Founder and CTO: @ComputeDesk. Exited Pace to FLYR. Formerly VP Research at Winton, particle physics at Harvard and CERN (Higgs discovery). MIT/Caltech alumnus

London, England
It's only a matter of time till neoclouds introduce dynamic pricing. If you are running at full capacity, by definition, you are pricing too low and leaving money on the table. As margins compress, that money will become a critical part of your revenue optimization.
Today we're introducing spot pricing for preemptible VMs on Nebius. From October 8 the price will be calculated dynamically from available capacity and demand, per GPU type and region, instead of a discount we set. Cap what you pay, or follow the price. Settings are live today.
2
11
1,090
David López Mateos retweeted
Big
Thanks @Google for having @ComputeDesk × @NodalExchange. Hyperscalers can soon trade and hedge compute, stay tuned.
1
16
996
“We track over 1000 different oil prices” @neilbradford . Who still thinks oil is a homogeneous commodity?
1
8
2,878
Vote, should I write a follow up blog post about how and when the free option disappeared? Also, congrats to all of those who actually took the free option! We won
Compute Desk CTO @SenorScience presciently wrote that "B300s at B200 prices is a free option" at the end of April. The market hadn't priced in the fact that KimiK3 would come out, uniquely suited to work on B300s. We could have sold you an option to hedge that. "If MoE keeps scaling, toward multi-trillion-parameter models with longer context, the per-card headroom matters even more". Here's the original article: computedesk.substack.com/p/b…
4
713
David López Mateos retweeted
Thanks for covering @ComputeDesk. Missing from analysis is how @Architect_Fi 's ComputeConnect, built on Compute Clear bridges price hedges and capacity hedges though an Exchange-for-Physical mechanism. Otherwise a good primer !
1
2
36
2,572
Do you want to be knowledgeable like @BrettHarrison ? Get the data
B300 prices have broken away from the pack, one-year contract rates up >$1/hour in the last 30 days while other models are flat. The compute capacity block trades our customers have brought Architect this month are almost all B300/GB300 or VR. Few are asking for Hoppers at size.
1
7
785
David López Mateos retweeted
Thanks @Google for having @ComputeDesk × @NodalExchange. Hyperscalers can soon trade and hedge compute, stay tuned.
1
6
38
2,650
1/ A common way to build a GPU forward curve: take the 6-month reserved rate and the 12-month rate, subtract the six months already priced, call the remainder the price for months 7 to 12. That subtraction doesn't work.
4
3
25
2,850
3/ The subtraction loads the extra discount onto the marginal months. Each extra point of discount on the 12-month contract becomes two points of apparent decline in months 7 to 12. From 2-year to 3-year contracts the factor is three. B200 year-3 delivery comes out at 21% of today's rate.
1
2
206
4/ Compute cannot be stored, so nothing ties the forward to spot. The curve has to be estimated with the discount separated out. Ours puts B200 forwards a year out at or above spot; most of the reserved-rate discount is the value of committing. Method write-up coming.
6
168
David López Mateos retweeted
Picture this. It's 1960. "I'm a skeptic of oil indices and oil futures. We haven't seen oil trade like a commodity yet, and I'd hate to see these go the way of onion futures of the 1950s. The indices are meaningfully off of actual clearing prices because 7 sisters and army purchases are highly confidential deal terms. Fungibility is great in theory not in practice. Any refinery / scale user knows the value of a barrel of oil has too many dimensions to reach fungibility from barrel to barrel: specific API gravity, geo, sulfur content, the 10 different operators in the supply chain, and a thousand+ more chemical variations. If it works, we'll surely use it but until then we're building something else." Messing around with our Good(e) friends at Shadeform ! @edsgoode message me your office address for a copy of "The World for Sale" if you haven't read it already
I'm a skeptic of GPU indices and compute futures. We haven't seen digital infra trade like a commodity yet, and I'd hate to see these go the way of bandwidth futures in the 90s. The indices are meaningfully off of actual clearing prices because hyperscaler and lab purchases are highly confidential deal terms. Fungibility is great in theory not in practice. Any hedge fund / scale user knows the value of a cluster has too many dimensions to reach fungibility from cluster to cluster: specific fiber paths, shell, geo, BOM, the 10 different operators in the supply chain, and a thousand+ more variables. If it works, we'll surely use it but until then we're building something else
7
7
44
17,708
David López Mateos retweeted
"Nvidia compute is fungible". - @JensenHuang
1
1
17
1,094
David López Mateos retweeted
Nodal Exchange is seeking to launch futures contracts on computing power for artificial intelligence chips, joining a host of firms rushing to do the same. bloomberg.com/news/articles/…
5
14
28,690
David López Mateos retweeted
💻 Compute futures are coming to Nodal Exchange. Nodal's huge array of power contracts along with superior risk margin protocol provide an edge that allow users to compute globally while hedging locally. “We chose to partner with Nodal, for their world-leading coverage of power and electricity, an obvious nexus to compute.” @andrawesbahou
Replying to @NodalExchange
@NodalExchange and @ComputeDesk today announced plans to launch a new suite of GPU (Graphics Processing Unit) compute futures contracts on Nodal Exchange. Read more: nodalexchange.com/nodal-exch…
2
3
13
1,249
Compute is becoming one of the largest asset classes in the world and it still trades on handshakes and spot quotes. Power futures and the compute basis trade are a big part of what changes that. Nodal already clears the power markets data centers run on. Right venue, right time.
Compute Desk is partnering with @NodalExchange to launch compute futures settled against our compute indexes. With the compute world’s nexus to electricity and power, Nodal is an exceptional venue for AI infrastructure companies, hyperscalers and financial institutions looking to hedge or gain exposure to compute, power or the spread between them. Reach out to @ComputeDesk to trade or hedge compute. H/T @ForgashEmily Bloomberg: bloomberg.com/news/articles/…
1
11
888
David López Mateos retweeted
Are you financing GPU infrastructure and looking for a backstop ? Get in touch to explore how we can structure one for you.
Nvidia’s GPU backstop deals are equivalent to writing put options at the residual floor and buying a smaller notional of call options at the same strike. If Nvidia could sell GPU rental price forwards or futures they could hedge the partial synthetic long.
2
8
2,123
Love working with the best.
The only place in the world an undergrad can research, design, and tapeout a full ASIC is ETH Zurich. It's an arduous process, and only the best make it through, but they'll fund the full foundry tapeout $$, which is insanity. Here's a binarized ConvNet accelerator (Tsukiji) I taped out 10 years ago. H/t to @LucaBeniniZhFe and @ETH_en for unambiguously the best place in the planet to for end-to-end ASIC design research.
8
1,000
This is what tradable indices look like. No sudden unexplained jumps, impossible to manipulate. Built to IOSCO standards. And yes, the supply crunch is still here.
Even with ~13 GW of new US power coming online for compute this year and more efficient models, the demand is still outpacing supply by a large margin. Every one of our indices is up ~30%
1
2
16
2,284