Pinned Tweet
AI is great at hitting explicit goals, but often at the cost of the hidden ones.
Terence Tao just wrote about this. He points out: AI is the ultimate executor of Goodhart’s law, i.e. when a measure becomes the target, it stops measuring what we care about.
Take a call center. “shorten average call time.” Sounds reasonable: shorter calls should mean faster resolutions, happier customers.
At first, it works. Agents become more efficient. But soon, people start gaming it: nudging customers to hang up when the problem is tricky, or just dropping the call themselves.
The numbers look amazing. Call times plummet. But customer satisfaction? Straight into the ground.
Now replace “call time” with “prove theorem X.”
If human mathematicians did it, they’d refine definitions, polish lemmas, contribute back to Mathlib, train juniors, deepen the understanding of math structures, and strengthen the community.
The AI, by contrast, optimizes only for the explicit goal. It might generate a 10,000-line proof in hours. Perfectly correct, but unreadable, unusable, and useless for human learning.
The summit is reached but the forest along the way is gone.
We need to start making our implicit goals explicit and design systems that protect the values we actually care about, not just the numbers we can measure.
Terence Tao just wrote about this. He points out: AI is the ultimate executor of Goodhart’s law, i.e. when a measure becomes the target, it stops measuring what we care about.
Take a call center. “shorten average call time.” Sounds reasonable: shorter calls should mean faster resolutions, happier customers.
At first, it works. Agents become more efficient. But soon, people start gaming it: nudging customers to hang up when the problem is tricky, or just dropping the call themselves.
The numbers look amazing. Call times plummet. But customer satisfaction? Straight into the ground.
Now replace “call time” with “prove theorem X.”
If human mathematicians did it, they’d refine definitions, polish lemmas, contribute back to Mathlib, train juniors, deepen the understanding of math structures, and strengthen the community.
The AI, by contrast, optimizes only for the explicit goal. It might generate a 10,000-line proof in hours. Perfectly correct, but unreadable, unusable, and useless for human learning.
The summit is reached but the forest along the way is gone.
We need to start making our implicit goals explicit and design systems that protect the values we actually care about, not just the numbers we can measure.
77
114
952
88,016
Jasper retweeted
Thinking about hosting a math Olympiad for AI founders. You'd be up against our founder @zjasper (@hyperbolic_labs ) who's hella cracked. Finished his math PhD at Berkeley in 2 years. Tag someone who could take him on 👀
2
2
5
1,090
As probably the only mathematician building an AI infrastructure + fintech company, I wrote about what I’ve learned and how I see the future of @hyperbolic_labs through the lens of math.
Hyperbolic geometry, utilization, and why I care about intelligence per watt.
Hyperbolic geometry, utilization, and why I care about intelligence per watt.
5
2
61
4,569
@hyperbolic_labs Here’s the full piece, including what our $30M Series A1 will fund: hyperbolic.ai/blog/hyperboli…
1
0
2
329
See you soon!
Happy Wednesday. On today's show:
- @markgurman (Bloomberg)
- @zac_rash (Coco Robotics)
- @BenjamineYLiu (Formation Bio)
- @AmiramShachar (Upwind)
- @zjasper (Hyperbolic)
See you on the stream.
5
0
13
890
I sat down with @LipBuTan1.
Are we overbuilding AI? His take: the momentum will be even bigger than the internet. The real question is how fast we can scale through memory, GPU, power, and financing shortages.
Thank you, Lip-Bu Tan, Shankar Chandran, and @walden_catalyst for leading our $30M Series A1.
youtube.com/watch?v=8CbK6X…
Are we overbuilding AI? His take: the momentum will be even bigger than the internet. The real question is how fast we can scale through memory, GPU, power, and financing shortages.
Thank you, Lip-Bu Tan, Shankar Chandran, and @walden_catalyst for leading our $30M Series A1.
youtube.com/watch?v=8CbK6X…
3
0
28
4,194
Adding more GPUs is only part of the answer. The bigger opportunity is making the compute we already have easier to use, orchestrate, and trade.
Thanks @Geoff_Weiss and @BusinessInsider for the conversation.
Thanks @Geoff_Weiss and @BusinessInsider for the conversation.
Thanks @BusinessInsider and @Geoff_Weiss for covering our $30M Series A1 led by @walden_catalyst.
GPU capacity is fragmented across hundreds of providers. @zjasper on how we're making compute easier to use and easier to trade.
16
6
83
4,144
We started Hyperbolic to make the world’s compute unified, productive and liquid.
Today, we’re announcing our new $30M round to keep building toward that.
Grateful to our team, customers, and investors for believing in us. There’s a lot more to build.
Today, we’re announcing our new $30M round to keep building toward that.
Grateful to our team, customers, and investors for believing in us. There’s a lot more to build.
We’ve raised an oversubscribed $30M Series A1 led by @walden_catalyst to build the Global Compute Grid.
We’re turning fragmented GPU capacity into one unified platform where AI teams can access, manage, sell, and orchestrate compute.
Our mission: make the world’s compute liquid.
32
10
175
11,215
Great to see @CNBC covering the emerging market for GPU futures and featuring our work at Hyperbolic!
AI compute is becoming a new asset class, but the market infrastructure is still in its early days.
Financial contracts can help hedge price risk, but the bigger opportunity is t where compute can actually be bought, sold, and delivered.
That's exactly what we're building at Hyperbolic. Through Forge, we're unifying fragmented GPU infrastructure and enabling compute to move more efficiently across clouds.
The future of AI compute isn't just about more GPUs. It's about making compute liquid, productive, and unified.
Thanks to Oliver Renick and CNBC for featuring our vision!
AI compute is becoming a new asset class, but the market infrastructure is still in its early days.
Financial contracts can help hedge price risk, but the bigger opportunity is t where compute can actually be bought, sold, and delivered.
That's exactly what we're building at Hyperbolic. Through Forge, we're unifying fragmented GPU infrastructure and enabling compute to move more efficiently across clouds.
The future of AI compute isn't just about more GPUs. It's about making compute liquid, productive, and unified.
Thanks to Oliver Renick and CNBC for featuring our vision!
Regulators have yet to approve GPU index futures but ex-Citadel quant @zjasper is already looking ahead to physical settlement.
And no, it doesn't mean shipping you a data center. Ngl, this was quite illuminating ⬇️⚙️
Ft @drwconvexity & @carmenli
cnbc.com/2026/10/26/gpu…
9
3
38
3,943
@CNBCeave’s debt situation would be less concerning if providers had more ways to monetize available GPU capacity during a slowdown, especially when long-term contracts carry thin margins. Cash-settled futures can help falling prices, but they do not by themselves put idle GPUs to work.
0
0
0
221
Even @AnthropicAI will need to resell its idle compute on platforms like @hyperbolic_labs 🤫
3
0
24
1,360
Jasper retweeted
Regulators have yet to approve GPU index futures but ex-Citadel quant @zjasper is already looking ahead to physical settlement.
And no, it doesn't mean shipping you a data center. Ngl, this was quite illuminating ⬇️⚙️
Ft @drwconvexity & @carmenli
cnbc.com/2026/10/26/gpu…
And no, it doesn't mean shipping you a data center. Ngl, this was quite illuminating ⬇️⚙️
Ft @drwconvexity & @carmenli
cnbc.com/2026/10/26/gpu…
3
1
15
8,099
Jasper retweeted
Cash-settled GPU futures hedge the price. Physically settled futures deliver the compute.
Our CEO @zjasper on why physical delivery matters for the future of compute markets, and how we're building toward it with Forge, our unified control plane for GPU infrastructure.
Thanks @OJRenick for the deep dive.
Our CEO @zjasper on why physical delivery matters for the future of compute markets, and how we're building toward it with Forge, our unified control plane for GPU infrastructure.
Thanks @OJRenick for the deep dive.
Regulators have yet to approve GPU index futures but ex-Citadel quant @zjasper is already looking ahead to physical settlement.
And no, it doesn't mean shipping you a data center. Ngl, this was quite illuminating ⬇️⚙️
Ft @drwconvexity & @carmenli
cnbc.com/2026/10/26/gpu…
2
4
23
2,651
GPUs are becoming money-printing machines.
Lock in B300s at $4–5/GPU-hour and sell the capacity at $10–20/hour.
That's 2–5x the rental price.
Neolabs are already hoarding GPUs.
If you're sitting on H100s, B200s, or B300s, Hyperbolic can help turn your GPU capacity into revenue.
Start hoarding compute and monetizing it.
Lock in B300s at $4–5/GPU-hour and sell the capacity at $10–20/hour.
That's 2–5x the rental price.
Neolabs are already hoarding GPUs.
If you're sitting on H100s, B200s, or B300s, Hyperbolic can help turn your GPU capacity into revenue.
Start hoarding compute and monetizing it.
There is an absolute barrage of inference providers popping up that get to $1M/mo extremely quickly.
This is the rough economics of inference providers:
Even if you rent on-demand B300s at $8/GPU-hr, many open source models (GLM 5.3, Kimi K3, DeepSeek 4.1 Flash / Pro) can get easonable throughputs (100tok/s). DeepInfer from DeepSeek directly shows they can get $38/GPU-hr on off-peak hours. Every chip / model has a different throughput / interactivity curve, but there is a lot of arbitrage between even high rental spot prices and token prices. New chips (Vera Rubins) also change this meaningfully. Your ability to be making gross margin relies on your utilization and how good your stack is.
Gateways like OpenRouter, Vercel and more help solve that problem by aggregating demand to you. May inference providers discount their tokens to get longer term commits too. I've even heard stories of selling back Chinese model tokens to Chinese labs!
One thorn in the economics of these players is that many Chinese open source model labs are now asking for rev share if you're at a certain scale, espp for pre-release access to their models for day 1 support. That further pushes the threshold to be in the money.
The barrier to entry seems low and falling. There are even stories of smaller ones who can "vibe code" a way to stand up OSS models on a cluster they rent. And as a result, we're seeing 10s of such companies pop up almost every week and immediately skyrocketing revenue at low-ish gross margin and raising rounds on the back of that.
Unsurprisingly, because the actual service provided seems commodity, this eventually drives up compute prices and drives down token prices. In theory, it would seem like neoclouds who have the lowest cost of compute and themselves have started inference businesses would win the race to the bottom (Crusoe Coreweave Nebius and GMI all have inference businesses), with larger players (Modal Fireworks Baseten) being second because they are well capitalized to make bigger compute commits cheaply and have demand (high utilization), but the reality is the pie is growing so quickly that even upstarts are absolutely crushing it and growing aggressively (inference dot net, DeepInfra, Wafer, Relace, Morph).
Strategies vary across the board: neoclouds want to serve inference, inference players want to be neoclouds and do higher margin reserved instances / post training, and some want to go heterogeneous hardware or creatively flexible compute constructs.
The risks are: do open source models continue to be competitive? Will their creators rent seek more? Will their shape change compute needs? How do different chips change economics? Who gets compute in a supply crunch?
Below is a aggressive sketch of how gross margin scales with utilization for various models / chip sets.
For now, there are many inference providers and they’re all ripping!
16
13
278
23,431
Jasper retweeted
Berlin rooftop, cocktails, and great conversations about compute, models, and inference 🍸
Thanks to everyone who joined us for Cocktails & Compute at @NVIDIAGTC Berlin!
Which city should we host next?
Thanks to everyone who joined us for Cocktails & Compute at @NVIDIAGTC Berlin!
Which city should we host next?
0
2
8
2,275
Excited to launch Flux, the next generation of Hyperbolic inference! 🚀
Efficient inference requires both optimized model serving and smart orchestration. How you allocate GPUs, place workloads, and improve utilization directly shapes the cost and performance of every token.
With Flux built on Forge, we’re bringing model serving and compute orchestration together, working toward a future where inference workloads can dynamically use available capacity across our global compute grid.
Huge thanks to the team for bringing this to life across engineering, product, and marketing. Proud of what we’ve shipped and excited for what comes next.
Give Flux a try and let us know what you think!
Efficient inference requires both optimized model serving and smart orchestration. How you allocate GPUs, place workloads, and improve utilization directly shapes the cost and performance of every token.
With Flux built on Forge, we’re bringing model serving and compute orchestration together, working toward a future where inference workloads can dynamically use available capacity across our global compute grid.
Huge thanks to the team for bringing this to life across engineering, product, and marketing. Proud of what we’ve shipped and excited for what comes next.
Give Flux a try and let us know what you think!
Introducing Flux, Hyperbolic’s next generation of inference, built on Forge.
Choose from 10 open models through one OpenAI-compatible API.
Start building with Flux → app.hyperbolic.ai/models
Explore the models and architecture: hyperbolic.ai/blog/flux-infe…
4
0
21
1,666
Proud to partner with @EagleCorp as an official launch partner for EAGLE-4 - bringing their fastest spec decoding solution to @hyperbolic_labs and accelerating the future of AI!
Congrats to the @EagleCorp team on EAGLE-4, their fastest yet, hitting 961.6 tokens/s on Qwen3.8-27B!
Proud to be an official launch partner, with EAGLE-4 coming soon to Hyperbolic.
2
0
8
1,039
AI compute demand is accelerating, and we’re seeing it firsthand at @hyperbolic_labs.
Great conversation with the @blockspace team about GPU demand, supply bottlenecks, and the future of the AI compute market. Thanks for having us!
Great conversation with the @blockspace team about GPU demand, supply bottlenecks, and the future of the AI compute market. Thanks for having us!
Why $NVDA GPUs Sell Out In Hours
Blockspace sat down with @hyperbolic_labs. The SemiAnalysis rated neocloud is one of the fastest growing companies in compute.
We discuss:
• Velocity on B300 sales
• Bottlenecks for AI Labs & neoclouds
• Trends in compute sales from the front lines
0
5
28
3,401
Many neoclouds are essentially hardware leasing businesses, taking on massive debt and long-term commitments while having limited software differentiation. That creates significant risk when GPU prices decline.
This is exactly why we’re building Hyperbolic differently. Rather than betting on the value of hardware, we’re building the software and market infrastructure to unify fragmented compute supply, improve utilization, and provide liquidity across the global GPU market.
We believe the long-term winners won’t necessarily be those who own the most GPUs, but those who can most efficiently allocate, orchestrate, and monetize compute.
In fact, declining GPU prices could benefit platforms like ours by making compute more accessible and expanding overall demand.
This is exactly why we’re building Hyperbolic differently. Rather than betting on the value of hardware, we’re building the software and market infrastructure to unify fragmented compute supply, improve utilization, and provide liquidity across the global GPU market.
We believe the long-term winners won’t necessarily be those who own the most GPUs, but those who can most efficiently allocate, orchestrate, and monetize compute.
In fact, declining GPU prices could benefit platforms like ours by making compute more accessible and expanding overall demand.
Ben Horowitz explains why a16z doesn't invest in neoclouds after living through the dotcom bubble
"If you get into a pathological debt situation, that can wreck anybody. It's why, by the way, we're not invested in any of these neoclouds, for that reason. That whole thing scares it."
"It can end up being a very dangerous arbitrage business where you can get upside down easily."
"You look at Loud Cloud just as an example, here's a simple thing that happened to us in 1999. Bandwidth was $1,500 a megabit. And you couldn't get it, it was very hard to get. And so you had to get a long-term contract, three years."
"By 2001 it was $50 a megabit. So if you were holding bandwidth, you were just dead."
"So that kind of thing is so out of the control of the entrepreneur. And so different than what we'd like to fund. That, you know, it just scared me to death."
11
2
59
6,235