Aug 2, 2026 · 1h 9m · 20vc
Arena CEO: There Will be a $100BN US Open-Source Model & Data is a Trillion Dollar Market
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 20VC, host Harry Stebbings interviews Anastasios Angelopoulos, Co-Founder and CEO of Arena, exploring the rapid evolution of open-source AI, US-China geopolitical competition, severe AI security threats, and Arena's milestone of surpassing $100 million in ARR.
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 30% of the talking time here. How this is scored →
Harry as informed peer 6.0 Guest teaching 5.0 Guest disagreement 3.0 Harry pushing back 4.3
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 47:14 Calling out venture investors on concentration risk
Anastasios aggressively dismisses VC orthodoxy, labeling investors 'total bitches' for worrying about revenue concentration when generational businesses like TSMC and Anduril prove otherwise.
Hardest push from Harry ▶ 28:48 Harry challenges margin disclosure price depressionHarry rejects Anastasios's claim that public IPO filings force price discounting, citing Palantir and luxury fashion pricing power to demonstrate true market leverage.
Biggest teaching moment ▶ 5:11 Deconstructing OpenRouter metricsAnastasios educates Harry on why top open-source rankings on OpenRouter represent a skewed sample caused by per-token margins rather than cannibalization of proprietary inference.
Harry holds his own ▶ 1:05:36 Explaining the South Korean memory crashHarry demonstrates superior public market domain knowledge, walking Anastasios through South Korea's market drop and the SK Hynix / Samsung bonus-taking dynamic.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Model Commoditization and the Breakthrough of Chinese Open Source | 4 | 6 | 2 | 2 | Harry asks whether the sheer volume of models signifies total commoditization. Anastasios explains how Chinese open-source models like Kimi K3 broke the US distillation narrative by beating frontier American models on front-end coding benchmarks. | |
| OpenRouter Metrics, Enterprise Adoption, and AI Sovereignty | 5 | 7 | 3 | 3 | Anastasios corrects Harry's interpretation of OpenRouter leaderboard data, pointing out that OpenRouter's token margin business model skews usage toward open-source failovers rather than dominant proprietary inference. He then introduces the strategic necessity of enterprise AI sovereignty. | |
| Building a $100B American Open-Source Company and Monetization Models | 6 | 4 | 3 | 6 | Harry pushes back on Anastasios's monetization model for open source, noting that Forward Deployed Engineering (FDE) and enterprise modernization are already core strategies for OpenAI, Anthropic, and Microsoft. | |
| Team Transience, Thinking Machines, and US vs. China AI Tailwinds | 6 | 5 | 3 | 4 | Harry cites co-founder departures at Thinking Machines and shares feedback from Chinese AI researchers regarding work ethic and state support. Anastasios counters by detailing US chip ecosystem advantages and hardware constraints facing China. | |
| Semiconductor Export Controls and National Security Strategy | 5 | 5 | 2 | 3 | The conversation covers the strategic trade-offs of semiconductor export controls. Anastasios frames the 2x2 matrix of market restriction versus hardware addiction and global mindshare. | |
| Model Backdoor Risks, AI Jailbreaks, and Frontier Lab Lobbying | 6 | 6 | 3 | 5 | Anastasios disabuses Harry of the notion that local hosting eliminates backdoor risks by explaining prompt sequence jailbreaks. Harry then argues that frontier labs will successfully lobby for protectionist restrictions due to political leverage. | |
| Nvidia's Open-Source Incentives and Arena's Strategic Dependency | 6 | 3 | 3 | 7 | Harry directly probes Arena's existential vulnerability, questioning whether the company has a viable business if the market consolidates into an oligopoly of two closed frontier labs. | |
| Enterprise Hesitation and the Top American Open-Source Contenders | 6 | 5 | 3 | 5 | They discuss enterprise fear of both Chinese models and frontier labs, listing US open contenders. Harry questions the routing layer's defensibility when multiple competitors and fintechs like Ramp launch their own routers. | |
| Inference Pricing Trajectories, Public Margins, and Pricing Power | 8 | 4 | 4 | 8 | When Anastasios argues that public IPO filings will reveal Anthropic's gross margins and create downward price pressure, Harry aggressively counters using Palantir's cost-plus evolution and luxury pricing power as counterexamples. | |
| The AI IPO Race: Anthropic vs. OpenAI | 5 | 6 | 2 | 3 | Anastasios highlights Anthropic's free cash flow advantage for IPO readiness and discusses the security breach requiring open-source defense, arguing for external AI guardian models. | |
| Critique of Government Model Licensing and Outcome-Based Regulation | 4 | 7 | 5 | 3 | Anastasios derides government pre-approval boards for model releases and shocks Harry by recounting how Arena routinely uncovers entirely synthetic AI candidates passing live technical interviews. | |
| The Silicon Valley Talent War, Neo-Labs, and Startup P&L Realities | 6 | 5 | 3 | 4 | Anastasios outlines the extreme compensation demands for elite researchers and predicts that two-thirds of the ~75 existing Neo-labs will be wiped out due to unsustainable P&Ls. | |
| AI Startup Valuations and the Downside Protection Calculation | 7 | 4 | 4 | 6 | Harry introduces the downside floor calculation made by VCs backing multi-billion dollar pre-revenue teams, citing Mistral and ElevenLabs as revenue-generating exceptions. | |
| Data as a Scaling Complement and a Trillion-Dollar Market | 6 | 6 | 2 | 2 | Anastasios explains data as an economic scaling complement to compute, arguing that data is more durable and less commoditized than hardware, predicting a $100B to $1T market by 2030. | |
| The Complexity of Data Sourcing and Venture Capital Revenue Concentration | 6 | 5 | 7 | 5 | Anastasios forcefully dismisses VC anxiety over revenue concentration, calling investors 'total bitches' and pointing to TSMC and Anduril. Harry banters back defending venture capitalists. | |
| Enterprise Data Expansion and Arena's Consumer Intelligence Flywheel | 6 | 6 | 3 | 4 | Anastasios reveals Arena's 30M+ monthly visitor flywheel for agentic evaluation, explaining how organic knowledge-worker traces allow them to evaluate performance across cost and latency. | |
| Margins in AI: Resellers vs. High-Value Intelligence Platforms | 7 | 5 | 3 | 6 | Harry presses Anastasios on Arena's monetization efficiency ($100M ARR against 30M prosumers) and examines low gross margins in token and GPU reselling businesses. | |
| Model Labs Expanding into the Application Layer | 8 | 4 | 4 | 7 | Harry identifies a contradiction in Anastasios's narrative regarding enterprise willingness to adopt frontier lab apps. He highlights that entrenched GTM and partner relationships protect vertical apps from generic lab models. | |
| Enterprise Entrenchment and Reevaluating the SaaS-pocalypse | 6 | 4 | 2 | 3 | Both agree that the 'SaaS-pocalypse' is overstated for deeply entrenched systems of record like Salesforce and ServiceNow, while surface-level tools remain vulnerable. | |
| Early Operational Mistakes and the Power of Deep Focus | 5 | 5 | 2 | 3 | In a quick-fire sequence, Anastasios reflects on the necessity of extreme operational focus and warns about compute debt insolvency risks if open-source models undercut frontier revenues. | |
| Overhyped Infrastructure Markets and South Korean Market Dynamics | 8 | 2 | 1 | 4 | Harry educates Anastasios on the South Korean market crash and semiconductor bonus cycles, arguing that physical mechanical infrastructure and cooling systems remain underhyped. | |
| Underrated AI Labs and the Future of AI in Medicine | 6 | 5 | 2 | 2 | Harry and Anastasios discuss Black Forest Labs, Demis Hassabis's vision for AI bio, and why data infrastructure rather than compute remains the central barrier to medical breakthroughs. |
Statements from this episode (45)
Angelopoulos: Kimi K3 Beat Top US Closed Models in Web Development
“And for the first time ever, we saw a couple of weeks ago that Kimi K three actually beat the best closed source American models on a, you know, pretty important subset of tasks, for example, front end coding, like web development, which a huge fraction of dev…”
Angelopoulos: Chinese AI Progress Involves More Than Distilling US Models
“That doesn't mean that they're not distilling. They may still be using distillation as a sub-step in their training procedure, But it does mean that distillation is only part of the story and that there's something that those labs are doing above and beyond di…”
Angelopoulos: Most global AI inference spend remains on proprietary first-party APIs
“If you look at the whole space of all inference, most of it is still being consumed on first party APIs and on proprietary models. That's why anthropic revenue has been just a total hockey stick. It's not, you know, it's not like they're being completely canni…”
Angelopoulos: Software will cease to be a business moat within five years
“Software is no longer really a moat because it can be produced instantaneously, right? Let's project out five years. That's what's going to happen. And so what moats exist? Network effects exist and data moats exist.”
Angelopoulos: US Will Have a Multi-Hundred Billion Dollar Open-Source AI Company
“I believe that we're going to have at least one massive, you know, multi-hundred billion, if not trillion dollar American company focused on American first open source.”
Angelopoulos: Forward-deployed engineering plus open-source models will outlast third-party APIs
“I do think the combination of FDE plus Open American model May be a more sustainable model for the future of American or even Western businesses, because it, because they might not want to be building on top of external third party services. They might want to…”
Angelopoulos: Thinking Machines' Inkling ranks #1 in US open source, behind nine Chinese models
“And within that time, they'd become the number one American open source model. But then the less generous take would be the companies existed for a year and a half. And then they've come up with, yes, the number one American open source, but there's nine Chine…”
Angelopoulos: Chinese AI labs face severe hardware constraints and rely on black-market chips
“So they're way hardware constrained over there. And they've been trying to like black market import chips because of this. And you see this in the news, right? The information just reported on this.”
Angelopoulos: Chip export controls risk incentivizing China to build an independent hardware ecosystem
“The downside of export control is that it can incentivize them to build their own ecosystem, and then what do we do?”
Angelopoulos: China Has Already Restricted American AI Models Domestically
“And by the way, it's worth noting that China has already restricted the use of American models within China, right? So if you look at the two by two matrix of US China restrict, not restrict, you know, like export import stuff They have already restricted the …”
Angelopoulos: China Unlikely to Ban Chinese AI Models in the US
“I don't really see them banning the use of Chinese models in the U.S. I don't think it makes sense for them.”
Angelopoulos: Local Hosting Does Not Eliminate Pre-Trained AI Model Backdoor Risks
“What if the other side that's interacting with the chat bot can, you know, build in a certain code word or a certain like character sequence that then jail breaks that model and gets it to reveal all the data to me. So it can sort of like vomit out all of the …”
Angelopoulos: US Likely to Restrict Chinese Open AI Models Within Three Years
“Yeah, my guess my guess would be that we will. I, I'm not saying I support it, but I think that it is likely where the world is headed. If I had to like place a bet, it would be there, but I think it's very uncertain at the moment.”
Stebbings: Sam Altman and Dario Amodei will successfully lobby for AI restrictions
“I just think Sam Altman is someone who I would never, ever bet against, and I think he's the best politician in the world, and I think when he says something, he says it with intent, and when he says we should give five percent away to the administration, he's…”
Angelopoulos: Open-source AI growth reduces Nvidia's revenue concentration
“Of course, Jensen is in some sense self-serving with this letter, because the more open source models are developed, the more companies are going to be training on GPUs. They're going to be fine tuning on their own data. And it's just more and more spend. It d…”
Angelopoulos: Enterprises fear relying on frontier AI labs and Chinese open-source
“It's not only true that they're terrified of working with the frontier labs, but they're also terrified of working with the Chinese open source.”
Angelopoulos reveals Arena uses Alibaba's Qwen in its tech stack
“And I said, you know, yeah, we use Quinn for X, Y, Z.”
Angelopoulos: Google's Gemma sits on the performance-versus-cost Pareto curve
“Gemma, by the way, is pretty good in terms of efficiency. If you look at arena, you'll see the, on the Pareto curves of like performance versus cost. Gemma's on there.”
Angelopoulos: Anthropic currently enjoys "disgustingly high" inference gross margins
“For example, one of the things that's going to happen is that like right now, Anthropic has like disgustingly high growth, gross margins in their inference.”
Angelopoulos: Going public will exert downward pricing pressure on Anthropic's inference
“And after they go public, the whole world is going to see that, right? Like we're going to see their margins because those are going to be public information. And that's going to exert downward pricing pressure on their inference.”
Angelopoulos: Enterprises Will Need Guardian AI Models Because Humans Are Too Slow
“So I think we need guardian models and also, you know, agents within our businesses.
What is a guardian model?
Something that can witness the traces basically that's looking over the shoulder of every agent within a business and then saying, okay, this is a sa…”
Angelopoulos: Fake AI applicants passed Arena's live technical engineering interviews
“They come in, they're like, hey, I want to be an infrastructure engineer at Arena, which is a great job that we're hiring for you. But then the other side of it is some guy that looks perfectly normal. They're getting, you know, they're passing all of our tech…”
Angelopoulos: Arena considering mandatory in-person onboarding to verify real humans
“We're gonna change our whole hiring process because of this kind of stuff. It absolutely worries me. Well, at first you need to verify that person is real. So all of our onboarding, we're considering at least making all of our onboarding in person because of t…”
Angelopoulos: Two-thirds of 75 AI neo-labs will fail or be acqui-hired
“There's at least 75 Neolabs. And For sure, like two thirds of those are going to be worth nothing or like they're going to be bought out for parts, right? That's going to be like an aqua hire.”
Angelopoulos: A $10B AI startup needs $4B revenue in 2-3 years to 10x
“And the thing that you really need to believe is that if the valuation is ten billion today, that you're going to generate the revenue, let's say it's a 30 X revenue multiple or 25 X revenue multiple to become a hundred billion dollar business. And so what tha…”
Stebbings: ElevenLabs has $800M in revenue and is raising at $22B
“I think 11 Labs is at eight hundred million in revenue. Raising it twenty two billion.”
Stebbings: AI data providers Handshake, Mercor, and Surge each surpass $1B revenue
“There's so many providers at a billion dollars plus in revenue. Handshake's over a billion. McCaw's over a billion. Serge is over a billion.”
Angelopoulos: Frontier AI labs spend 10% to 20% of GPU compute budgets on data
“Companies are spending on it, usually within Frontier Labs, at about 10 to 20% about the amount that they're spending on GPUs.”
Angelopoulos: AI data market will reach $100B to $1T by 2030
“I believe it's going to be at least a hundred billion dollars by 2030, if not a trillion.”
Angelopoulos: Data is the hardest part of model training as algorithms commoditize
“The data is really the hardest part of model training. Because you need to source it. It's so dirty. Nobody wants to do that shit. Nobody wants to hire all these people to generate data and then, you know, turn that into basically data plus GPUs equals model. …”
Angelopoulos: Silicon Valley investors are "total bitches" about revenue concentration
“The first is that I think that Silicon Valley investors have become total bitches with respect to revenue concentration. It's like, what the, what are you talking about? Like TSMC has revenue concentration. There's businesses that are like many hundreds of bil…”
Angelopoulos: Arena has 30M+ monthly visitors, bigger than Hugging Face and xAI
“People don't know this, but Arena's one of the largest consumer AI apps in the world. We're bigger than, like, XAI. We're bigger than, like, Hugging Face, and Manus, and GenSpark, where it's so massive, like if you, like outside in, it's like 30 plus million m…”
Angelopoulos: Arena has passed a $100M annualized revenue run rate
“So we're past a hundred million in annualized revenue run rate, and that's based on like Q, Q two times four.”
Stebbings: AI infrastructure companies operate at mid-thirties gross margins
“We're seeing a load of businesses like your fireworks of the world, where they're at the 30% style, mid thirties margin base. And that's very different to software margins that were 65 to 80.”
Stebbings: Claude Design is starting to eat away at Figma
“We see Claude Design has actually really started to eat away at Figma.”
Angelopoulos: Model labs must enter application layer to avoid commoditization
“If, like, inference is going to commoditize, then, of course, the next best thing is for the model providers to be moving up the application layer in order to own more of the application stack so that they ensure that they're not commoditized and they're getti…”
Angelopoulos: Harvey's CEO views model labs as his biggest competitive worry
“Harvey, the CEO of Harvey himself is saying that, you know, his biggest competitive worry is the model labs.”
Stebbings: GTM and complex deployment defend legal AI against model labs
“If you're Lagor or Harvey, dude, you've got to go into Cooley or Clifford Chance or any of the, build relationships with fifty-year-old white male partners who want to play golf and be told that they're great and that
You know, life is awesome. And then you go…”
Angelopoulos: The AI 'SaaS-pocalypse' is overstated due to data moats
“I think that the SaaS-pocalypse has been a little bit overstated overall. Because people don't understand always the dynamics of those businesses and how tough it is to replicate what they've built just also from a network perspective and a data perspective.”
Angelopoulos: Legal AI startups need not fear Anthropic due to differing priorities
“It's like priority number 12 for Anthropic is probably not high enough for Harvey and LaGuardia to be too scared.”
Angelopoulos: Nvidia is probably leading the race to a $10T valuation
“I think it's hard to say not NVIDIA. I think NVIDIA is probably in the lead there.”
Angelopoulos: Enterprise AI adoption will 10X Nvidia reliably
“But I think the enterprise adoption of AI is going to be another 10 Xer for the industry. I think it'll 10 X NVIDIA very reliably.”
Angelopoulos: Open-source AI pressure could cause insolvency for leading frontier labs
“And I think that the reason to be worried is because if the open source ecosystem somehow makes the cost saving opportunity for businesses much more salient and therefore decreases the revenue of companies like OpenAI and Anthropic within the enterprise, that …”
Angelopoulos: AI Will Eradicate Diseases Like Open Problems in Math
“I think that the, like, level of just human flourishing that's going to happen as we start to one by one eradicate diseases the same way that we're currently eradicating open problems in math is going to be incredible.”
Angelopoulos: Value in AI Biology Will Accrue to Data Layer
“That's exactly one of the areas where the data layer, where you can clearly see that the data layer is where value is going to accrue. Because the GPUs Are the same GPUs in both cases. The problem is that, that data infrastructure, the flywheel, the data colle…”