As AI-driven workloads push data center infrastructure to new limits, the challenges of power delivery, thermal management, and IT equipment operability within the new data center infrastructure space are becoming increasingly complex, causing a significant hit on deployment speed.

To better respond to a rapidly changing industry, senior leaders at Flex are helping shape the next generation of AI infrastructure. Chris Butler, president of Flex’s embedded and critical power business, leads efforts to enable customers to harness power from the grid and efficiently deliver it all the way to the chip level.

Rob Campbell, president of Flex’s communications, enterprise, and cloud business, leads initiatives focused on compute, networking, and storage at scale, along with systems integration, liquid cooling, and deployment strategies designed to support the evolving demands of AI and cloud environments.

The currents shaping the AI data center

While many foundational elements of the data center – such as racks housing IT equipment, power infrastructure, and thermal management – have long been central and remain so in the AI era, the way these elements are designed, scaled, and operated continues to evolve, alongside additional core concepts that are now fundamentally shifting.

Earlier generation data centers were built around general-purpose CPUs running at roughly 150 watts per chip. By contrast, modern AI-focused data centers increasingly rely on high-performance processors and accelerators, including GPUs that can consume closer to 3,000 watts per chip – representing a dramatic increase in power density. While not every data center will transition to GPU-driven AI workloads, those that do are actively exploring more efficient ways of thermal management and power delivery.

Beyond obvious upgrades in technological capability, the most significant change in the AI data center is how these facilities are cooled. Traditional air cooling is no longer sufficient, with liquid cooling solutions steadily infiltrating larger portions of the data center white space. For example, in addition to AI-focused processors and accelerators being almost entirely liquid cooled, AI networking platforms increasingly require dedicated liquid cooling systems to manage the added thermal output. Simply put, where there is more power, there is more heat.

These facilities not only require more power, but they also need equipment capable of handling it. Butler further contextualizes this challenge:

“Power, heat, and scalability challenges are hallmarks of the AI era. We work closely with hyperscalers and chipmakers well ahead of the introduction of new architectures, standards, and products to ensure data center power infrastructure is up to the task as facilities are built or retrofitted for AI and HPC workloads.”

Campbell adds: “An example of this is our strategic collaboration with AMD, where Flex is extending our advanced manufacturing capabilities, resilient supply chain, and US footprint to build the AMD Instinct platform. By combining our advanced manufacturing capabilities with AMD’s high-performance computing leadership, we are helping data center operators scale AI infrastructure.”

The structure that carries the AI data center wave

To shoulder the rising wave of AI infrastructure at scale, we can break the challenge into distinct sets: power density, thermal complexity, balancing deployment timelines with quality and reliability, and multi-vendor integration.

It’s powerful and it’s dense

The sharp rise in power required by modern AI systems increasingly strains fixed physical footprints and requires tackling the power density challenge head-on. Butler contextualizes the issue and Flex’s standing in the industry:

“Critical and embedded power equipment comprise data center power infrastructure. Critical power systems deliver energy from the utility grid to the rack, while embedded power systems distribute it throughout the rack to IT equipment and cooling systems. Flex brings deep expertise across both. Our portfolio spans the complete power chain, which is unique among our competitors and allows us to deliver not only specific products, but also integrated, modular solutions that enable rapid leaps in data center compute capacity and efficiency.”

That end-to-end control of power delivery, Butler explains, is what enables Flex to rethink data center architectures as power density and scale accelerate:

“As data center operators grapple with increasing data center power demands, we’re collaborating with customers to develop power architectures that will support today’s AI and HPC workloads along with the next generation of AI factories. We recently announced the industry’s first globally manufactured, fully integrated platform that unites power, cooling, and compute into pre-engineered, modular reference designs for gigawatt data centers. It lets them deploy infrastructure up to 30 percent faster, which is a huge advantage in a hypercompetitive space.”

Keeping cool…or not

As noted earlier, more power means more heat – and that heat has to be dissipated via cooling systems to keep IT equipment performing optimally.

JetCool, a Flex company, provides end-to-end liquid cooling infrastructure based on novel microconvective cooling technology, from cold plates and manifolds to coolant distribution units (CDUs) supported by Flex’s global manufacturing capabilities to enable scalable deployment across hyperscale AI environments.

“The combination of Flex and JetCool gives us an end-to-end solution that covers data center power and all of the cooling hardware and secondary fluid networks needed to bring cooling down to the chip level. This integrated capability helps hyperscalers, neocloud providers, and chipmakers deploy high-density AI systems faster and at scale. JetCool has some unique cold plate designs that are more efficient than others in the market,” says Campbell.

Moreover, in a development that underscores the fast-changing dynamics of the industry, Nvidia’s announcement at CES of its intent to design next-generation chips that do not require chilled fluid for liquid cooling – relying instead on water at ambient temperatures – sent shockwaves through the HVAC sector.

Campbell explains that this shift has raised concerns among industrial chiller manufacturers, but notes that Flex's JetCool technology is well-positioned to support efficient chip cooling without chilled water, an advantage that could become increasingly important as the industry moves in this direction:

“We have strong capabilities in liquid cooling at the cold plate and rack level, while partnering where needed at the chiller level. JetCool technology enables effective chip cooling at ambient temperatures without chillers – a model we’ve successfully used with customers for years and one that supports the industry’s push away from chilled water.”

Getting flexible and modular

Modular infrastructure components, such as the aforementioned Flex AI infrastructure platform, combine IT hardware and CDUs into prefabricated units that can be integrated into the data center environment quickly.

These modular building blocks – such as IT, cooling, and power modules – allow customers to start with pre-defined solutions, customize them as needed, and preassemble them in the factory. This significantly reduces the time it takes to deploy new data centers from concept to completion, enabling faster, more scalable, and more efficient deployment of next-generation AI data centers. It also derisks onsite deployment. Butler explains:

“Rather than mobilizing hundreds of technicians to cut, weld, and assemble thousands of components across remote sites, modular units are built in a controlled, de-energized factory environment.”

Campbell builds on this perspective, noting that the shift to modularization is being driven as much by labor and location constraints as by speed:

“Data center operators, especially the hyperscalers, have come to the realization that the time and effort required to install hardware onsite – whether it's power hardware, IT hardware, or cooling hardware – is entirely too long.

“It also requires skilled tradespeople who aren't always available when and where they’re needed, as data center operators in search of land, water, and power choose sites farther afield from urban centers. I think we'll see a big move toward modularization. It’s a move that Flex enables and fully supports, whether in the form of units such as PDUs and CDUs or fully integrated power pods and skids.”

Campbell further emphasizes the global scale of Flex’s operations, noting that the company has a significant footprint in the Americas, EMEA, and Asia Pacific – more than 48 million square feet of advanced manufacturing capacity globally.

Butler adds: “By integrating and delivering the full grid-to-chip solution, we help customers deploy faster and reduce vendor complexity. We build or supply most components in-house, while bundling additional elements into fully integrated systems. This gives customers a one-stop shop – from design and integration to delivery of fully commissioned systems they can turn on upon arrival.”

Core coordination and multivendor integration

The race to deploy AI hardware and accelerate time to revenue is driving collaboration across siloes and bringing the value of partnering with integrated solution providers such as Flex into focus. Butler explains:

“A year ago, separate teams handled energy sourcing, critical power systems, data center infrastructure, embedded power solutions, and IT equipment decisions. Last summer, there was an epiphany that IT-driven increases in energy demand – think about those rapid advancements in AI accelerators that are now outpacing Moore’s Law – weren’t being coordinated with the power team, creating a major risk.”

Butler goes on to note that the industry has begun rethinking how to better align these organizations. Today, there is a far more informed and unified understanding that a core power strategy requires all parts of the business to work together and align on what’s feasible and necessary.

That level of partnership extends beyond internal collaboration to company-customer relationships. For those wondering how a company that appears consistently at the forefront of its industry benchmarks success with few direct comparisons, Flex takes a customer-centric approach – tracking its progress through the real-world outcomes of its customers.

Customers rely on Flex to solve practical, high-stakes challenges, rather than theoretical ones. By investing across its broad umbrella of specialities, Flex has built an extensive ecosystem of relationships and partnerships. As a result, customers often engage Flex two or three years ahead of deployments, providing early visibility into future requirements and enabling Flex to develop and design new architectures well in advance.

Butler highlights one example in which a hyperscaler approached Flex for help addressing severe power spikes in an AI rack deployment. Flex’s solution – using a capacitor-based energy storage system to smooth power peaks – demonstrates the company’s ability to innovate and solve complex problems rooted in reality.

This approach underscores that trusted partnerships and proven results serve as the most meaningful indicators of Flex’s success, even in the absence of direct industry comparators.

Never coasting

Rounding up, Butler leaves us with the reminder that it’s important to recognize how early we still are in the AI deployment cycle. Many of today’s implementations are effectively pilot projects rather than full-scale rollouts.

“True scale will emerge as 400 to 800 VDC data center power infrastructures and 1MW racks are deployed at volume,” he says, adding:

“Voltage levels will likely continue to rise beyond 800 VDC, as they did with the solar industry’s shift to higher DC voltages. Longer term, this could include medium voltage DC, onsite AC-to-DC conversion, and even new power sources such as small modular reactors in the 2030s. We’re actively working with customers to plan for shifts coming over the next three to ten years.”

AI deployments and ramp-ups are expected to accelerate significantly. One of the largest opportunities to improve efficiency and reduce data center footprint lies in bringing medium voltage power directly into the data hall. While this shift is likely still five years away, it is clearly on the industry’s roadmap.

Flex provides advanced manufacturing capabilities, innovative power and cooling products, and end-to-end lifecycle services that solve data center power, heat, and scale challenges in the AI era. For more information on how Flex accelerates data center deployment worldwide, visit flex.com.