joseph retweeted
Forward GPU capacity gets priced in the open on December 2.
During @NVIDIAGTC, Compute Exchange will run a virtual two-sided auction for two standardized contracts: NVIDIA B200 and B300. Each is one eight-GPU bare-metal node in a U.S. data center, on a 12-month term starting April 1, 2027.
How it runs:
• Bids and offers are submitted online, 9:00 a.m. to 3:00 p.m. ET
• Matched contracts clear at a single price
• Individual bids and offers stay confidential
Most GPU capacity still trades in privately negotiated deals, each on its own terms, which makes prices hard to compare. Common contract specs and one clearing price give buyers and providers a market reference for April 2027 capacity.
Participation is open to eligible U.S. organizations, subject to verification.
We are holding two info sessions on the auction on Monday, November 23:
• 11:00 a.m. PST (2:00 p.m. ET): us06web.zoom.us/webinar/regi…
• 1:00 p.m. PST (4:00 p.m. ET): us06web.zoom.us/webinar/regi…
Contract specifications and registration: compute.exchange/gtc-dc/auct…
TomorrowX is heating up so fast. My calendar is a complete nightmare right now. Spending the next 10 hours on my HK → Dallas → Austin trip sorting it out… the best way to catch me is at our Compute Party!!
Are you coming to @TomorrowXSummit? Come to our Compute House block party!
luma.com/vf8dn8du
joseph retweeted
We invented a recursive megakernel generator called Hyperkernel, the most complete solution to megakernels and GPU efficiency.
Hyperkernel-generated megakernels achieve state of the art performance on 11 different domains: inference (DeepSeek, Kimi and GLM), pretraining (Qwen), hardware fault recovery, robotics (π₀-FAST, ACT) diffusion and real time world models (Matrix-Game), full self driving (OpenPilot), voice models (Moshi), genomics (Evo 2) and protein folding (ESMFold).
All of this in a single kernel launch.
For context, Hyperkernel generated fully optimized megakernels for robotics, genomics and voice in under 50 minutes, and doubled the efficiency for >5 models generally using only about 60-100 lines of code per megakernel. This speaks to the unparalleled standard for code quality in the automated generations.
Our first release today open sources the megakernels themselves. Over the next couple weeks, we will release the entire Hyperkernel stack, along with the complete serving stack for Kimi and GLM (20-30% inference efficiency gains across all concurrencies).
Out of the box GPU machinery performs terribly for unorthodox research. We built Hyperkernel only 20 days ago for internal use; but given its multi-modality, we have decided to open source it.
Creative research should not be bottlenecked by the lack of dedicated infra teams. Hyperkernels offer a one stop shop infra stack for all research. We would love for the open source community to add to it :)
GitHub repo: github.com/Hyperstitioncc/Hy…
NOTE: We primarily use NVIDIA B300s, but our kernel generation extends to other hardware (B200s, AMD, TPUs).
1/9
We absolutely crush NVIDIA and achieve the best low latency inference for DeepSeek v4.1 Flash.
Compared with NVIDIA’s published vLLM results, Hyperkernel delivers:
• 36.7% higher per-user generation speed at concurrency 1.
• 28.2% higher per-user speed while serving two users, versus NVIDIA serving one.
• 15% higher per user speed while serving four users, versus NVIDIA serving one.
• 90.6% higher per-user generation speed at concurrency 4, compared with the highest published C4 AgentX interactivity for TP4 B300s.
• 28.5% higher per-user generation speed at concurrency 8, compared with NVIDIA's published result.
• 36% faster median per-request decode speed.
Cool!
This is my conversation with Michael Kagan, the founder of Mellanox and current CTO of NVIDIA
We speak about his life, the progression of networking over the span of his career, and how the work that NVIDIA is now doing to build "Planet scale" computing
I hope you enjoy the conversation as much as I did
00:00 Highlights
01:16 Growing Up in Russia and Moving to Israel
03:42 Intel, the PC Revolution, and a Passion for Technology
06:14 Sponsor: Liquid
06:33 Lessons from Intel: Software and Networking
10:10 Founding Mellanox and the Vision for Cloud Computing
14:34 From Supercomputers to the Cloud
17:46 How GPUs Sparked the AI Revolution
20:52 Sponsor: Compute Exchange
21:14 NVLink and the Rise of Rack-Scale Computing
24:57 The Engineering Challenges Behind NVL72
27:39 Designing Hardware for Future AI Workloads
29:43 BlueField DPUs and the AI Factory
32:55 Making AI Agents Safer
35:37 What Mellanox Could Have Done Differently
37:03 NVLink, InfiniBand, and Spectrum-X
39:24 Connecting AI Factories at Planetary Scale
42:42 Keeping Massive AI Systems Running Through Failures
1
103
joseph retweeted
This is my conversation with Michael Kagan, the founder of Mellanox and current CTO of NVIDIA
We speak about his life, the progression of networking over the span of his career, and how the work that NVIDIA is now doing to build "Planet scale" computing
I hope you enjoy the conversation as much as I did
00:00 Highlights
01:16 Growing Up in Russia and Moving to Israel
03:42 Intel, the PC Revolution, and a Passion for Technology
06:14 Sponsor: Liquid
06:33 Lessons from Intel: Software and Networking
10:10 Founding Mellanox and the Vision for Cloud Computing
14:34 From Supercomputers to the Cloud
17:46 How GPUs Sparked the AI Revolution
20:52 Sponsor: Compute Exchange
21:14 NVLink and the Rise of Rack-Scale Computing
24:57 The Engineering Challenges Behind NVL72
27:39 Designing Hardware for Future AI Workloads
29:43 BlueField DPUs and the AI Factory
32:55 Making AI Agents Safer
35:37 What Mellanox Could Have Done Differently
37:03 NVLink, InfiniBand, and Spectrum-X
39:24 Connecting AI Factories at Planetary Scale
42:42 Keeping Massive AI Systems Running Through Failures
joseph retweeted
Come join us for Cocktails & Compute in SF on Oct 28th!
luma.com/6h2xhmyg
Excited to announce our partnership with @nebiusai at @NVIDIAGTC Berlin!
I love seeing infrastructure providers like Nebius embrace transparency. This is how we move compute toward a more open, data-driven marketplace.
Check out @stevehou blog on our initial analysis: lnkd.in/gmMUPV_5
#SiliconData #Nebius #GTCBerlin #AIInfrastructure #ComputeEconomy
@dee_bosa
🚨New: Our Head of Research @stevehou breaks down the implications of @nebiusai’s new dynamic pricing for spot/preemptible GPU compute.
silicondata.com/blog/gpu-spo…
joseph retweeted
🚨New: Our Head of Research @stevehou breaks down the implications of @nebiusai’s new dynamic pricing for spot/preemptible GPU compute.
silicondata.com/blog/gpu-spo…
Dynamic pricing is cool 😎. It’s good for market efficiency, promotes price discovery, increases liquidity, and lowers costs for all. Kudos to @nebiusai for starting this and hope more will join soon! 😊
🚨New: Our Head of Research @stevehou breaks down the implications of @nebiusai’s new dynamic pricing for spot/preemptible GPU compute.
silicondata.com/blog/gpu-spo…
joseph retweeted
Our CEO @Carmenli was on the cover of WSJ Finance today
Only index provider at NVIDIA’s GTC analyst Q&A, so of course I asked @JensenHuang about GPU residual values. 😆
His take: newer is better, but older GPUs matter when everything is sold out.
He knows exactly what we do. Then he asked: “How did you get your data, and how can we help?”
Surreal. 💚 Also love running into Alison!
Excited to announce our collaboration with @nebiusai alongside Marc Boroditsky.
Tune in to our live conversation with @dee_bosa at 10:10 a.m. Berlin time.
Join us live from Nebius House 📹
nitter.cf/i/broadcasts/1DGLdZnpg…
Thank you @GavinSBaker for citing our data. We just announced live on @BloombergTV from @NVIDIAGTC Berlin that we’re excited to expand our partnership with @CMEGroup and advance our joint product roadmap, with tradable products based on @silicondata’s LLM Token Indices planned for the near future.
An exciting step toward bringing greater transparency, price discovery, and financial innovation to the AI compute economy!
Gavin Baker says token prices are falling, but the price of renting a GPU hour “has gone vertical,” because each GPU hour now produces more useful tokens
BAKER: “By the way, on Jevons Paradox, it is interesting, and I think it’s very important to be precise in our terms. What we have seen, Silicon Data, an Atreides and I believe NVIDIA portfolio company, has this token index. It captures price and mix."
"And as the price of tokens has gone down and the mix has shifted to cheaper tokens. The cost to rent an hour of GPU has gone vertical."
"And that is because as you’re getting more tokens that are more useful out of a given GPU hour, that GPU hour becomes more valuable. So we’ve actually seen as the price of tokens has gone down, the prices that you were charging companies per your last call has gone vertical. You raised prices how many times in the last three months?”
VOLOZH: “Ugh, nobody counts. Now we raise prices every 12 minutes on the auctions. Not only raising, by the way. Sometimes they go down.”
BAKER: “But token, we need to be precise in our terms. Price of tokens going down, shifting to cheaper tokens, does not mean the price of renting a GPU from Arkady is going down.”
Live from @NVIDIAGTC Berlin. Thank you for the much needed vest @gcore_official
joseph retweeted
The era of compute is upon us. Compute lending. Compute trading. But what does it mean?
In this beautifully illustrated article in @arenamag, I sketch out some early players who are trying to make a world of commoditized compute beyond the clouds
It’s also about coffee ☕️
"The man who does not use computation cannot compete with the man who avails himself of its secrets. No hedge fund can price second order effects faster than a 200-megawatt Jane Street data center. No lone genius can produce a better language model than Anthropic."
@CooperVeit3 on the secretive, unborn market for compute:
arenamag.com/articles/counti…
If you’re going to Berlin, reach out! #gtc
Heading to Berlin next week for @NVIDIA GTC!
Excited to join the Capital Providers Panel on October 22, 3:45–4:30 PM, to discuss GPU financing, compute markets, and how pricing transparency is shaping the future of AI infrastructure.
See you in Berlin! Pin @josephcompute if you want to meet up for beers.
2
299
joseph retweeted
What a turnout at the Compute Trading Happy Hour for #SFTechWeek. 850+ people requested a spot and the Chotto Matte rooftop was packed. Thanks to everyone who came, and to @silicondata and @robotventures for hosting with us.
joseph retweeted
Replying to @stevehou
@stevehou representing @silicondata at the NeoCloud summit on a panel today 🔥
joseph retweeted
Come join us :)
We’re looking for an exceptional Product Manager for @computeexchange
Salary is $150-$230k depending on experience level / background. Comes with equity
Preference to be based in the Bay Area or NYC
To apply DM me or @computeexchange your background. Let us know why you’d be a good fit working with Compute Exchange, and share your CV/Linkedin


























