Prediction Open AI assessment confidence: 85% certainty 4/5 debate potential 3/5

Belizaire: Batch computing will quickly consume more energy than real-time compute

John Belizaire · Energy is now the ‘primary bottleneck’ for AI · May 21, 2024 · at 35:27

John Belizaire, CEO of Soluna, discusses the growth trajectory of batchable AI workloads compared to real-time computing services.

0:00 / 0:17exact quote · 17.1s

720p mp4 · rendered on demand · StarZero watermark

“If you look at the whole pie of the types of compute that's out there the batchable portion is growing faster than the rest, and it's become, it will quickly become a larger portion of the over, overall pie, and it will quickly have more energy demands than the rest.”

quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →

The claim, plainly

Batchable computing workloads will quickly become a larger share of total compute and consume more energy than real-time computing.

a quote alone can be ambiguous, so this restates it with the references and context of the conversation around it resolved. AI-written; the tape below the player is the source.

Assessment

On May 21, 2024, John Belizaire stated that batchable computing workloads would quickly become a larger share of total compute and consume more energy than real-time computing. As of September 2026, industry data from Gartner, the IEA, and EPRI indicate that batchable compute has not surpassed real-time/interactive compute: conventional always-on servers still represented 69% of data center power in 2026, and within AI workloads, real-time inference has grown to represent 60% to 80% of compute demand compared to batch training. However, because the prediction has an open/default horizon (set to May 21, 2029), the claim remains open and cannot be definitively resolved as true or false yet.

Gartner reported on June 10, 2026, that conventional interactive/real-time servers represented 69% of total data center power consumption in 2026, compared to 31% for AI-optimized servers.
IEA and McKinsey analyses from April–August 2026 show AI inference (predominantly real-time and latency-sensitive) accounted for 60% to 80% of all AI compute, with training (batch) remaining a minority.
EPRI studies (November 2025–February 2026) confirm that the vast majority of current data center IT power draw remains continuous and inflexible rather than batchable.

Checked against

[1] aimultiple.com [2] introl.com [3] digitaledgedc.com [4] ironmountain.com [5] civo.com [6] gartner.com [7] iea.org [8] europa.eu

batchable workloads data center energy demand response EPRI IEA report batchable compute vs real time data center workloads energy share Gartner IDC EPRI IEA batch computing vs interactive real time data center energy consumption share AI training vs inference compute share energy Gartner IDC EPRI IEA data center "batchable" "real-time" data center workloads percentage

assessed as of 2026-09-02

This assessment is an AI reading of a public statement, made against the cited references with the confidence shown. It can be wrong. The quote, the tape and the references are all on this page; if something is off, use report an error below. full disclaimer →

⏳ next review: 2029-05-21; checked again automatically on the next resolution wave

Mentioned

batchable portion compute

⚑ Spot an error? Report it

More from John Belizaire

Opinion

Belizaire: Data centers should leave population centers and go behind-the-meter

“Maybe data centers shouldn't be at population centers anymore. They should all be behind the meter and be dispatchable processes”

John Belizaire May 21, 2024 ▶ 52:47 Energy is now the ‘primary bottleneck’ for AI

Assertion Supported

Belizaire: AI servers require ten times the power density of traditional compute

“When you look at today's data center design it's designed for traditional compute processes and servers that use a 10th of the power that this new technology uses. It uses 10 times the amount of power 10 times the amount of power density is in a particular fac…”

John Belizaire May 21, 2024 ▶ 24:51 Energy is now the ‘primary bottleneck’ for AI

Assertion Supported

Belizaire: Hyperscalers are designing data centers dedicated exclusively to AI

“Some of the big hyperscalers are working on all their own designs. And these facilities will only be doing AI. It's at that level where the entire building and facility would just be AI processing.”

John Belizaire May 21, 2024 ▶ 25:15 Energy is now the ‘primary bottleneck’ for AI

Assertion Supported

Belizaire: 300 TWh of wasted energy is available globally each year

“We think there's about 300 terawatt hours of this wasted energy on a global basis available every single year.”

John Belizaire May 21, 2024 ▶ 34:50 Energy is now the ‘primary bottleneck’ for AI

Belizaire: Batch computing will quickly consume more energy than real-time compute

More from John Belizaire

Turn any episode into a week of clips.