AI capex and cloud infrastructure: the new race for chips

Competition in AI infrastructure is shifting to chips, cloud and connectivity: costs, lock-in and capacity become the real strategic levers.

AI capex and cloud infrastructure: the new race for chips
Data centre and AI chips in a cloud infrastructure strategy context

Competition in AI infrastructure is entering a more aggressive phase: it is no longer enough to offer computing capacity, you must also control costs, performance and dependence on suppliers. In this scenario, those who buy cloud and chips look for more predictable alternatives, while the large players push proprietary semiconductors, custom CPUs and satellite connectivity.

The strategic message is clear: massive investment in data centres and hardware is no longer presented as a gamble, but as a response to demand that is already there. For decision makers, the point is not who “wins” the media comparison, but how negotiating leverage, technological lock-in and the availability of critical capacity change over the next spending cycles.

AI chips: price, performance and supply chain control

The narrative around the AI supply chain is shifting from a market dominated by a few accelerators towards a more varied mix of proprietary chips and heterogeneous platforms. The value lies not only in raw power, but in the ability to reduce the total cost per inference and training, improve integration with cloud services and guarantee availability at scale.

For enterprise customers, this means looking more carefully at the portability of workloads, the maturity of the software stack and how easily models and pipelines can move between different architectures. In such a concentrated market, even a small change in price-performance can shift budgets and roadmaps.

Custom CPUs and workload migration

The widespread adoption of CPUs designed in-house by hyperscale clouds signals a structural shift: competition is no longer only about the price of an instance, but about the overall efficiency of the service. When a platform can replace general-purpose architectures with optimised silicon, it gains better margins and a stronger negotiating lever over its customers.

For companies running mixed environments, the consequence is concrete: the benefits of application modernisation increase, but so does the cost of staying on legacy stacks. Organisations that want to preserve flexibility need to invest in containers, observability, compatibility testing and rigorous governance of hardware dependencies.

Satellite connectivity and new enterprise markets

Space connectivity is also becoming a piece of the digital infrastructure. Positioning a satellite service as an enterprise alternative concerns not only remote areas, but also the resilience of multi-site networks, operational continuity and the expansion of services for sectors with global coverage requirements.

The fact that transport operators, telecoms and public institutions are testing or adopting these services shows that the market is moving from the experimental phase to the commercial one. For IT and operations managers, the question is no longer whether alternative connectivity exists, but where it makes sense to place it within the business continuity strategy.

Capex, market expectations and the risk of overinvestment

When a hyperscaler commits hundreds of billions in capex, the real issue becomes the timing of the return. Data centre expansion makes sense if demand remains strong, but it also exposes the company to two risks: unused capacity and margin compression if growth slows. Public narrative tends to emphasise confidence, while the market evaluates execution, time to monetisation and visibility on contracts.

At the same time, large agreements with AI customers strengthen the investment thesis, but they do not eliminate the problem of demand concentration. If a few large buyers absorb enormous capacity, the risk of commercial dependence remains high. For technology leaders, this is a key lesson: infrastructure must be scaled together with the diversification of the customer portfolio.

Operational takeaways

  • Evaluate AI strategy not only on the vendor, but on total cost, portability and software maturity.
  • Reduce lock-in with modular architectures and workloads that are portable across different chips and clouds.
  • Integrate alternative connectivity and network resilience into business continuity plans.
  • Tie infrastructure investments to contracts, visible demand and realistic usage scenarios.
  • Monitor the shift from competition on hardware to competition on ecosystem and execution capability.