Yotta 2024, held October 7–9, 2024, at the MGM Grand in Las Vegas, was framed around a practical problem: how to expand AI infrastructure without running out of power, cooling capacity or environmental headroom. Drew Robb’s September 26, 2024 preview in Data Center Knowledge used the event agenda to show how data-center design was being reshaped by high-density compute, edge inference and future workloads.
The preview was not a product comparison or a current market forecast. It documented the issues and speakers scheduled for a 2024 industry event, along with estimates attributed to Goldman Sachs, Omdia analyst Alan Howard, EPRI and academic sources. Those figures should therefore be read with their original date and assumptions in mind.
Why AI was changing the design brief
The central argument of the Yotta 2024 preview was that AI growth affects every layer of a facility. More powerful processors increase rack densities; higher densities alter electrical distribution and thermal design; and the resulting demand makes energy sourcing, water use and construction schedules strategic issues rather than operating details.
Rebecca Sausner, CEO of Yotta Events, said the conference had “assembled the brightest minds across digital infrastructure to discuss actionable solutions that will shape the future.” She also said the industry needed to “scale responsibly” as it dealt with AI and machine learning.
#1 Best Overall
Power availability and architecture
The agenda included a session on power architecture for AI workloads. That focus reflects a distinction between simply obtaining more utility capacity and designing a distribution system that can deliver large, rapidly changing loads reliably at the rack. Developers evaluating an AI-ready site need to examine utility interconnection timing, substation and switchgear capacity, redundancy, rack-level distribution, protection, backup generation and the ability to add capacity in phases.
Thermal management for extreme density
A dedicated session addressed cooling extreme density, while the expo included a Liquid Cooling Coalition Pavilion. The design question is not merely which cooling technology is available; it is whether the building, heat-rejection plant, piping, controls, maintenance procedures and staff can support the intended rack power for the facility’s full life.
Sustainable expansion
The sustainability track covered energy efficiency, renewable power sources and advanced cooling. Schneider Electric and Shell Energy were named in that context, alongside contributors including Jennifer Huffstetler of Intel and Kim Greene of Georgia Power. The preview presented these organizations as participants in the event, not as endorsements or evidence that one supplier is superior.
Rank #2
What the 2024 agenda covered
From models to the edge
“From LLMs to Inferencing at the Edge” broadened the discussion beyond centralized training. Edge inference can shift compute closer to users, machines or sensors, changing requirements for site location, connectivity, physical security, serviceability and power availability. A portfolio may need both large training campuses and smaller distributed facilities.
Quantum and post-AI computing
Another session previewed quantum computing and developments after AI. The topic signaled a planning principle: facilities should preserve enough electrical, mechanical and architectural flexibility to accommodate technologies whose cooling, vibration, environmental or power requirements may differ from today’s accelerators.
Deployment at yottabyte scale
The agenda also examined AI deployment at yottabyte scale. That theme points to storage, data movement, networking, backup, governance and physical space—not just accelerator count. Capacity plans need to account for data ingest and egress, high-speed fabric expansion, cable pathways and the energy used to move and store data.
Enterprise strategy and spending
A panel considered generative AI’s effect on enterprise infrastructure strategy and spending. Wendy Schuchart, editor-in-chief of ITPro Today and Data Center Knowledge, said, “As technology adoption tends to lag behind capability, most enterprises are already behind when it comes to generative AI.” She said she wanted to hear how organizations would integrate generative AI at scale and how different industries were refining their processes.
Forecasts cited in the event preview
The following figures were reported in the 2024 article with the attributions shown. They are not independently rechecked current estimates.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →| Figure | Attribution in the 2024 preview | How to interpret it |
|---|---|---|
| Nearly 8% of all U.S. power by 2030, versus less than 2% at the time | Goldman Sachs | A U.S.-specific forecast cited in 2024; geography, methodology and subsequent revisions are not established here. |
| Up to 160% growth in data-center power demand by 2030 | Goldman Sachs | An upper-bound forecast quoted by the article, not a guaranteed outcome. |
| Almost 14% annual IT-load-capacity growth through 2030 | Alan Howard, Omdia analyst | A capacity-growth estimate attributed to Howard. |
| Almost half of data-center capacity used for AI by 2030 | Alan Howard, Omdia analyst | A forecast about the share of capacity, not a measurement of every facility. |
| About 45 GW of total demand during 2024–2026, conditional on procuring the power | Alan Howard, Omdia analyst | The condition is material: projected demand depends on obtaining sufficient electricity. |
Resource-use figures—and their limits
The preview also cited an EPRI estimate of 0.0029 kWh per ChatGPT question. It reported figures of 10 gigawatt-hours of electricity to train one large language model and 700,000 liters of freshwater, linking them respectively to the University of Washington and an arXiv paper. The article did not provide enough detail about model size, hardware, utilization, electricity mix, cooling system or accounting boundaries to generalize those numbers across models. They are best treated as illustrative evidence of resource intensity, not universal per-query or per-model constants.
Design criteria for an AI-ready facility
Yotta 2024 did not establish a winning product or architecture. Its agenda suggests a set of criteria owners and designers can use when comparing real options:
- Power: Confirm available utility capacity, interconnection dates, redundancy, power quality and a practical path to higher rack loads.
- Cooling: Match heat-rejection and distribution equipment to target density, with provisions for liquid cooling where air systems are insufficient.
- Efficiency and energy supply: Model efficiency at expected load, evaluate renewable procurement and account for water as well as electricity.
- Scalability: Build in phased expansion for electrical rooms, cooling plants, structures, network routes and commissioning teams.
- Workload flexibility: Plan for centralized training, edge inference, storage-heavy deployments and possible future compute types.
- Operations: Ensure controls, monitoring, maintenance access, spare capacity and staff capability keep pace with changing hardware.
Expo themes and companies named
The expo was organized around an Ampere AI Pavilion, a Liquid Cooling Coalition Pavilion and Innovate Arena categories covering rack-level technology; energy and critical power; cooling; AI-driven optimization and automation; and edge and cloud hosting. The preview also mentioned Vertiv, Digital Realty, Ampere, Iceotope Technologies, Shell Energy and Schneider Electric in event contexts. Their appearance indicates participation or thematic relevance only; it does not constitute a comparative test, endorsement or purchasing recommendation.
What the preview means for construction planning
For a construction team, the lasting lesson is coordination. Electrical design, mechanical systems, structural loading, water strategy, controls, network pathways and utility negotiations must be developed as one capacity plan. A facility sized only for today’s average IT load can become constrained by transformer lead times, cooling-plant space or permitting long before the server halls are full.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Yotta Events’ Sausner described the conference as “not just another conference” but a place to tackle issues that would define the next decade. In the documented 2024 agenda, those issues were concrete: secure power, remove heat, reduce resource intensity and preserve flexibility as workloads evolve.
Source: Drew Robb, “Industry Experts Look to the Future of Data Center Design at Yotta 2024,” Data Center Knowledge, September 26, 2024: https://www.datacenterknowledge.com/build-design/industry-experts-look-to-the-future-of-data-center-design-at-yotta-2024.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
