Jun 30, 2026 · 1h 32m · invest-like-the-best
The Two Harvard Dropouts Who raised $800M to take on NVIDIA · Invest Like The Best
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of Invest Like The Best, host Patrick O'Shaughnessy interviews Etched co-founders Gavin Uberti and Rob Wachen, who detail how their startup is building specialized custom silicon to disrupt AI inference and challenge NVIDIA. The founders discuss overcoming widespread skepticism, their novel architectural breakthroughs, high-urgency execution culture, and the macroeconomic future of global token production.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Patrick holds 22.4% of the talking time here. How this is scored →
Patrick as informed peer 3.8 Guest teaching 5.3 Guest disagreement 1.9 Patrick pushing back 1.1
speaking balance: gold is Patrick, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 56:10 Dismissing hyperscaler in-house silicon efforts
Rob bluntly asserts that proprietary ASIC efforts from Google, Meta, Microsoft, and OpenAI are non-existential side projects incapable of matching dedicated chip companies in architectural risk-taking.
Hardest push from Patrick ▶ 1:11:10 Pushing on investor base rates vs startup naivetePatrick challenges conventional investment dogma by highlighting how all mainstream semiconductor experts labeled the Etched thesis foolish, arguing that rigorous base-rate investing would destroy outlier venture bets.
Biggest teaching moment ▶ 7:28 Deconstructing Dennard scaling and voltage limitsRob educates Patrick on why adding peak FLOPS without lowering operating voltage causes thermal throttling, proving low-voltage architecture is physically necessary for inference.
Patrick holds their own ▶ 25:40 Synthesizing robotics playbook to startup operating modelPatrick displays deep strategic grasp by connecting Gavin's high school robotics winning strategy directly to Etched's vertical integration and strict communications blackout.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Patrick as informed peer | Guest teaching | Guest disagreement | Patrick pushing back | Why |
|---|---|---|---|---|---|---|
| Overcoming Industry Skepticism as Young Founders | 3 | 5 | 3 | 2 | Patrick sets the stage by recounting industry consensus that 20-somethings cannot start semiconductor firms. Rob counters conventional EDA tool assumptions, schooling Patrick on why temperature corners can be discarded in specialized AI data center environments. | |
| Early Validation and Recruiting Veteran Leadership | 4 | 7 | 3 | 1 | Gavin and Rob deliver an in-depth technical masterclass on pre-fill vs. decode disaggregation, Dennard scaling, low-voltage power delivery, and cluster-scale interconnect bottlenecks. Patrick listens attentively and offers simple analogies to follow along. | |
| Sponsor Message: Ramp, WorkOS, and Rogo | 2 | 5 | 1 | 1 | Following the mid-roll ad read, Patrick asks why throughput and token cost are the ultimate decade-long bottlenecks. Rob and Gavin explain concurrency scaling and draw an analogy to mass production economies of scale. | |
| Founder Origin Stories: Personal Motivations and Early Career Beats | 3 | 3 | 1 | 0 | Rob recounts his battle with stage four bone cancer and how ChatGPT diagnosing his MRI sparked his realization around compute constraints, while Gavin shares his teenage kernel engineering work at Xnor. Patrick acts as a supportive interviewer. | |
| High School Robotics Strategy and Extreme Vertical Integration | 4 | 4 | 2 | 2 | Patrick connects Gavin's two-person high school robotics team strategy directly to Etched's aggressive vertical integration and lack of public communications. Gavin and Rob validate the parallelization thesis. | |
| Recruiting Philosophy: Legends and First-Principles Thinkers | 4 | 5 | 2 | 1 | Rob and Gavin detail their bimodal recruiting philosophy pairing industry veterans with raw first-principles builders. Patrick actively probes and categorizes their talent taxonomy. | |
| High-Urgency Execution: The Bangalore Sabbatical and Prefetching | 5 | 6 | 2 | 2 | Gavin describes moving top engineers to Bangalore to save a critical tapeout schedule, while Rob explains prefetching and hardware-in-the-loop emulation. Patrick compares their aggressive operational tactics to Elon Musk's zone-flooding playbook. | |
| Semiconductor Supply Chains, TSMC Partnership, and Power Infrastructure | 4 | 6 | 2 | 1 | Patrick brings up global supply chain shortages and memory stock hype. Gavin and Rob reframe the discussion around TSMC's exceptional collaborative service culture and the fundamental constraint of power availability. | |
| Evaluating Compute Density, Interactivity Curves, and Mega-Clusters | 5 | 7 | 3 | 1 | Patrick invokes Cerebras and interactivity curves for Blackwell and AMD. Rob breaks down ISO-interactivity concurrency metrics, and Gavin highlights the massive gap between current 4000ns interconnect latency and physics limits. | |
| Kernel Engineering and Hardware-Software Philosophy | 4 | 7 | 3 | 1 | Patrick asks about the longevity of human kernel engineering. Gavin and Rob explain why Etched chose kernel-first over graph compilers, a contrarian bet validated by hiring heavily from high-frequency trading. | |
| Vertical Integration Limits and AI Chip Market Dynamics | 6 | 5 | 3 | 2 | Patrick questions the boundaries of vertical integration given hyperscalers designing internal ASICs like OpenAI's Jalapeno. Rob and Gavin counter that hyperscalers lack existential urgency, which limits their willingness to take architectural risks. | |
| Overcoming the Hardest Technical Challenge in Chip Design | 3 | 6 | 2 | 0 | Gavin describes a near-catastrophic clock crossing domain flaw requiring 50-picosecond phase synchronization that caused early team members to quit before Etched engineered a drifting lock mechanism. | |
| Sponsor Break: Vanta Trust Platform | 4 | 5 | 1 | 1 | After the sponsor break, Patrick prompts the founders to recall their harrowing early fundraising days. Rob and Gavin detail the sheer capital intensity of sub-5nm tapeouts and how Silicon Valley passed before they secured their $103M Series A. | |
| Winning Over TSMC at Age 22 | 6 | 4 | 1 | 2 | Gavin tells the story of convincing a senior TSMC VP over dinner napkin math at age 22. Patrick breaks the fourth wall, articulating his investor perspective on base rates and embracing contrarian risk. | |
| Cultivating a Find-a-Way Culture and Silicon Validation | 4 | 5 | 1 | 1 | Rob describes the terrifying moment when their first wafer sort returned all red squares before team veterans methodically isolated the issue. Patrick reflects on the rarity of an institutional 'find-a-way' culture. | |
| Next-Generation Architecture and Supply Chain Strategy | 4 | 6 | 2 | 1 | Rob outlines the shift toward gigawatt-scale manufacturability for next-generation chips and explains how Etched avoids zero-sum wafer allocation by intentionally utilizing alternative process nodes and memory tiers. | |
| The Future of AI Models, Agent Swarms, and Data Centers | 4 | 7 | 3 | 1 | Gavin argues machines process memory fundamentally differently than biological brains, predicting trillion-dollar token factories and billion-agent swarms. Patrick probes on whether trillion-dollar data centers are physically and economically feasible. | |
| The Alien Framing: Communicating the Global AI Opportunity | 3 | 6 | 2 | 0 | Patrick poses the alien thought experiment. Gavin and Rob articulate why token production economics and scale-up cluster design represent the most valuable industrial frontier. | |
| The Kindest Thing: Rob Wachen's Personal Story | 1 | 1 | 0 | 0 | Patrick closes with his traditional question about the kindest thing anyone has done. Rob shares the moving story of his parents relocating to Boston during his grueling cancer treatments. |
Statements from this episode (46)
Wachen: The semiconductor stack is over-buffered for general-purpose use cases
“The entire semiconductors and data center industry is built on buffer. And what I mean by that is every part of the stack from the EDA tools to the power modules, to the circuit boards, to the chip design and standard cells, everything is built to be general p…”
Wachen: EDA tools default to freezing temperature constraints irrelevant to AI
“You know, the default configurations for a lot of these EDA tools assume that you're going to be running your chips in freezing temperatures. Now, I don't know about you, but I've never seen an AI data center with ice in it. So, you know, we can feel pretty co…”
Wachen: Bitcoin mining ASICs run at under a quarter the voltage of GPUs
“Bitcoin miners run at under a quarter of the voltage of GPUs.”
Wachen: All AI chips in the future will be low-voltage
“And we think all AI chips in the future are going to be low voltage chips. They're going to have to cram way more flops in the same silicon area, and without thermal throttling run at way lower voltages.”
Uberti: NVIDIA Blackwell point-to-point latency is about 4,000 nanoseconds
“For example, on Blackwell chips, it can be about 4000 nanoseconds to go point to point.”
Wachen: Etched AI chip operates at under half the voltage of rivals
“We're now bringing forward our first generation of this low voltage inference technology Which is running at under half the voltage of any other AI chip.”
Wachen: Giant AI models will serve billions of users in a few years
“A few years from now, there's going to be giant models serving billions of users.”
Wachen: Only a few million people globally pay for AI model subscriptions
“We're very much in the early innings of AI today, where the paid plans, there's only a few million users in the world using paid plans of AI models.”
Wachen: Inference clusters will quickly scale to tens of thousands of chips
“And, you know, the inference side, you know, today people usually think about it as an eight chip cluster, or maybe just, you know, NVL as the scale up domain but very quickly, this is going to become thousands of chips and tens of thousands of chips.”
Wachen: Existing GPUs and TPUs are retrofitted pre-ChatGPT hardware
“Every GPU, every TPU, every AI chip that was serving these models were just, like, fundamentally built before this and are retrofit to serve these modern models.”
Wachen: Software COGS per incremental user will become high due to inference
“Like, the cost structure of every software company, the COGS is not going to be, like, zero anymore for an incremental user. It's going to be, like, quite high, and it's going to be a function of inference.”
Wachen: AI inference will become the biggest market in the world
“It feels like we are on a, like, decade march for inference to become, you know, the biggest market in the world.”
Uberti: High-speed AI decode is bottlenecked by data movement, not math
“When you do this sort of kernel's work, what you realize is that the math is relatively easy. But to get high speed decode, the thing that matters is data movement. Almost all the work that you do is optimizing how do you move data around a single chip or acro…”
Wachen: Etched is the only startup building custom racks and chips
“We're actually, I think we're the only startup right now that's building its own rack as well as its own chips.”
Uberti: Etched sent a dozen engineers to Bangalore to fix delays
“We went out and shipped a dozen of our top engineers across the world to Bangalore for six months. I was there as well. I lived in Bangalore for four and a half months personally”
Wachen: Over $1B in AI Compute Revenue Generated Every Day
“Every day there's over a billion dollars of revenue in this category, and a lot of it's inference.”
Wachen: Etched emulated full reticle chip on 700 FPGAs before silicon
“We took over 700 FPGAs and put the entire full reticle chip on an FPGA cluster and ran a dozen different models with our full inference stack on them before the chips came back.”
Wachen: Etched ran rack inference in 40 days versus rival's 10 months
“There is another very famous AI chip company that took 10 months to go from getting their silicon back to having them running inference in Iraq. And this was like publicly announced to their investors and it was a really big deal. We were able to do it in 40 d…”
Uberti: TSMC funded an experiment on Etched's recommendation and updated its line
“TSMC customer service is way, way better than I have seen at any other company in any other industry. It's the kind of thing where if you say, hey, you can approve your yield by making this change, you can go make them a recommendation, and then we'll go run a…”
Uberti: TSMC will continue to win due to superior customer service
“It is why they're the number one and why they're going to win.”
Wachen: Colossus charges $12/hour for Blackwells as the only 20k-cluster provider
“Why is Colossus charging 12 dollars an hour for Blackwells? It's because they're the only place you can buy 20,000 of them at once, right?”
Wachen: Etched hardware achieves 10x higher user concurrency at equal interactivity
“So when people are doing that evaluation, our hardware is going to generally be able to get you an order of magnitude more concurrency at a given level of interactivity.”
Wachen: AI agents will soon build complete web browsers in under an hour
“I thought it was really cool months ago when Cursor published that they had
a bunch of coding agents build an entire browser from scratch in a week.
Totally nuts.
And that, that will soon happen in under an hour.”
Uberti: Future AI token factories will cost $40B to $100B per mega-cluster
“You could have the same kind of thing for some futuristic mega cluster.
Forty billion dollars, hundred billion dollars as a giant mega token factory serving one or a handful of models for a massive number of users to get that same economies of scale thing.”
Uberti: The best kernels are written by human-AI collaboration
“Today, it's all very hybrid, and the best kernels are still written by human AI collaborations.”
Uberti: AI models will eventually do all kernel generation superhumanly
“And when the models keep getting smarter, they'll eventually do all of it. They will become superhuman.”
Wachen: Codex got GPT-OSS running overnight from Etched docs alone
“And we had Codex actually get GPT-OSS running from scratch just based off of our docs completely by itself. And it did it, I think, overnight.”
Wachen: Etched hired dozens from HFT over shared kernel-first philosophy
“And we've had dozens of people from high frequency trading join the team because they saw this philosophy too.”
Wachen: The Largest Token Producer Will Be the World's Most Valuable Company
“Whoever produces the most tokens is going to be the most valuable company in the world.”
Wachen: In-House AI Chip Failure Is Not Existential for Hyperscalers or OpenAI
“Meta won't fail if MTIA fails, Microsoft won't fail if Maya fails, and OpenAI won't fail if Jalapeno fails.”
Uberti: Etched aligned two clock signals to within 50 picoseconds
“We realized there was one and only one way to solve it. As we had to go line up two clock signals on our chip to within 50 picoseconds. That is literally 50 trillionths of a second. And we had to go get these signals aligned to this super small granularity and…”
Uberti: Employees quit Etched over deemed unsolvable chip design problem
“We had people quit. That people literally were like, this problem is unsolvable, and, ah, best of luck guys.”
Wachen: Semiconductor physical design vendor agreements cost at least $40M–$50M
“We needed to sign an agreement with a physical design vendor, which, you know, will cost you at least 40, fifty million dollars.”
Wachen: Historical semiconductor Series A rounds topped out at $40M–$50M
“You know, at the time, the biggest semiconductor fundraiser for a Series A was around, like, 40, fifty million dollars.”
Uberti: Synopsys allowed Etched multi-year deferred payment for emulators
“Synopsis actually went ahead and let us get some of their emulators on extremely favorable terms where we pay over many years.”
Uberti: TSMC VP Emailed Wanting to Partner with Etched After One Dinner
“And the following day, I get an email from TSMC saying, Gavin, I want to work with Etched. Find a way to make it happen.”
Wachen: Traditional semiconductor funds missed all AI chip companies
“I think, ah, a lot of the traditional semiconductor funds missed the entire AI chip, like, all the AI chip companies.”
Wachen: Modern EDA and FPGA tools changed chip tape-out failure rates
“I think it's very hard to realize the constraints have changed, and like, you know, when you've looked at tape outs for 20 years, and you've seen so many of them not work on the first try, or the second try, or the third try, like, you couldn't even run a work…”
Wachen: Serving global AI tokens requires three specific chip primitives
“Fundamentally, if you want to run a majority of tokens in the world, you need to do three things. You need to build a chip with the most flops, And given power budget, you need to build a chip that has the lowest latency between other chips. So the biggest sca…”
Wachen: Etched first-gen chip avoided Rubin supply chain by using 4nm
“And that's why actually for our first gen product, we built it on a different supply chain than the Rubens. So, you know, we're on four nanometer Rubens on three nanometer, you know, we're on a different HPM than Rubens and so forth.”
Uberti: Compute costs will drop faster than memory costs due to DRAM limits
“Generally, loading data is very expensive, and doing math is very cheap. And as time goes on, you'll end up finding that math gets cheaper at a rate that is faster than memory gets cheaper, due to this fundamental limit on any kind of DRAM device.”
Wachen: AI inference will eventually make up the majority of global GDP
“I firmly believe we are on a global march of inference becoming majority of global GDP. And it may take more than 10 years, but it's going to happen.”
Wachen: AI agents will outnumber human knowledge workers by 2027
“I think this is the second to last year where a majority of the workforce is going to be human. I think in 20, 27, you're going to see there's going to be more agents doing knowledge work than humans, and it's going to be extremely interesting to see what happ…”
Uberti: Individual trillion-dollar data centers are inevitable
“Absolutely. It is a matter of time. It's like asking, what do you say, a billion dollar fab, or a ten billion dollar fab, or a hundred billion dollar fab? It is inevitable that the economies of scale don't stop at, oh, forty billion dollars is the magic number…”
Wachen: Society Faces a Multi-Decade AI Token Supply Shortage
“We are in a new era of intelligence where the cost of producing intelligence is dramatically so much cheaper than the value of the intelligence that we are in a many year, probably many decade supply shortage of these tokens.”
Uberti: AI Token Serving Requires Non-Linear Cluster Scaling Economics
“That the way you want this to scale is not that, oh, if I want to go serve 10 times more tokens, I buy 10 times more servers. It must be some solution where if I want to serve 10 times more tokens, then I get some economies of scale benefit with my say cluster…”