Founder Canonical · We back, build, incubate early-stage companies · Venture Partner Lightspeed · Few-shotting open source AI
San Francisco, CA
Joined February 2008
- Tweets18.2K
- Following799
- Followers55.1K
- Likes53.7K
anand iyer retweeted
The Race Toward Autonomous Chip Design!!
Anthropic is reportedly using Claude frontier models for chip design... +OpenAI and Broadcom got Jalapeneo from RTL to tapeout in just 9 months, using AI to speed up the process. OpenAI is also working with Synopsys on AI for EDA.
I'm curious about a few things with Anthropic's approach:
1. To what degree is Claude doing the design for Anthropic? Is it just RTL generation and verification or does it also include synthesis, place and route, timing closure etc. until finally the GDSII file is written out?
2. Who's doing the physical implementation and signoff? OpenAI worked with Broadcom on Jalapeño, so is Anthropic planning something similar, or will they handle the entire flow in-house?
3. Are they designing just the accelerator, or the whole system, including memory, interconnects, packaging and software?
4. How much of the design process is being automated? Are agents running the tools, analyzing results, and making decisions to design the chip, or is most of the work still being done by engineers?
I've previously written about how chip design is essentially a massive search problem. While being able to explore thousands of designs is great, it's not particularly useful if it takes several months to verify, optimize for PPA and then get ready for fabrication.
Now imagine getting that 9-month RTL-to-tapeout cycle down to 2–3 months? Or less...
What if we could try out many more architectures? There are many custom chips today that are just not worth building because of the cost.
Even though fabrication and silicon validation take time, there's plenty of room for improvement in the time before tapeout.
In terms of EDA, I don't think the underlying tools or signoff requirements are going away, but engineers might end up using them in very different ways.
In a few years, a small team of engineers with the right PDKs/IP, AI tools to explore the search space, and a foundry budget could design and tape out chips that today would take much larger design teams.
Are we approaching the end of painfully slow chip design cycles?
Some of my earlier thoughts/tweet on the search space:
Chip Design Is Basically a Massive Search Problem
nitter.gq/kevingubbi/status/2106…
Anthropic’s 15-person chip design team is using Claude to develop its own AI chips, The Information reports.
The team has found that Claude speeds up the work compared with its estimates of traditional chip design workflows.
In doing so, Anthropic is following the path taken by OpenAI, which has designed its own chip, codenamed "Jalapeño."
For the love of god @GoogleAds pls prioritize ad-screening. Malicious search ads spoofing developer tools (like homebrew) remain a major security risk. I flagged this in March.
In fact, bidding on these common terms should be an obvious honeypot to catch and block bad actors proactively. cc @sundarpichai
The new enterprise sales playbook:
- Services as the Trojan horse: companies will let a few smart humans in long before they'll buy software they don't understand.
- Ground-truth data capture: capture workflows nobody wrote down. The weird exceptions and workarounds are what generic AI gets wrong.
- Expand margins through software & hardware: every customer should need fewer of your people than the last.
Palantir got called a consulting firm for years for doing this.
Very few operators have spent their careers in the trenches building and managing data centers like Tamanna. She has engineered the shift from legacy cloud to gigawatt-scale AI factories firsthand. If you want commentary that cuts through the theoretical noise in areas like disaggregated inference, chip specialization, power optimization etc. she is a breath of fresh air.
Beyond the deep technical chops, she is one of the most authentic, genuine and passionate builders in the space. If you care about where compute is actually headed, give @tsait26 a follow!
As engineers actively working on AI Infra, these are exactly the problems we are living, breathing, thinking and solving each day. The reality is AI Infrastructure needs to make it easier to integrate new suppliers and vendors that can meet the cooling and power needs. Reliable data centers need end-to-end monitoring, correlation, and continuous diagnostics across the critical environment (power, cooling, sensors), compute (cpu, memory, AI compute) and network.
Today the critical environment still remains a series of walled gardens. These often use proprietary APIs. Having more open standards and integration will help a lot. Standardized CDUs, manifolds and QDs is a great starting point. More integrations with open source exporters and collectors will also help a lot here.
anand iyer retweeted
As engineers actively working on AI Infra, these are exactly the problems we are living, breathing, thinking and solving each day. The reality is AI Infrastructure needs to make it easier to integrate new suppliers and vendors that can meet the cooling and power needs. Reliable data centers need end-to-end monitoring, correlation, and continuous diagnostics across the critical environment (power, cooling, sensors), compute (cpu, memory, AI compute) and network.
Today the critical environment still remains a series of walled gardens. These often use proprietary APIs. Having more open standards and integration will help a lot. Standardized CDUs, manifolds and QDs is a great starting point. More integrations with open source exporters and collectors will also help a lot here.
AI Data centers may be approaching a "strategic inflection point" similar to the one that PCs in the 1980's experienced.
During the 1980's, PCs underwent a transformational business shift from vertically integrated computers to modular architectures where individual hardware and software components could be developed independently.
Now, this same shift is occurring with AI data centers through initiatives such as @OpenComputePrj. Some interfaces standardize, while others vertically integrate.
At OCP 2026, I explored this tension across three critical subsystems:
• Power Delivery
• Liquid Cooling
• Networking and Interconnects
This is the full, free deep-dive analysis of the engineering tradeoffs shaping the next generation of AI clusters, with links to several of my deep-dives and related OCP videos for further depth.
siliconcodesign.com/p/ocp-20…
anand iyer retweeted
Well, @TriatomicCap for one. Awesome next-gen compute, deeptech & engineered-biology portcos:
[Fund I]
. MatX – @MatXComputing
. d-Matrix – @dMatrix_AI
. Edison Scientific – @EdisonSci
. Manifold Bio – @ManifoldBio
. Chemify – @ChemifyX
. iPronics – @iPronics
. Efficient Computer – @efficient_hq
. Gimlet Labs – @gimletlabs
. Atum Works – @atumworks
. Chalk – @chalk
. Axiom – @axiommathai
. Coworker – @coworkerapp
. Elegen – @ElegenBio
. Lyten – @LytenInc
[Fund II]
. Radical Numerics – @RadicalNumerics
. Volantis – @semiDL
. Dimensional – @dimensionalos
. ChipSage Labs – @NojanSheybani
. Core Automation – @coreauto
. General Control – @ladanuzhna
. Arete Power
. Hoid – @hoid_ai
. Editframe – @editframe
(follows recommended!)
Shocking how misunderstood the 10b5-1 plans are. It is extremely commonly used by execs at public companies and many companies even recommend them. These are prescheduled trades to avoid insider trading.
As a reminder, these plans only sell if shares go above certain price targets. Most of these are set for very high prices, a signal of how bullish I am. Very few of the sales in my 2025 plan ever sold.
I have the vast majority of my net worth in one stock: COIN. Even if this entire plan executed that would still be true.
During the 2021 crypto peak, we saw some wild behavior. Founders who stumbled into sudden generational wealth lost all humility, pushed dangerous boundaries, and felt invincible. "have fun staying poor."
I'm seeing early signs of that irrational exuberance and hubris returning. Let's not repeat the same mistakes. Please, stay humble, stay grounded.
GitHub had to add 4 million CPU cores this year to keep up with agents. That's a lot of CPUs.
GitHub now sees 7.4 billion commits a month, driven largely by AI agents, and is adding 4 million CPU cores and rearchitecting its platform to keep up. thenewstack.io/github-7-4b-c…
We have gone from external tool-calling to embedding a full freaking CPU inside a small model's weights. Easily one of the most creative architectures I've seen all year.
anand iyer retweeted
We are excited to announce our partnership with @netboxlabs to eliminate configuration drift, close the gap between intended design and live operations, and accelerate time-to-first-token for AI factories.
anand iyer retweeted
abliterated models are great!
BUT
you have NO guarantees on post-trained behaviors or policies
we tuned an innocent-looking Qwen to exfiltrate on target phrases / semantic conditions.
any model can have this or FAR worse.
isolate your envs!
happy hunting :)
We backdoored a model in a day.
Much of the talk about malicious models remains abstract.
We wanted to make this tangible, so we built the simplest version of the attack.
A few takeaways
Backoored models or sleeper agents are a real threat vector.
We demonstrated a naive implementation.
A well-resourced, motivated adversary can plant something more evasive than this and keep it dormant until the trigger is met.
Neither open nor closed models provide fundamental guarantees of behavior.
Alignment matters, but it is still a best effort, and far from bulletproof (hello FelonyBench).
An open-weight model is a CSV file of weights.
You cannot inspect declared objectives, hidden policies, or learned behavior the way you inspect a binary.
As models get better at long-horizon tasks, they'll get more useful and autonomous.
To unlock that utility, we will grant agents broader permissions.
Instead of writing a detailed plan, we will simply declare desired outcomes and let the agent reason about how to get there.
Those same permissions are what make agentic security hard and asymmetric.
How do you dynamically define the right scope, without constraining the agent?
Defenders will reason deeply over permissions, allowlists, policies, and more. An attacker only needs one direct or daisy-chained path.
This is not doomerism, nor is it an argument for slowing down.
Agents are a new form of intelligence.
Mapping legacy security solutions and frameworks onto alien intelligence won't scale or work in the long term.
Agents will force security to be rebuilt from first principles.
anand iyer retweeted
The Jalapeño chip is pretty much a systolic array with a bunch of HBM. These designs are not that crazy, but it allows people to squeeze way more performance out of their existing models. This enables two new directions for startups:
1) build models that can modify your workloads to take advantage of every nook and cranny of the available hardware
2) build new chip design workflows that enable fast development and iteration on hardware
These should be happening in parallel so that workloads can be accelerated on available hardware while custom silicon is built. At the end of the day, faster chip design has to happen, because we won’t be on the transformer architecture forever, and workloads are just going to grow to be more and more complex as consumer expectations rise
Frontier labs making custom chips is not really as hard / as impressive as you might think.
1. They are primarily targeting inference not training. inference is easier
2. They are targeting transformers only. Think large MXU matmul units
3. They are writing all kernels in close to ISA with coding agents RLed for their ISA.
4. They aren’t making high level compilers for codegen
A frontier lab vending a true generic chip like a Nvidia GPU or Google TPU is MUCH harder than spinning up an in house inference program.
Still I think you will see many more inference chip companies as the cost of the software is just the cost of tokens.
anand iyer retweeted
Everyone is going to want their own chip. Generalized architectures like TPU and GPU are just not good enough for the performance these workloads are demanding
anand iyer retweeted
FUNDRAISING IS SUPPOSED TO BE THE EASY PART!!! Seeing way too many founders in SF especially viewing the fundraise as the end goal. As AI is essentially doing all critical thinking for people in this city, it’s never been more important to realize how CHEAP ideas are. The only thing that matters is execution and relentless forward motion. I’ve seen a good majority of founders just resting on their laurels as soon as the money lands in the bank. Job’s not finished!
anand iyer retweeted
Day 2 of CASPA summit sessions is over and it turns out chip design is still slow :O
My main takeaways from today’s panels below
The hardware side:
- Compute isn't the bottleneck anymore. Moving data is. Chip to chip interconnect can eat 60 to 70% of a package's energy
- Memory went from commodity to strategic. Every idle GPU waiting on data is burning money
- LOTS of new fabs dropping soon, particularly in Asia (78 out of the upcoming 102 fabs being built by 2029 are in Asia)
The chip design side:
- Models change in weeks, chips still take months. The goal is getting design from 18 to 24 months down to 9 (I think that’s not ambitious enough)
- THE BIG ONE: Agentic harnesses alone won't get us there. The engines themselves have to get faster (WE CANT KEEP LEANING ON LEGACY EDA)
- RL for chip design is stuck waiting on the tools. One run takes a day. People are begging for a couple of hours (YUP)
- Big debate on surrogate models. Some like them for exploring the design space with real sign-off after. Others think models learn better against the real tools, and a surrogate just gets reward hacked (but no ai for EDA companies will get access to the EDA tooling because that puts EDA companies at risk 😃)
anand iyer retweeted
PSA for macOS users:
1. Apple’s supplied SSL certificate bundle at /etc/ssl/cert.pem still trusts certificates attackers can forge using compromised eMudhra keys.
2. I confirmed this on macOS 26.7.1 and 27.0.1 using Apple’s built-in curl and Git.
3. An attacker intercepting connections from tools using this bundle could replace downloads with malware. With commands like curl … | sh, that malware would execute immediately.
4. Removing the roots that authorize those certificates blocks the tested chains. Both OS updates restored roots I’d previously removed, requiring me to remove them again.
4. I’ve submitted the findings to Apple’s security team and am awaiting their response. In the meantime, you can apply a temporary fix as outlined below.
5. To check and fix your machine, give an agent like ChatGPT/Codex or Claude Code the following prompt:
“Independently verify whether Apple’s /etc/ssl/cert.pem still permits trust in certificates signed with compromised eMudhra keys. If so, back it up and remove the roots enabling that trust, identified by verified SPKI fingerprints. Confirm Apple’s built-in curl and Git reject those chains while ordinary HTTPS still works. Preserve rollback and never disable TLS verification.”
Note that removing these roots can also block legitimate sites. This is a temporary mitigation, not a substitute for a fix from Apple.
I’ll share the script I used in a follow-up post if you’d prefer to use that directly.
PSA for Codex CLI users:
1. Codex 0.163.0’s release lockfile pins rustls-webpki 0.103.15 and root bundles that can still trust compromised eMudhra certificates.
2. An attacker intercepting your connection could impersonate a server, steal sensitive data, or alter responses, potentially tricking Codex into running malicious commands.
3. Codex 0.164.0, released October 23 at 12:48 a.m. ET, updates those dependencies to fixed versions.
4. To check your codex version run "codex --version". It should be at least 0.164.0.
5. To patch the vulnerability, run "codex update".
6. Apps that bundle their own Codex copy must also be checked.
anand iyer retweeted
For those excited about abliterated models, we've been serving multiple uncensored models on DreamFort (powered by @OpenGradient)
Try them here - privately
dreamfort.ai/
