Arushi Sharma Frank
@ArushiSF
Followers
984
Following
1K
Media
233
Statuses
2K
Personal acct.Pvt Equity. CSIS. EmeraldAI. Big Batteries. Data Center Power & Grid Expert. Ex - TSLA, Constellation,EPSA,AGA,Tax Fdn. Lives on 🌍. STEM Lawyer.
Earth
Joined January 2024
@EmeraldAi_ @EPRINews #DCFlex.Packed day at Summer Seminar. Glad to be here w/ @vsiv. Super stoked to see APS SRP demos' flagship results shared with the US and multinational #datacenter investment ecosystem.
1
0
5
#DCFLex @EPRINews @EmeraldAi_ @nvidianewsroom Nvidia, all the hyperscalers, multiple cap table participants, and everybody talking about flex results from multiple demos. Very cool.
0
0
1
A good report or two are out there on what we need long-term. This article is about what we could really use tomorrow/yesterday.
1
1
1
AI workloads, pumped-up electrification targets. ERCOT, PJM, CAISO—they’re all buckling under queue congestion, load clustering, & 20th-century planning apparatus. That’s quite the problem statement. The answer, we posit, is Gas + X.
5
2
7
"While not meaningless, historical utility dd forecasts consistently .overestimate both peak and average demand. Rigorous models of data center energy use are needed to adequately calibrate both future investment and public .concern."
VERY important piece @tylerhnorris!. Data centers may not run at 90%+ capacity 24/7 as widely believed. Key confusion: load factor ≠utilization rate. Maintenance, hardware failures, workload variation, & thermal limits mean lower usage, but utilities plan around peal demand.
1
1
10
Utilities consistently emphasize data centers- specifically AI as drivers of projected .increases in electricity dd. In parallel, specialized microelectronics analysts made more reserved estimates of short- and medium-term load growth due to greater emphasis on .manufacturing.
0
0
0
"@Microsoft challenged Georgia Power’s modeling methods, highlighting concerns .that the model is undervaluing renewable energy while overestimating data center load for projects that have not yet selected Georgia Power as their provider (Allsup 2024)."
0
0
0
Tech and reg - are working against each other w/ a grid connect planning framework which is risk-averse, static. It's largely a behavioral incentives problem across two sectors which don't quite chat with each other at the level they should.
csis.org
Power utilities still plan for worst-case scenarios in both grid forecasting and interconnection studies. But AI data centers do not have to be built—or operated—in ways that trigger those worst-case...
2
2
2
Contrary to the assumption of constant high utilization, this strategy reflects real-world variability, challenging the narrative of data centers as always-on loads, supported by utility reports citing actual load factors of 80-85% versus planned 90%. @tylerhnorris @ShanuMathew93.
1
0
3
This approach drives regional network upgrades rather than site-specific ones, aligning with findings from the 2024 United States Data Center Energy Usage Report (LBNL-2001637), which notes data centers often operate below 90% capacity due to maintenance and thermal limits.
2
1
2
Utilities plan for loads to have firm demand - all the time. They assess load contribution to system peak and apply the tail risk analysis for all hours as firm dd. That goes into area network upgrade analysis. The upgrades are regionally planned this way, not just for one site.
VERY important piece @tylerhnorris!. Data centers may not run at 90%+ capacity 24/7 as widely believed. Key confusion: load factor ≠utilization rate. Maintenance, hardware failures, workload variation, & thermal limits mean lower usage, but utilities plan around peal demand.
2
2
17
RT @duncan__c: On new years day of 2023 @DER_Task_Force released a pod titled "The Clean Energy Abundance Platform". In it we broke down h….
0
5
0
I hear this from friends in other countries too. Most recently, in Buenos Aires.
Spoke to a Chinese friend today who was recently home in a Chinese major city for a few weeks. He said al of his late 20s friends are bummed out and don’t think their futures will be good. 9-9-6 is real but the work is performative to keep your boss happy. All bullshit. Tons.
0
0
0
0
1
0
There is also a GONG. I have whacked it. @basepowerco .Kudos to those lucky enough to attend their IEEE afterparty/Salon at HQ. The lab setup and the house demo are also 🔥.
We track tons and tons of metrics. This is the number you see when you first walk in the front door of the office @basepowerco . This one - and its rate of change - matters the most.
2
0
10
Capacity shortfalls in West London, data centers competing with new industrial parks. Can't validate accuracy, but can say the post is worth a quick look.
linkedin.com
It’s already hard enough to get a grid connection in London. Now it’s about to become nearly impossible. Are you feeling the pinch? London is now second only to California as the world’s largest data...
0
0
0
Schedule AI like generation. 💥🤯.👇🏽.
linkedin.com
When AI Loads Don’t Talk to Each Other, The Grid Listens Anyway What happens when three data centres, each training advanced AI models, ramp their power consumption independently: • No operator...
0
0
0