Dylan Patel

Down payment on a EUV machine!

@priya_kamdar @anikamirzaa You think I'm changing?

@anikamirzaa @dylan522p Thank god Apple is releasing a folding iPhone later this year

😔😔

Whenever I make a group chat green, I now get to proudly proclaim it as goblin mode

Are babies and relationships measured on the same time scale? After 18-24 months you start measuring in years.

People hate on SF men for group think, but every girl in SF has Tabi's now It's an epidemic

@iAmHenryMascot @AnthropicAI @chamath We are building more than ever yes

You went from couple of hundred thousands in token spend annually to now $11m, doubled from $5m in the last 2 weeks. Are you seeing massive rev lift to justify this or you are investing way ahead of the future hoping for this? @chamath said big enterprises are yet to see the productive rev lift from token spend yet, and that needs to happen to justify all the revenue accruing to @AnthropicAI and co.

Claude Code spend had gotten to $10.95M runrate peak at SemiAnalysis But then Opus 4.7 saved me. More token effecient for tasks, smarter, and no fast mode. Thank you @AnthropicAI You saved me from bankruptcy

@iAmHenryMascot @ArtificialAnlys @Kimi_Moonshot @dwarkesh_sp You see it in the Agentic coding usecase as AA described. And it's that you see divergence happen this year. Which we are starting to since mythos.

@dylan522p means you were a bit wrong, you said we were going to see huge divergence on model capabilities this year cos the chinese OSS models don't have access to compute that OAI & Ant have in @dwarkesh_sp's podcast earlier in the year. This doesn't seem to be the case at all.

All three leading open weights models were released last week. Progress continues for open weights models alongside proprietary ones, with the gap to GPT-5.5, the leading proprietary model, sitting at 6 points on the Artificial Analysis Intelligence Index @Kimi_Moonshot’s Kimi K2.6 (Reasoning) and @Xiaomi's MiMo V2.5 Pro (Reasoning) tie as the leading open weights models on the Artificial Analysis Intelligence Index at 54, with @deepseek_ai's DeepSeek V4 Pro (Reasoning, Max Effort) at 52. This places the best open weights models within 3-6 points of the leading proprietary models: @OpenAI's GPT-5.5 (xhigh) at 60, and @Google's Gemini 3.1 Pro Preview and @AnthropicAI's Claude Opus 4.7 (Adaptive Reasoning, Max Effort) at 57. For context: just one year ago the highest-scoring open weights model was DeepSeek V3 0324 which achieved 22 on the Intelligence Index, and was ~13 points below the highest-scoring proprietary model, Claude 3.7 Sonnet (Reasoning) at 35. Key takeaways: ➤ The top three most intelligent open weights models are trillion-plus-parameter MoE architectures with permissive licenses. Kimi K2.6 (Reasoning) has 1T total / 32B active parameters with 256K context window, MiMo V2.5 Pro (Reasoning) has 1T total / 42B active with 1M context window, and DeepSeek V4 Pro (Reasoning, Max Effort) has 1.6T total / 49B active with 1M context window. ➤ The gap to proprietary remains wide on the hardest reasoning and agentic coding evaluations. On HLE (Humanity's Last Exam) the three top open weights models score 34-36%, vs 44% for GPT-5.5 (xhigh) and 45% for Gemini 3.1 Pro Preview. On CritPt (Research-level Physics) they score 4-12%, vs 27% for GPT-5.5 (xhigh). On TerminalBench Hard (Agentic Coding & Terminal Use) they score 43-46%, vs 61% for GPT-5.5 (xhigh) and 54% for Gemini 3.1 Pro Preview. ➤ Omniscience (knowledge + hallucination) shows a large gap to proprietary models, with DeepSeek V4 Pro (Reasoning, Max Effort) hallucinating significantly more than its open weights peers. DeepSeek V4 Pro (Reasoning, Max Effort) scores -10, MiMo V2.5 Pro (Reasoning) +4, and Kimi K2.6 (Reasoning) +6. By comparison, GPT-5.5 (xhigh) scores +20, Claude Opus 4.7 (Adaptive Reasoning, Max Effort) +26, and Gemini 3.1 Pro Preview +33.

AI Value Capture - The Shift To Model Labs Vera Rubin VR NVL72: V for Value - Rubin delivers a step jump in performance per TCO. ROI accruing to users, Neoclouds, Hyperscalers, AI Labs, Memory Vendors or GPU Manufacturers? READ NOW: newsletter.semianalysis.com/p/ai-value-cap…

@climatetech_vc @dwarkesh_sp Bro ur missing 75%+ datacenters chrrently under construction right now lmfao How are your numbers that low? Our pipeline is over 100gw ofc

And just to be clear @dylan522p is not the only girl in town tracking data centers. We saw on @dwarkesh_sp how well he understands the power system.

The Hedgeye post is misleading and clickbaity. The real story is that 30–50% of the global hyperscale (>50MW) data center pipeline for 2026 MIGHT not materialize. Our research is the primary source that Bloomberg and Tom’s Hardware are citing: sightlineclimate.com/research/data-…

@DomResidAgg @AnisBay919 Very well semianalysis.com/energy-model/

@dylan522p @AnisBay919 How well are you able to estimate metered load (not IT load) mr Patel? I’m assuming you don’t monitor anything RT yet but do you have anything about typical metered load throughout time of day, etc?

BREAKING: this account is retarded

every day i wake up and look in the mirror. i must choose between the two wolves within

@thesephist Banger

@AnisBay919 Vauge statements vs the only firm that properly tracks every datacenter in the world

@dylan522p You might be misinformed: tomshardware.com/tech-industry/… bloomberg.com/news/features/…

@mots_pod @creatine_cycle @jxnlco @baelentina_ @SemiAnalysis_ Frfr she's the real hero. @jxnlco was the unlocker

@creatine_cycle @dylan522p @jxnlco @baelentina_ @SemiAnalysis_ also important to shoutout michelle. iykyk

we spoke about the @dylan522p glow up thanks to @jxnlco. dylan before: - "he looks like a nice guy" - @baelentina_ - a friendly young chap - he looks like the guy an indian auntie shows a girl and the girl is like "idk" - @SemiAnalysis_? the newsletter? dylan after: - dashing - women flock to him - he RUNS SemiAnalysis - he OWNS SemiAnalysis

@RHouseResearch @SemiAnalysis_ Imagine if you had the tokenomics model which was way better, more granular data, more frequently. semianalysis.com/tokenomics-mod…

The irony of the WSJ dropping a hit-piece referencing last year's numbers is that the real-time signal on OpenAI came from @SemiAnalysis_ just a few days ago.. With GPT 5.5's release and Anthropic's capacity constraints, OpenAI probably is accelerating as we speak while the WSJ is writing about 2025, lol.

@creatine_cycle @jxnlco @baelentina_ @SemiAnalysis_ Bruh that was literally a year ago that we went shopping for a day

We agree. The real leading indicators are clear: breakout Codex growth ✅, enterprise offerings on every cloud ✅, the only consumer app that matters ✅, a compute strategy built to accelerate ✅, and the best researchers in the world ✅.

It’s crazy to me that @dylan522p even knew when my cofounders bday was! So much alpha!