Jun 13, 2024 · 43m · catalyst
Under the hood of data center power demand
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of Catalyst, host Shail Khan and former Microsoft energy executive Brian Janus examine how the generative AI boom is straining electrical grid infrastructure and reshaping data center siting. They discuss utility queue bottlenecks, behind-the-meter microgrid flexibility, and the technical strategies required to balance rapid computing growth with climate decarbonization goals.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Shayle holds 33.8% of the talking time here. How this is scored →
Shayle as informed peer 5.6 Guest teaching 5.7 Guest disagreement 1.4 Shayle pushing back 2.0
speaking balance: gold is Shayle, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 27:25 Janus dismantles the off-grid data center concept
Janus bluntly dismisses the popular narrative that data centers can bypass utilities by going off-grid, pointing out it merely shifts dependency to an equally constrained gas grid.
Hardest push from Shayle ▶ 23:20 Kann presses on peak-shaving training workloadsKann directly challenges Janus's assumption that AI training loads cannot participate in demand response or peak shedding given their flexible batch nature.
Biggest teaching moment ▶ 30:45 Dublin dispatchability regulatory solutionJanus explains how he helped craft a policy in Ireland requiring data centers to offer on-site dispatchability in exchange for grid connection to avoid outright moratoria.
Shayle holds their own ▶ 32:30 Kann questions tariff equity and ratepayer impactsKann demonstrates sharp market acumen by drilling into cost socialization, asking how utility capital allocation for hyperscalers impacts consumer rates and clean power deployment.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Shayle as informed peer | Guest teaching | Guest disagreement | Shayle pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome and Catalyst Swag Referral Promotion | 0 | 0 | 0 | 0 | Host monologue covering housekeeping, referral swag promotions, and introducing the premise of data center electricity demand without guest interaction. | |
| The ChatGPT Inflection Point and Scaling Pressures | 6 | 5 | 1 | 2 | Kann probes the early days of AI scaling at Microsoft and notes existing regional latency constraints. Janus details the rapid transition from megawatt increments to gigawatt denominators after ChatGPT's rollout. | |
| Cloud Architecture and Single-Site Gigawatt Campuses | 5 | 7 | 1 | 1 | Janus educates Kann on what constitutes a cloud region and explains why AI training models require massive, low-latency, single-campus gigawatt sites rather than multi-region distribution. | |
| Dissecting Utility Queues and Speculative Demand | 7 | 6 | 2 | 3 | Kann questions the legitimacy of massive utility queue announcements like AEP's 80-90 GW backlog. Janus contextualizes zombie requests versus real demand, highlighting AEP's 765 kV transmission infrastructure. | |
| Historic Resiliency Needs and Microgrid Foundations | 5 | 7 | 2 | 1 | Janus explains why high-availability requirements necessitate redundant backup systems, noting how historic diesel standards at distribution levels have persisted even with direct high-voltage interconnects. | |
| Mid-Roll Sponsor Messages: Bloom Energy and Engie | 6 | 5 | 2 | 4 | After mid-roll sponsors, Kann pushes on whether training workloads could curtail during system peaks. Janus explains that high capital expenditure on GPUs demands high utilization, making intermittent operation economically unviable. | |
| Creative Power Solutions: Microgrids and Grid Enhancing Technologies | 6 | 8 | 2 | 2 | Janus dismisses the viability of fully off-grid data centers as moving power problems to the gas grid, outlining practical behind-the-meter microgrid options and grid-enhancing technologies, citing his Dublin dispatchability precedent. | |
| Utility Alignment, Tariffs, and Regulatory Innovation | 7 | 6 | 1 | 3 | Kann asks about cost-allocation fairness and ratepayer protections when utilities invest for hyperscalers. Janus articulates utility economic motivations and the regulatory evolution needed for dispatchable tariffs. | |
| Reconciling Decarbonization Pledges with Load Surges | 7 | 6 | 1 | 2 | Kann and Janus discuss corporate climate pledges clashing with real load growth, acknowledging that 2030 corporate targets are slipping due to unforeseen grid bottlenecks and rapid AI adoption. | |
| Siting Priorities and the Fallacy of Efficiency Savings | 7 | 7 | 2 | 2 | Kann characterizes power access as overtaking all other siting criteria. Janus reinforces this with Jevons paradox, explaining that higher chip efficiency leads to denser computing rather than reduced power demand. |
Statements from this episode (18)
EPRI Report: Data Centers Could Consume 10% of US Electricity by 2030
“There are breathless reports coming out almost daily at this point claiming data centers could consume up to maybe 10% of U.S. Electricity by 2030. That was the high end of a recent EPRI report.”
Janus: AI technology is advancing faster than electric utilities can build power
“The problem we were going to have was going to be whether we would have enough power to support this technology that was moving at a pace that, as you know, moves way faster than the electric utility industry.”
Janus: Microsoft's energy demand grew from hundreds of megawatts to gigawatts
“Over that decade, we had gone from a baseline of, you know, a couple hundred megawatts, and we were growing at a relatively rapid pace, but that's also kind of a small denominator. So the incremental tranches every year that we had to go procure were measured …”
Janus: AI training latency constraints require single-site gigawatt data centers
“Because those training models are in and of themselves a big machine, and so you can't, and this is not my forte and my area of expertise around the actual architecture inside of a training model, but my understanding of sort of the constraints there is that y…”
Janus: AEP will not connect 90 gigawatts in next decade
“There's no reason to believe that AEP is going to connect 90 gigawatts in the next 10 years.”
Janus: AEP runs rare 765 kV transmission system attractive to hyperscalers
“They operate one of the highest voltage Systems in the country. So they operate a seven 65 KV system which is, you know, higher than what anyone else has. There's a little bit of seven 65 in New York as well.”
Janous: AEP's data center demand forecast surged 275% in five months
“If you go even back to January of this year, PJM had Roughly four gigawatts of anticipated data center demand for AEP by 2030, and by May, AEP revised that up to closer to 15, in just five months, right?”
Janus: Cloud Apps Cannot Seamlessly Move Regions During Data Center Outages
“I think people have this sense of the cloud as really being this sort of ethereal thing, and applications can just move around, and if one data center goes down, then, well, you can just move everything to another data center. That's not really how it works. W…”
Janus: Standard 48-Hour Diesel Backup Is Probably a Data Center Relic
“I will say that backup is probably a bit of a relic because when we first started building these data centers, 15 years ago, most of them were relatively small. They were five or 10 megawatts. They were connected at distribution voltage. So the standard was pu…”
Janus: AI Training Loads Will Not Run as Intermittent Renewable-Following Workloads
“So I think it's a little bit overblown to say, and I think this is also true of Something like the conversation around crypto, but that these are highly curtailable loads that you could just, you know, attach a training model to a wind farm and only run it, yo…”
Janus: Powering Gigawatt-Scale Data Centers with Solar and Storage Is Not Feasible
“You're not going to put solar in storage to supply a 500 to gigawatt, 500 megawatt to gigawatt scale data center. I mean, it's just not feasible.”
Janus: Nuclear Power Will Not Solve Data Center Bottlenecks This Decade
“The focus that these companies have right now is getting data centers online in the next, Two, three, four years, and nuclear is not going to solve that problem.”
Janus: Data Centers Account for Nearly 20% of Ireland's Electricity
“It's almost 20% of the entire country of Ireland's electricity are data centers, and they're almost all in Dublin.”
Janus: Every New Dublin Data Center Must Install On-Site Gas Generation
“So now every new data center in Dublin is getting a natural gas generator behind the meter so it can be flexible and, you know, avoid that contribution to system peak.”
Microsoft disclosed overall emissions rose 30% since setting climate commitment
“Microsoft announced that despite all their best efforts, they're Overall emissions have gone up 30% since making initial commitment.”
Janus: Hyperscaler climate targets are probably not achievable on time
“And so there is, I think at this point, a point of coming to grips with the fact that those targets are probably not achievable in the time horizons that they had all set.”
Janus: Hyperscalers have deprioritized day-one clean power for immediate grid capacity
“You know, if you go back a few years, you know, zero carbon energy day one was really becoming a non-negotiable that, you know, we wanted energy, wanted to be zero carbon. We wanted to have it Done, ready to go when we plug in the data center. I think the, pro…”
Janus: Chip efficiency gains will not reduce AI data center power demand
“If let's say Meta, for instance, finds a piece of land or a point on the grid where they can pull three gigawatts, Of power on one location. And then the next day, NVIDIA comes out with a chip that's twice as efficient. Is Meta going to build a 1.5 gigawatt da…”