H100 ornn index spot prices are falling, now at $2.42 per hour, roughly 40% below the May peak. The ecosystem is concerned that this is a sign that compute demand and by extension the appetite for AI is waning. (1/5)🧵
SemiAnalysis
2,549 posts
- Replying to @SemiAnalysis_Falling spot prices alongside rising contract prices are therefore not evidence of weaker demand. It is more likely a shift of opportunistic capacity usage toward committed production deployment. Serious buyers are locking in term capacity, and that is pushing contract pricingThe contract market is the one that better reflects where durable AI workloads are actually going and we track this in our GPU pricing index. (5/5)
- "So you're saying that your SRAM supply is infinite?" "Yes" "But the logic wafers on which the SRAM is fabbed is supply constrained?" "Yes Dave that's right"
- BREAKING NEWS: OpenAI’s Technical Lead for Compute has returned to OpenAI after joining Meta in April 2026. What is going on at Meta that people are quitting after only a couple of months? Did Anuj get reorged into a data-labeling role at Meta, and is that why he left?
00:18Today we're announcing our $320M Series A at a $2.3B valuation, led by @khoslaventures , with @generalcatalyst, @JeffBezos , @ericschmidt, @NicoRosberg, alongside leading researchers from frontier labs and academia.- US Grid Constraints: Towards 40GW+ of Behind-The-Meter Datacenter by 2028? Why the Grid Can't Keep Up, and Why that Drives Behind-The-Meter 50%+ of DCs/Year By 2028
- Our CPUs Are Back piece is the work weve been pointing people to whenever they ask why server CPU revenue, which was supposed to be dead, is suddenly behaving like an early-cycle category again. (1/4)🧵
Replying to @SemiAnalysis_What we walk through in the article is how four demand drivers are stacking at the same time: reinforcement learning, which needs CPU clusters running compilation and verification to keep accelerators busy; agentic models, which call tools and hit databases far more aggressively
On the supply side, we map the 2026 roadmap from AMD Venice and Verano (256 cores, TSMC N2) to Diamond Rapids and Coral Rapids (Intel 18A-P, but losing SMT) to Arm pivoting from licensing IP to selling Phoenix as a full chip with Meta as lead customer. The punchline, and where - Between colleagues attending dates with robots and ishowspeed getting mogged by a rizzbot, man is one step away from humanoid kind. EPISODE 16 LIVE NOW! youtube.com/watch?v=MD5zKs…
00:00 - It literally is a Slack bot lolReplying to @gallabytesThis is correct, I think a number of people on the tl didn’t read past the title and made inferences and comparisons that are just wrong and then use it as an opportunity to take cheap shots. This is not a “feature” like some crappy Slack bot and it’s certainly not a Claw, though
- IN-DEPTH: What Unitree's Evolution Means For Robotics Unitree Playbook, China's Scaling Law, Geopolitics, and Humanoids in the wild
- Chat develop a chip from initial design to tape out in 9 months, make no mistakes.
- NVIDIA POOR DRIVER QUALITY ALERT: There is a GB300 NVL72 firmware bug where the rack needs to be rebooted every 66.5 days. Although people tend to think of NVIDIA as having top-tier software, it turns out there are still many issues with NVIDIA drivers and firmware. The thing is,
- Meta leadership voting on a motion to re-allocate 7k engineers to data labelling org
00:00 - SemiAnalysis repostedHi i'm dwarkesh! Grew up all over the US, now sf-based and always down to nerd out about AI, science & history :) a lil about me: 🟠 Host of the dwarkesh podcast 🟠 Studied at UT Austin 🟠 Just published a book on the history of AI scaling Lets grab coffee or do a fun activity
