Jeremie Eliahou Ontiveros

Tracking hyperscalers, datacenters and energy infrastructure at SemiAnalysis. opinions my own

[Sriram Krishnan @sriramk]

interesting take from @SemiAnalysis_ that the data center moratoriums haven’t translated to actual megawatts lost ( yet)

https://t.co/keZBTevFTO

[Michele Tarawneh @micheletarawneh]

You’re mixing up things here…sub-5MW on existing interconnects is real and it's useful. But it's scavenging spare headroom, which is finite and thinning fast (+ will attract NIMBY^10). That fills the marginal demand. It isn't where the bulk of capacity comes from.

This isn't a training vs inference split. The inference-first buyers in front of us right now are asking for hundreds of MW, not 5. Serving a frontier model isn't one chip answering one request - weights are sharded across 100s of accelerators that have to behave as a single tightly-coupled system, with all-to-all traffic between them on every token. If you break that across 40 sites you don't get 40x the throughput. You get worse utilization, worse batching, no prefix cache reuse and a materially worse cost per token.

The companies chasing the lowest latency are the ones needing the densest, most tightly-coupled deployments. Not the most distributed ones.

And 5ms isn't the constraint anyway. Time to first token is hundreds of ms, generation runs for seconds. Nobody's bottleneck is 5ms of transit.

The demand we see is 300MW+, and almost nobody can deliver it on a timeline that matters.

Also - as a power guy, you’re ignoring the opportunity of co-locating renewables power generation + BESS to be financed by fully-priced PPAs…which is really why I dont’t get why you’re on this “small metro area DC” crusade in the first place.

[SemiAnalysis @SemiAnalysis_]

A decommissioned GPU's best use case is aura farming in the frame. With compute so cheap in the cloud, Jeremie takes a page from SpaceX's future trajectory.

That same logic, compute pricing shaping behavior, is exactly what drives SpaceX's 10 gigawatt buildout and the $300 billion ARR case behind it.

$100M/megawatt economics, SpaceX's 10GW bet, Microsoft's offtaker role, and more on this week's SemiAnalysis Weekly

[SemiAnalysis @SemiAnalysis_]

SpaceX 10GW in 2027 – Why It’s Real,
Will Drive $500B ARR for SpaceX,
and Why Microsoft Will Be the Largest Offtaker
Inference at 100B/GW/year, SpaceX's stellar pace,
Microsoft's 10GW 2026 Awakening,
Azure Can Grow Tiple-Digits

https://t.co/DJGWuKpQPE

[Daniel Romero @HyperTechInvest]

SemiAnalysis’s Head of Data Center and Energy Infrastructure Research:

“Anthropic, we estimate, had 1.5 gigawatts of capacity at the end of 2025. By 2027, they’re going to have 10, so that’s 9 gigawatts in two years.

That’s essentially the size of Google today, and you have four companies trying to do that within the same timeframe.”

[Shayle Kann @shaylekann]

Ask an AI expert to describe the strategy and competitive positioning of the individual hyperscalers, frontier labs and neoclouds with regard to chips and you'll get a detailed answer.

Ask them the same thing with regard to power and you'll get very little. This is a mistake.

So, to rectify the situation, I went deep with Jeremie Eliahou Ontiveros of @SemiAnalysis_ on @CatalystPod .

Watch: https://t.co/lVEJ8xGLhz

Listen: https://t.co/Hv4Wgn3rm3

[Jeremie Eliahou Ontiveros @JeremieEO]

Looking forward to this conversation!

Many thanks to @anissagardizy8 for the invite!

[Anissa Gardizy @anissagardizy8]

Excited to chat with this stellar crew about compute and infrastructure!

Wednesday, May 20. 11 a.m. PST.

RSVP: https://t.co/84LUm9kMOi @sk7037 @AnjneyMidha @JeremieEO

[Jeff Balkanski @balkanski]

First-try Gold, team smaller than a single hyperscaler standup - join us.

[Jeremie Eliahou Ontiveros @JeremieEO]

@institLPGP @SemiAnalysis_ We wrote a deep dive on 800VDC about 10 months ago for institutional clients. Flagged some key beneficiaries which are all up nicely since

The way Bloom benefits from 800VDC is much more nuanced!

[Rittenhouse Research @RHouseResearch]

Semi-Analysis's @JeremieEO on @tbpn last week:

"2025 was an insane year for the (data center) leasing market, and everything points to 2026, 2027 just accelerating."

Towards the end of 2025, Applied Digital $APLD, Cipher $CIFR, and Hut 8 $HUT signed the most attractive large scale (e.g. 200MW+) leases with investment grade hyperscaler counterparties to-date..

The incremental 830MW of power that Galaxy Digital $GLXY received final ERCOT approval on yesterday seems to have arrived at a fortuitous time..

[TBPN @tbpn]

Happy Friday. On today’s show:

- @bhorowitz (a16z)
- @arampell (a16z)
- @DavidGeorge83 (a16z)
- @jkhamehl (a16z)
- @JeremieEO (SemiAnalysis)

See you on the stream.

[Jeremie Eliahou Ontiveros @JeremieEO]

@CapArtichoke @dylan522p @fundaai Our first mention of GCP's TPU upside was August 7th. In early September we quantified revenue growth in 2026/27 and called out strong acceleration.
Our first AWS note quantifying growth / calling for Accel was Aug 5th. AMZN has well outperformed Meta, MSFT, ORCL since then.

[Jeremie Eliahou Ontiveros @JeremieEO]

@institLPGP @ShanuMathew93 @SemiAnalysis_ Yesterday's piece discusses near-term time-to-market solutions to power DCs when the grid can't keep up. In that context, solar is non-existent. Solar's role in energizing DCs is a separate topic.

[Jeremie Eliahou Ontiveros @JeremieEO]

@institLPGP @ShanuMathew93 @SemiAnalysis_ You are not understanding how Google and xAI's solar deployments work. They're completely different from the gas BTM deployments, and have a different purpose.

We'll cover that in a future piece. And we've explained it in detail to our Datacenter Model clients over a year ago.

[Jeremie Eliahou Ontiveros @JeremieEO]

@puffinancial @GridGodAjey @dylan522p @SemiAnalysis_ Thanks for the flag, we had a few typos. However Lazard's analaysis has a few issues. First, it's not applicable to an islanded datacenter project, it doesn't capture all the required overbuild.
Second, their capex estimates are simply way too low.

[Jordan Nanos @JordanNanos]

ClusterMAX 2.0 is here!

In this update we rate 84 neoclouds, up from 26
Our market view tracks 209 providers, up from 169 in March and 124 last October
We spoke to over 140 end users as part of the research
And we wrote over 43,000 words about our experience

This is more than Animal Farm, but less than The Great Gatsby

More thoughts in 🧵

[Jeremie Eliahou Ontiveros @JeremieEO]

@RichardMeyerDC @ShanuMathew93 Depends on the type of chiller, but agree 1,000 for 1gw is a bit too high. Chiller specs are public, it's not hard to figure this out. You generally see ~1.5mw per air-cooled chiller and much more for water-cooled units.

We explained everything here: https://t.co/Dufp9t2ful

[Aishwarya Mahesh @ash_locked]

Too often, AI and energy conversations happen in separate rooms. That gap inspired @themeghamachine and I to launch the Watt & Why Series ⚡

Excited to kick this off and big thanks to @CrusoeAI and @SemiAnalysis_ for co-hosting the first session.

A space that brings everyone from AI researchers to energy traders together for the conversations that will shape our future!

[Jeremie Eliahou Ontiveros @JeremieEO]

Amazon and Anthropic clearly aren't falling behind on GenAI :)

[SemiAnalysis @SemiAnalysis_]

Amazon’s AI Resurgence:
AWS & Anthropic’s Multi-Gigawatt Trainium Expansion
Anthropic multi-gigawatt clusters, Trainium ramp,
best TCO per memory bandwidth,
system-level roadmap,
Bedrock and internal models

https://t.co/GzZVZHip0V

[Jeremie Eliahou Ontiveros @JeremieEO]

Thanks for the invite @AndrewRangeley !! Looking forward to our next convo!!

[Yet Another Value Empire @YetAnotherValue]

Podcast #331 is up!

@JeremieEO returns to talk all things data centers / power

https://t.co/vDUsgDdjhI