Soben’s Data Centre Trends Report 2026 explores the trends and challenges shaping the industry this year. In this article, Joe Cusick discusses Trend 9 – New Chips on the Menu.
Chips are the new battleground for data center delivery, shaping everything from substation capacity to cooling architecture. Who controls supply and who designs effectively for technological change will determine who scales successfully. As the GPU and AI chip market evolves, Joe explains why this shift introduces cost and program risk for large-scale developments and what developers can do to protect long-term investment.
History repeating
For those who remember the semiconductor surge of the late 1990s, today’s market feels familiar. Then, rapid advances in chip performance reshaped entire industries. Now, AI-driven silicon is once again setting the pace – this time for data center development.
From processing to storage and networking, the data center industry depends entirely on these tiny slices of silicon. Today, Nvidia controls roughly 80% of the data center GPU market. For developers delivering large-scale programs, that level of concentration creates supply, pricing and roadmap risk – accelerating investment in in-house silicon and alternative providers.
Chips as critical program drivers
Nvidia and AMD design their chips but rely on manufacturers such as TSMC, Intel and Samsung for fabrication. This separation between design and production adds another layer of supply chain complexity.
At the same time, hyperscalers are developing proprietary silicon to regain control over performance and supply certainty. Chips are no longer simply equipment to procure. They are long-lead, critical-path dependencies that can determine when a data center becomes operational.
Because chip architecture drives power density and heat output, it shapes everything from substation capacity and cooling systems to rack configuration and long-term flexibility. Infrastructure design must now align to chip strategy from the outset.
When silicon shapes the schedule
As data center design becomes anchored to specific chip platforms, silicon vendors gain strategic leverage, influencing not only performance but delivery timelines and capital decisions. The shift is not just technological but commercial, with leading chip designers and platform providers increasingly backing developer programs to secure ecosystem adoption and long-term demand.
Where capital support or commercial alignment is in place, vendors move beyond product supply and into execution. With fabrication scheduled well before facilities come online, delays in construction can leave suppliers holding high-value inventory or reallocating constrained capacity, creating exposure across the supply chain.
In a tightly allocated market, schedule discipline increasingly matters not only to developers, but to the vendors supporting the underlying compute platforms.
Designing for variability
Vendor leverage is only part of the story. As hyperscalers advance proprietary silicon and new chip platforms enter the market, infrastructure variability is increasing. Different architectures bring different power densities and thermal profiles, requiring greater flexibility in electrical capacity, cooling systems and expansion planning.
Historically, facilities were designed around a known chip platform. That certainty has disappeared. Chip roadmaps evolve rapidly, availability can shift late in the program, and final specifications are often confirmed much closer to deployment. Flexibility is no longer optional.
Chip architecture now dictates substation sizing, electrical distribution, cooling strategy and IT floor space. For hyperscalers building for their own workloads, this may mean more regimented standards. For colocation developers, the challenge is greater.
Multi-tenant facilities must accommodate a range of clients, platforms and density requirements within the same building. A data hall designed today may need to support very different silicon by the time it becomes operational two years later. Without built-in adaptability, developers face redesign, underutilized capacity or missed leasing opportunities.
As silicon providers proliferate and chip performance advances, designing for a single end state is no longer viable. Developers must design for optionality, balancing flexibility, cost discipline and delivery certainty.
Delays are already emerging
The consequences are visible. Where facilities have been designed around fixed chip assumptions, late-stage specification changes are triggering redesign, procurement disruption and schedule pressure.
The industry has faced similar challenges before. When compute performance advances faster than construction timelines, infrastructure must be built with headroom to absorb change.
In practical terms, this means intentional flexibility: electrical capacity that can support higher densities, cooling systems that can transition from air to liquid, and space planning that accommodates future expansion. This is not overengineering. It is risk management.
Crucially, flexibility must be embedded early. Alignment between chip strategy, facility design and construction sequencing at concept stage is now essential. When chip roadmaps and build programs are considered together from the outset, developers reduce redesign risk and protect both schedule and capital efficiency.
Geopolitics and supply risk
Infrastructure flexibility is only part of the equation. Governments are investing heavily in domestic semiconductor manufacturing, particularly in the United States, reshaping global production capacity and supply chains. While this may expand long-term output, it also reflects a more fragmented and politically sensitive environment.
Tariffs, export controls and shifting trade policy introduce volatility. At the same time, constraints on labor mobility and specialist visas add pressure to large-scale programs that rely on globally distributed expertise.
These factors sit outside a developer’s direct control, but they cannot be ignored. Chip manufacturing remains concentrated and globally interdependent, meaning geopolitical disruption can quickly translate into procurement delay.
Resilience must therefore extend beyond facility design to supply strategy and program planning.
Turning uncertainty into strategy
The message is clear: flexibility must be embedded from the outset.
Chips are no longer a downstream IT decision. They drive cost, schedule and risk. The earlier their implications are assessed, the more options developers retain and the less expensive change becomes.
The real value lies in front-end alignment. Defining decision gates early, understanding how long chip specifications can remain open, and modeling the impact of different density and cooling scenarios protects both schedule and capital efficiency.
Most disruption occurs when chip assumptions shift mid-construction. Late changes cascade through electrical systems, cooling infrastructure and procurement, creating avoidable delay and cost pressure. Targeted flexibility and disciplined decision timing are far more effective than reactive redesign.
In a market evolving this quickly, certainty does not come from fixing a single outcome. It comes from designing for optionality.
Chips are now fundamental to data center delivery. Developers who treat them as an early strategic driver rather than a late procurement item will be best positioned to control cost, protect schedule and capture opportunity.
Soben is now part of Accenture’s growing Infrastructure and Capital Projects practice where our hands-on expertise in delivering complex capital projects is complementing Accenture’s deep digital and industry knowledge across the data centre project lifecycle. Our combined team is helping organizations around the globe to design, build, and run next generation data centers that set the stage for growth, agility, and smooth operations.
Click here to find out more about Accenture’s Data Centre Services capabilities.