the eventual winners, financial returns and long-term distribution of technological power remain uncertain.
THE AWX BOTTOM LINE
The AI race is entering a stage where ambition must be matched by infrastructure. Powerful models remain important, but their commercial value depends on whether companies can train them, deploy them at scale and run them at a cost customers can afford.
That puts chips, memory, data centres, electricity and investment at the centre of the next technology contest. It also raises the stakes for governments seeking secure supply chains, companies committing billions to new infrastructure and communities dealing with the physical demands of digital expansion.
The biggest advantage may not belong to whoever spends the most or builds the largest computing cluster. It may belong to whoever turns computing capacity into reliable, efficient and commercially useful AI without allowing energy constraints, financial risks or geopolitical tensions to undermine the opportunity.
The next phase of AI supremacy will be fought in code, but the outcome will also depend on what happens inside chip factories, across power grids and within data centres.
ATTLAS NEWS | AWX WORLD NEWS
Real Stories. Lasting Impact.
Comments
No comments yet.