Sep 29, 2025 · 1h 31m · 20vc
Groq Founder, Jonathan Ross: OpenAI & Anthropic Will Build Their Own Chips & Will NVIDIA Hit $10TRN · 20VC with Harry Stebbings
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
Groq founder Jonathan Ross discusses the macroeconomics of the AI industry, explaining how compute, energy, and custom silicon dictate geopolitical power and enterprise success. He challenges industry assumptions about market bubbles, details Groq's unique supply-chain advantages, and shares a visionary outlook on the democratization of software development through AI.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 16.8% of the talking time here. How this is scored →
Harry as informed peer 3.9 Guest teaching 4.8 Guest disagreement 2.3 Harry pushing back 3.2
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 11:53 Dismissing latency acceptance as 100% wrong
Jonathan strongly rejects Harry's suggestion that users are content with inference latency, calling the opinion 'a hundred percent wrong' and citing historical internet engagement metrics.
Hardest push from Harry ▶ 28:21 Challenging guest on GPT-5 efficiency vs compute scale conflictHarry refuses Jonathan's framing that compute directly scales quality, pressing him on whether Sam Altman's focus on GPT-5 efficiency directly contradicts Jonathan's central thesis.
Biggest teaching moment ▶ 1:17:03 System-level SRAM vs DRAM economic breakdownJonathan corrects common market misunderstandings around SRAM pricing by breaking down system-level memory duplication, showing how GPUs consume 500x more memory copies.
Harry holds his own ▶ 1:16:29 Demonstrating architecture expertise with SRAM calloutHarry correctly identifies SRAM architecture as the underlying driver of Groq's multi-user serving capability, impressing Jonathan with his technical memory from their prior interview.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Analyzing the AI Market and the "Vibe Investing" Bubble | 3 | 4 | 3 | 4 | Harry asks if the AI market is in a bubble, and Jonathan immediately reframes the prompt, advising him to ask what smart money is doing instead. Harry pushes back on Jonathan's assertion that aggregate market returns exceed spend by highlighting Nvidia's extreme revenue concentration. | |
| Hyperscaler Capex and the Valuation Imperative | 4 | 5 | 3 | 5 | Harry presses Jonathan on hyperscaler capex, insisting that at some point financial returns must materialize regardless of Mag Seven status. Jonathan counters by explaining that capex spend is driven by existential career risk rather than pure financial ROI metrics. | |
| Analyzing the NVIDIA Loop and the Supply Bottleneck | 5 | 6 | 4 | 5 | Harry questions whether Nvidia investing $100B into OpenAI is an infinite money loop, calculating that 60% flows directly back into Nvidia's stock. Jonathan rejects the premise, explaining compute supply constraints and calling Harry's suggestion that users tolerate latency '100% wrong'. | |
| High Bandwidth Memory and the Strategic Value of Custom Chips | 3 | 6 | 2 | 3 | Harry asks if OpenAI and Anthropic can successfully move into custom chip production. Jonathan educates him on the difficulty of chip software and introduces Nvidia's monopsony over HBM memory supply. | |
| Capital Planning, Memory Shortages, and Data Center Costs | 3 | 6 | 3 | 3 | Harry asks if HBM shortages explain Sam Altman's call for hundreds of billions in capital. Jonathan directly shuts down the theory with a blunt 'No', explaining that data center amortization over 10 years dominates overall spend. | |
| Upgrading Silicon and the Economics of Legacy Hardware | 5 | 5 | 2 | 4 | Harry challenges accounting assumptions around 3-to-5-year chip amortization given rapid annual silicon iteration cycles. Jonathan breaks down the economic distinction between capex payback thresholds and opex retention thresholds. | |
| Groq's 6-Month Lead Time and the 18-Month Advantage | 2 | 4 | 1 | 1 | Jonathan outlines Groq's core pitch: delivering LPUs in 6 months versus Nvidia's 2-year advance checks. Harry reacts approvingly, framing it as an 18-month strategic chasm. | |
| Designing Models for Silicon and NVIDIA's Future | 5 | 6 | 3 | 5 | Harry challenges Jonathan's assertion that compute always equals performance, citing GPT-5's pivot toward efficiency over raw scale. Jonathan rejects the framing, explaining that models are tailored to hardware architecture and compute scale directly drives application quality. | |
| Training vs. Inference Costs and the Geopolitical AI Race | 4 | 6 | 3 | 3 | Jonathan debunks popular market claims regarding Chinese AI models like DeepSeek, revealing they are 10x more expensive to run than US open models. Harry questions whether CCP energy subsidies negate this cost penalty. | |
| Prompt Recycling and Building a Competitive Moat | 3 | 5 | 2 | 3 | Harry inquires about open-sourcing models to counter Chinese distillation techniques. Jonathan explains prompt compatibility lock-in and contrasts US aversion to errors of omission with European aversion to errors of commission. | |
| Japanese Decisiveness, Permitting Hurdles, and Geopolitical Power | 3 | 5 | 2 | 4 | Jonathan praises Japan's decisive execution on 2nm fabs and nuclear restarts. Harry pushes back on European adoption feasibility, questioning whether governments can deploy infrastructure at necessary speeds. | |
| The Tourist Economy Threat and Allied Energy Collaboration | 4 | 6 | 3 | 4 | Harry suggests sovereign European models like Mistral offer a compelling defense against US control. Jonathan dismisses model sovereignty without compute infrastructure as meaningless for long-term competition. | |
| The Shifting AI Market and the Virtuous Loop of Hardware | 4 | 5 | 3 | 3 | Harry asks if inference growth threatens Nvidia's dominance given GPU inefficiency for inference. Jonathan clarifies that inference growth creates a virtuous cycle that amplifies demand for training GPUs. | |
| Debunking Unemployment Myths and the Rise of Vibe Coders | 3 | 5 | 2 | 2 | Jonathan outlines a counterintuitive macroeconomic view where AI induces labor shortages rather than mass unemployment through deflationary wealth effects. Harry agrees, highlighting the gap between public perception and market realities. | |
| Margin Strategy, Volatility, and the Jevons Paradox | 5 | 5 | 2 | 3 | Harry cites startups like Lovable and Replit to question whether gross margins matter during exponential growth phases. Jonathan explains that high margins provide structural stability against market volatility. | |
| AI Cost Reduction and Solving Customer Problems | 4 | 4 | 1 | 2 | Harry reflects on his previous mistaken assumptions regarding Canva's AI costs. Jonathan explains that reducing customer friction dramatically expands the total addressable market. | |
| Economic Downturns, Predictions, and Market Stability | 4 | 4 | 2 | 3 | Harry asks about potential market pullbacks given S&P concentration in tech. Jonathan details economic reflexivity and laments that excess capital encourages top engineers to launch solo ventures instead of joining Groq. | |
| The Economics of the Tech Talent War | 4 | 4 | 2 | 3 | Harry asks which incumbents Jonathan worries about, noting Google's turnaround. Harry presses Jonathan on whether Google's window of distribution advantage is closing relative to OpenAI. | |
| OpenAI, Anthropic, Google, and the Dynamics of Coding Tools | 4 | 5 | 2 | 3 | Harry questions developer tool moats given low switching costs between AI coding assistants. Jonathan responds that enterprise contracts foster sticky adoption and asserts both OpenAI and Anthropic remain deeply undervalued. | |
| Groq's Recent $750M Fundraise and Margin Strategy | 5 | 5 | 2 | 4 | Harry challenges Groq's recent $750M fundraise, asking if it is sufficient capital and confronting Jonathan on past negative software margins. Jonathan clarifies hardware vs software unit economics. | |
| The Chip Market Future & NVIDIA's $10T Valuation | 4 | 4 | 2 | 2 | Harry prompts Jonathan for 5-year chip market predictions. Jonathan predicts Nvidia will maintain over 50% revenue share despite selling a minority of total physical chips. | |
| What the Market Misunderstands: SRAM vs. DRAM | 5 | 5 | 2 | 3 | Harry demonstrates technical knowledge from past interviews by identifying Groq's SRAM architecture. Jonathan breaks down system-level memory cost math, demonstrating why SRAM is cheaper at scale than DRAM. | |
| Groq's Stance on Going Public | 6 | 5 | 3 | 4 | Harry brings up Cerebras going public and asks rapid-fire questions citing Hamilton Helmer's Seven Powers. Jonathan corrects Harry on Cerebras pulling their IPO and refutes Cuda lock-in claims for inference. | |
| Concluding Thoughts: LLMs as the Telescope of the Mind | 1 | 1 | 0 | 0 | Jonathan offers a closing reflection comparing LLMs to Galileo's telescope, framing AI as expanding human consciousness. Harry expresses gratitude and concludes the episode. |
Statements from this episode (84)
Ross: Countries controlling compute will control AI
“The countries that control compute will control AI, and you cannot have compute without energy.”
Ross: Nvidia will be worth $10 trillion in five years
“I personally would be surprised if in five years NVIDIA wasn't worth 10 trillion, but I can't predict the outcome.”
Ross: OpenAI and Anthropic revenues would double with double compute
“If OpenAI were given twice the inference compute that they have today. If Anthropic was given twice the inference compute that they have today, within one month from now, their revenue would almost double.”
Ross: Microsoft withheld Azure GPUs because internal usage made more money
“Microsoft in one quarter deployed a bunch of GPUs, and then announced that they weren't going to make them available in Azure because they made more money using them themselves than renting them out.”
Ross: 35 to 36 companies generate 99% of AI token spend
“35 companies or 36 companies are responsible for 99% of the revenue or at least the token spend in AI right now.”
Ross: Total AI ROI will exceed capital invested despite individual failures
“Right now, people are making more money than they're spending. It's just very lumpy. As an aggregate. Plenty of people are gonna lose their shirts, but overall, less money is gonna go in than is gonna come out.”
Ross: Hyperscaler AI spending is driven by existential survival, not ROI
“That's how the hyperscalers feel. So of course they're going to be spending like drunken sailors because the alternative is that they're completely locked out of their business. So it's not a purely economical framework that they're using. It's a, do we get to…”
Ross: Big tech must spend on AI to protect Magnificent Seven valuations
“If you're not a member of the mag seven, You're not going to be able to get anywhere near the valuation. And so what do you do to stay there? You spend. And it's worth it because the stock value stays up because you're in the top seven or 10.”
Groq deployed a customer feature in four hours with zero human code
“Four hours later, it was in production. Not a single line of code was written by a human being. There was no debugging done by a human being. It was all prompting.”
Ross: Shipping features during customer meetings will win deals against competitors
“Qualitatively, when you can do that before the customer meeting is over, you're gonna be able to win deals that your competitors won't.”
Ross: Google ran three parallel chip projects, but only TPU succeeded
“So, people look at the TPU as a big success, and what they don't realize is that there were about three chip efforts at Google at the same time and only one of them ended up outperforming GPUs.”
Ross: Tesla's Dojo custom AI chip project was recently canceled
“And when you look around the industry, you've got a bunch of people building chips, some of them are getting canceled, like Dojo recently got canceled.”
Ross: Building custom AI chips to rival NVIDIA is like replicating Google
“Going off and saying, I'm gonna build my own AI chip to compete with NVIDIA, It's a little bit like saying, you know, that Google search is pretty nice. Let's go replicate it. It's insane. Like the level of optimization, the level of design and engineering tha…”
Ross: CPG profit margins correlate directly with response speed
“What is the number one thing that a high margin correlates to in CPG? It's the speed at which the ingredient acts on you. So that dopamine cycle, how quickly something occurs, determines your brand affinity. And so when something has a very quick response, you…”
Ross: Every 100ms speed improvement increases conversion by roughly 8%
“Every 100 milliseconds of speed up results in about an eight percent conversion rate.”
Ross: OpenAI, Anthropic, and all hyperscalers will build custom chips
“I have no doubt that OpenAI will be able to build its own chips. I have no doubt that eventually, Anthropic will be building their own chips, that every hyperscaler will build their own chip.”
Ross: NVIDIA effectively holds a monopsony on High Bandwidth Memory
“The thing is, Nvidia effectively has a monopsony on HBM.”
Ross: NVIDIA will produce ~5.5M GPUs this year despite 50M die capacity
“If Nvidia wanted to, they could build fifty million of those GPU die, Per year. But they're going to build about 5.5 million GPUs this year.”
Ross: Securing AI Memory Allocations Requires Paying Two Years in Advance
“The problem is you have to write that check more than two years in advance.”
Ross: Memory Suppliers Restrict HBM Supply to Protect High Profit Margins
“There's also this situation where the margin on HBM is so high, That no one wants to actually increase the supply, because then the margin goes down.”
Ross: Hyperscaler AI CapEx of $100B Annually Is Not Overspending
“So when you hear the hyperscalers talking about that, you know, seventy five billion to a hundred billion dollar a year investment, because they're building out the capacity for data centers, they're putting a lot of money up for returns that they're expecting…”
Ross: Groq plans to upgrade its AI chips annually
“We're looking at upgrading chips about once a year.”
Ross: Five-year AI chip amortization schedules do not make sense
“And in our case, we actually don't think that five years makes any sense.”
Ross: NVIDIA H100 GPUs remain highly profitable to operate despite age
“They're getting close to five years old. And they're still operating well, they're still earning more than their operating costs by quite a bit. You would never deploy an H 100 today, but they're still profitable to run, right?”
Ross: Groq LPUs Ship in 6 Months Versus 2 Years for GPUs
“You have to write a check two years in advance to get GPUs. For us, you write us a check for a million LPUs, and the first of those LPUs starts showing up six months later.”
Ross: Hardware incumbents win because AI models are optimized for existing GPUs
“So if you are the incumbent, you have an advantage because people are designing their models for your hardware. It doesn't even matter if there's a better architecture out there. It's not going to run well, so it's not a better architecture.”
Ross: NVIDIA will keep selling chips despite AI labs building custom silicon
“NVIDIA still keeps selling chips.”
Ross: Data center demand has exceeded projections for ten consecutive years
“For the last 10 years, infrastructure for data centers, you're planning that out two, three, four, five years in advance, right? And what happens is everyone, everyone's predictions are wrong. They end up building too little. This has just been what, what's ha…”
Ross: AI product quality scales directly with compute spent per query
“AI doesn't work the way SAS does. In SAS, you have a bunch of engineers who go out and build a product, and the quality of that product is determined based on what those engineers did. That's not the case in AI. In AI, I can improve the quality of my product b…”
Ross: Chinese AI models cost 10x more to run than US open-source
“The cost to run the OSS model is about one-tenth that of the Chinese models.”
Ross: China will build 150 nuclear reactors to power domestic AI compute
“They're going to build a 150 nuclear reactors. So they're going to have enough energy, even though their chips aren't as energy efficient.”
Ross: US will maintain AI hardware advantage over China for 2-3 years
“So my expectation is that right now for the next two to three years, the United States has a clear advantage in that away game over China.”
Ross: An AI model itself is not a clear competitive advantage
“I think the model itself is not a clear advantage.”
Ross: Anthropic should open-source older models to counter Chinese AI adoption
“I think that Anthropic should be open sourcing their previous generation in order to get people using them instead of the Chinese models.”
Ross: Prompt compatibility functions like software compatibility to create ecosystem moats
“And just like you have software compatibility, you have prompt compatibility.”
Ross: Missing out costs more than fumbling in growth economies
“There's two kinds of risk. There's mistakes of commission, where you do something that's a mistake, and then there's mistakes of omission, where you don't do something and it's a mistake. And the United States is terrified of making mistakes of omission. When …”
Ross: Norway could match total US energy output via wind and hydro
“Norway has about an 80% utilization rate of wind. So, like, 80% of the time you can be generating energy. They have enough hydro that if you deployed an five x the wind power of the hydro, Norway itself could provide as much energy as the United States and cou…”
Ross: Japan built a two-nanometer fab and is producing wafers
“Japan decided to build a two nanometer fab. When I was there last, they were showing off these two nanometer wafers that they produced. Now, the yield's not where it needs to be. This is not production grade, but they built a two nanometer fab, and they are pr…”
Ross: Japan allocated $65 billion toward AI initiatives
“They've allocated sixty five billion dollars for AI.”
Ross: Saudi Arabia will build 4 gigawatts of data center power
“They're going to build out three to four gigawatts in the very near future.”
Ross: US nuclear power permitting costs triple actual construction costs
“He said they spend three times as much on the permitting in the United States than on the nuclear power plant.”
Ross: Europe will become a tourist economy without rapid AI energy expansion
“Then Europe's economy is going to be a tourist economy. People are going to come here to see the quaint old buildings, and that's going to be it.”
Ross: 10x compute scale will outperform a 10x smarter AI model
“You could have a model that is 10 times smarter than OpenAI's model. And if you have 10 times the compute, OpenAI's model's gonna be better.”
Ross: NVIDIA Will Sell Every Single GPU It Builds
“NVIDIA's gonna sell every single GPU that they build.”
Ross: AI Training and Inference Form a Virtuous Hardware Demand Cycle
“The more inference you have, as mentioned before, the more you need to train the model to optimize for the inference. And the more training you have the more inference you want to deploy to optimize for the cost of that training, to amortize the cost of the tr…”
Stebbings: 10% of the global population uses ChatGPT weekly
“10% of the world's population is a GPT weekly active user.”
Ross: Compute is the most predictable lever for AI progress
“Compute is the easiest knob because it just keeps getting better and better and better every year. And if I write enough you know, if I write a check for enough money and I'm willing to wait a little while, I'm going to get more compute. It's the most predicta…”
Ross: There is no upper limit to usable AI compute
“There is no limit to the amount of compute that we can use.”
Jonathan Ross: AI will cause massive labor shortages, not mass unemployment
“I believe that AI is going to cause massive labor shortages. Yeah, I don't think we're gonna have enough people to fill all the jobs that are gonna be created.”
Ross: AI deflation will lead to earlier retirement and shorter workweeks
“And what that means is people will need to work less. And that's gonna lead you to number two, which is people are gonna opt out of the economy more. They're gonna work fewer hours, they're gonna work fewer days a week, and they're gonna work fewer years. They…”
Ross: Trump Administration Policies Are Definitely Helping Advance US AI
“Definitely help. All of the moves that have been made are things that are going to help with AI. For example you know, the permitting issues, right? Overall, it's been a very positive experience on AI.”
Ross: Coding will become a required skill for non-technical roles like marketing
“And coding is gonna become the same thing. For you to be in marketing, you're gonna have to be able to code. For you to be in customer service, you're gonna have to be code be able to code.”
Ross: High margins trade competitive moat for market stability
“The real reason why you need higher margins is volatility, because if you have a razor thin margin and the market moves, you may not be able to raise more money, you may not be able to get a loan, and so what a margin does is it gives you stability and staying…”
Ross: Groq will drive cash flow via volume while minimizing margins
“I want my margin to be as low as I possibly can make it while keeping my business stable, and I'm going to make my cash flow by increasing the volume.”
Harry Stebbings: AI implementation costs have dropped by 98%
“And it's just such a naive approach to ask that question even, because now the cost of implementation has gone down by 98%.”
Ross: PE firms are aggressively seeking cheap AI compute from Groq
“PE firms are all over us. They want access to cheap AI compute, because every time they get more cheap AI compute, they can bring the, they can change the bottom line of their businesses.”
Ross: AI compute expands labor supply, an unprecedented economic shift
“The most valuable thing in the economy is labor. And now we're going to be able to add more labor to the economy by producing more compute and better AI. That has never happened in the history of the economy before.”
Ross: Easy funding prevents AI startups from building critical talent mass
“Right now the biggest problem I see in AI is if you see a good engineer, one that you would have hired before, they can go out and they can raise 10, twenty, hundred million, a billion dollars, and then rather than contributing to one of the other AI startups,…”
Ross: Markets are only overheated if macro factors hinder corporate success
“In terms of whether or not the economy is overheated, I think one of the best predictors of that is, is the economy getting in the way of the success of the companies. If it's not getting in the way, then I don't think it's overheated.”
Ross: The tech talent war is more aggressive than ever in history
“It's definitely much more aggressive than it's ever been in history but only in tech.”
Ross: Gemini's integration into Gmail and Google products is practically unusable
“It's like, it's in Gmail, but it's practically unusable. It's in pretty much every product, and it seems thrown in, kind of, like, half thought through”
Ross: OpenAI is highly unlikely to fail or go away
“At this point, it would be hard to imagine a scenario where OpenAI goes away. I just, I don't see how that happens.”
Groq engineers recently switched from Anthropic tools to OpenAI Codex
“Our engineers recently started using codecs more than using the Anthropic tools.”
Groq mandates AI tool usage for engineers while leaving tool choice open
“We have a philosophy, we don't tell our engineers what tools to use. We do tell them you must use AI, because otherwise you're just not going to be competitive.”
Enterprise AI adoption is sticky due to long-term software contracts
“Enterprises make these long-term deals, and they stick with whatever their deal they made a year ago.”
Jonathan Ross: OpenAI and Anthropic Are Both Highly Undervalued
“I'd want to invest in both. They're both undervalued. Highly undervalued.”
Ross: Groq Will Never Create Its Own AI Models
“In our case we found an area where we will not compete with our customers, which is we will not create our own models. So we just won't do it.”
Groq raised $750 million at a valuation near $7 billion
“So we raised seven hundred fifty million. ... Yeah, almost seven billion.”
Groq originally planned to raise only $300 million in recent round
“It is, in fact we were only gonna raise three hundred million”
Groq operates with positive gross margins on hardware sales
“So we have when we sell hardware, those hardware units actually have positive margin.”
Ross: NVIDIA will retain 50%+ revenue share with 10% chip volume
“My prediction is that in five years, NVIDIA will still have over 50% of the revenue. However, they will have a minority of the chips sold. They might have, you know, minority share. They might have 51% of the revenue, and they might have 10% of the chips sold.”
Ross: Groq can produce more compute than anyone without supply constraints
“We don't have the same supply chain constraints. We can build more compute than anyone else in the world. The most finite resource right now, compute, the thing that people are bidding up and paying these high margins for, we can produce nearly unlimited quant…”
Jonathan Ross: Groq is currently operating 13 data centers globally
“We're now at 13 data centers. We have data centers in the United States, in Canada, in Europe in the Middle East.”
Ross: A customer recently requested 5x Groq's total chip capacity
“Like I said, last week someone came to us and asked for five times our total capacity.”
Ross: Groq's latest fundraise was 4x oversubscribed on double target
“I mean, with this fundraise, we ended up raising you know, more than twice what we were, you know, expecting to raise, and then we were forex oversubscribed over over what we did raise, and so we could have raised a lot more money, it would have been more dilu…”
Ross: Lowering AI chip prices 50% leads customers to buy double
“If we lower what we charge 50%, people are gonna buy twice as much. They're spending as much as they're making because whatever they spend increases the quality of the output.”
Ross: Groq is not currently preparing for an IPO
“Our focus is purely on execution right now. Whether or not you go public, you know, that's like that's a completely different game than we're playing right now. Right now, all that matters is can we satisfy the demand for compute?”
Ross: NVIDIA's software moat applies to training, not inference
“That NVIDIA's software is a moat. Yeah it's true for training, but it's not true for inference.”
Ross: Groq has reached 2.2 million registered developers
“I mean, we have 2.2 million developers on us now. That's how many have signed up.”
Ross: Starting a new AI chip company today is too late
“I wouldn't do chips. That, that chip has already sailed. It takes too long to build a chip.”
Ross: Amazon lacks AI DNA compared to Google, Meta, and Microsoft
“Amazon, I think doesn't have AI DNA. And they, like, if you compare them, so you didn't mention Meta, right? But Meta and Google always had the AI DNA, and Microsoft bought it with OpenAI, but that bought them time. Amazon still doesn't have that DNA, but they…”
Jonathan Ross: LLMs Are the Telescope of the Mind
“I think over time, we're gonna realize that LLMs are the telescope of the mind. That right now, they're making us feel really, really small. But in a hundred years, we're gonna realize that intelligence is more vast than we could have ever imagined, and we're …”
Shorts cut from this episode
▶ Nvidia worth $10T ?! · 20VC with Harry Stebbings (@0:19)
▶ Missing out is MORE Expensive · 20VC with Harry Stebbings (@35:52)
▶ Is AI a Bubble? · 20VC with Harry Stebbings (@4:04)
▶ Can Europe win the AI Race? · 20VC with Harry Stebbings (@0:00)
▶ The Future of AI Chips · 20VC with Harry Stebbings (@13:08)
▶ "The Countries that Control Compute Will Control AI" · 20VC (@0:00)