How new AI demands are fueling the data center industry in the post-cloud era

The growing use of synthetic intelligence (AI) means a fast enhance in data use and a new era of potential data center industry progress over the subsequent two years and past.

This shift marks the starting of the “AI Era,” after a decade of industry progress pushed by cloud and cell platforms, the “Cloud Era.” Over the previous decade, the largest public cloud service suppliers and web content material firms propelled data center capability progress to unprecedented ranges, culminating in a flurry of exercise from 2020 to 2022 as a result of the surge in on-line service utilization and low-interest-rate financing for tasks.

However, there have been vital shifts throughout the industry in the previous 12 months, together with a rise in financing prices, construct prices and construct instances, mixed with acute energy constraints in core markets. For instance, typical greenfield data center construct instances have prolonged to 4 or extra years in many world markets, roughly twice so long as a couple of years in the past when energy and land had been much less constrained.

Meanwhile, the largest web firms are partaking in an accelerating race to safe data center capability in strategic geographies. For every of the world expertise firms, AI is each an existential alternative and a menace with distinctive challenges for data center capability planning. These dynamics are prone to consequence in a interval of elevated volatility and uncertainty for the industry, and the stakes and diploma of problem of navigating this atmosphere are increased than ever earlier than. 

Flexible data center capability planning can permit for altering inputs in quickly altering markets. Looking again, the Cloud Era gave rise to a very new set of market-propelling prospects with completely different wants than earlier generations. Industry gamers that had been in a position to tackle these evolving wants received an outsized share throughout the final industry cycle. 

Key concerns for data center industry executives and their traders ought to be:

• Scenario planning to capitalize on the evolving wants of the market. 
• Proactive, but versatile, methods for market choice, facility design and different future selections.

New era of shopping for data center capability: Programmatic shopping for is over

During the Cloud Era, public cloud service suppliers turned extra refined in forecasting the ramp-up of demand and adopted a extra programmatic strategy to procuring capability. For a number of years, these patrons usually procured comparatively normal quantities of third-party capability structured with an preliminary dedication, adopted by a reservation and a proper of first supply for the identical amount. However, as demand finally outpaced the unique forecasts, cloud service suppliers (CSPs) needed to return to the marketplace for extra capability. 

Over the previous two years, buyer habits has notably shifted: With the advantage of hindsight, data center prospects are now more and more keen to signal considerably bigger offers, notably in markets the place energy is presently comparatively extra out there to keep away from last-minute scrambles for extra capability and sophisticated footprints. They have additionally demonstrated a willingness to lease capability at increased costs in markets the place capability is constrained. 

Key consideration for executives and traders: 

• Prior fashions and expectations might have adjustment to replicate this Evolution in buyer shopping for habits. 

Self-build data center improvement approaches are evolving

The largest cloud and web firms, the hyperscale patrons in the data center industry, have traditionally most well-liked to construct capability themselves in markets the place there’s vital anticipated demand, potential financial benefit and manageable danger.

However, intense competitors has led these gamers to rely extra on leased capability from third events to get a extra environment friendly path to market. In response, there are indicators that the self-build technique could also be shifting.

Hyperscaler organizations acknowledge that it’s unrealistic to self-build every little thing, and leasing will proceed to play an vital position in capability procurement. As a consequence, hyperscalers are relying extra closely on leasing for velocity to market benefit, whereas additionally contemplating smaller self-builds to doubtlessly offset future demand. This suggests a possible enhance in the whole variety of self-builds and a extra heterogeneous mixture of self-builds and leased capability inside cloud areas and even particular person availability zones. For third-party suppliers, assessing the menace of potential future migration danger, given this dynamic, can be more and more vital. 

Key consideration for executives and traders: 

• The shifting mixture of self-build vs. leasing throughout the industry and inside particular native markets might alter the measurement of the addressable market, execution decision-making, and potential dangers.

Increased energy demand for AI workloads, cooling shift to liquid

AI workloads require power-hungry graphics processor items (GPU), ensuing in a lot increased energy density necessities inside the data center. Currently, the AI market is comparatively homogenous at the server infrastructure degree, with Nvidia holding about 95% of the GPU marketplace for machine studying (ML).

Therefore, the majority of high-end AI workloads are run on related {hardware}: Specifically, chassis consisting of eight of Nvidia’s newest AI-specific GPUs (H100s), with every chassis consuming 5 to 6kW of energy. Up to 6 chassis can match in a single data center rack, ensuing in whole rack densities in the 30 to 40kW vary, in comparison with roughly 10kW/rack densities for commodity public cloud workloads. 

As a consequence, hyperscalers and data center operators should discover methods to successfully cool the tools. Some main hyperscalers have introduced plans to shift to liquid cooling options or elevate the temperatures inside their data facilities to assist these increased densities. 

Key concerns for executives and traders: 

• Current designs ought to assist the future wants of power-dense workloads as densities shift over time.
• Selecting completely different cooling expertise choices might have to think about each financial and sustainability considerations.

Environmental, Social and Governance (ESG) demands

The data center industry ESG concerns are primarily targeted on sustainability. To obtain their sustainability targets, industry members have introduced ambitions associated to renewable vitality utilization, water utilization and discount of their carbon footprints. Data center operators are using a wide range of methods, the place out there, to fulfill these targets:

• Efficiency enhancements
  • Energy-efficient designs utilizing applied sciences reminiscent of free cooling, environment friendly energy distribution and environment friendly lighting methods
• Renewable vitality utilization
  • Procuring renewable vitality from the grid
  • On-site renewable era, together with photo voltaic and wind
  • Power buy agreements (PPAs) for long-term renewable vitality, specifying quantity and worth
• Water utilization
  • Air-cooled methods
  • Closed-loop water methods to scale back water use
  • Rainwater harvesting and water recycling
  • Water-free cooling, reminiscent of evaporative cooling or adiabatic cooling
• Carbon neutrality
  • Energy restoration utilizing warmth from IT tools
• Waste discount 
  • The skill to make the most of these methods will range extensively by market relying on native local weather, native vitality combine and different components reminiscent of the want for employee security. 

Key concerns for executives and traders: 

• ESG technique ought to be differentiated from opponents. 
• An ESG technique ought to try to handle desired, measurable change, or it might run the danger of being labeled as “greenwashing.”

AI plugins: Next wave of ecosystems

OpenAI has lately introduced plugins to assist third-party providers, reminiscent of in style on-line ordering and reservation purposes. These plugins are designed to assist builders entry and combine exterior data feeds immediately into OpenAI’s language mannequin, permitting for extra refined coaching and prompting capabilities.

This new performance might doubtlessly reshape current data center ecosystems round particular industries or data sources. As this dynamic evolves, it is going to be important for operators to determine future “magnets” for these communities of curiosity and supply a related set of connectivity merchandise to assist the wants of those ecosystem members. 

Key concerns for executives and traders:

• To assist ecosystem improvement, the proper set of merchandise, companions and infrastructure is important. 
• It is vital to determine the highest-value prospects in this new market atmosphere and decide how the gross sales group is provided to focus on them.

Conclusion

The stakes have by no means been increased for data center industry members to develop proactive, versatile methods to navigate this new era and construct the proper data center capability in the proper markets. AI is driving elevated data storage demand, which is positioned to outstrip provide in the close to time period. Builders, traders and customers will profit from versatile data center infrastructure methods that may harness the AI revolution and result in outsized progress.   

Gordon Bell is EY-Parthenon principal for technique and transactions at Ernst and Young LLP.

Lillie Karch is EY-Parthenon senior supervisor for technique and transactions at Ernst and Young LLP

The views mirrored in this text are the views of the authors and don’t essentially replicate the views of Ernst & Young LLP or different members of the world EY group.