Arena's Angelopoulos: A trillion-dollar data market and America's coming open-source giant
Key takeaways from the 20VC interview with Arena co-founder and CEO Anastasios AngelopoulosDealroom has a profile for this one. Try Dealroom → (August 2026):
Arena measures AI in the real world. Angelopoulos positions Arena (formerly Chatbot Arena / LMArena) as the central evaluation platform for AI, ranking models on how they perform in the hands of real people rather than on static benchmarks — tracking factuality, steerability, hallucination and whether users actually get their jobs done.
An American open-source giant is coming. He argues at least one "multi-hundred-billion, if not trillion-dollar" US company will be built on American-first open source, naming Thinking Machines as a key contender, and says labs are finally cracking sustainable open-source business models via revenue-share and forward-deployed-engineer strategies.
Kimi K2 broke the distillation narrative. Angelopoulos calls it a big moment that China's Kimi K2 beat the best US closed-source models on an important subset of tasks such as front-end web development — proof, he says, that Chinese labs are doing more than just distilling American models.
Data is a trillion-dollar market. He frames data as a "scaling complement" — like petrol to cars — that grows with every new and larger model, with Frontier labs spending 10-20% of their GPU budget on it. He sees the data market reaching at least $100BN by 2030, potentially a trillion, citing $1BN-plus revenue players including Mercor, Handshake, Surge and Scale.
Enterprises will demand AI sovereignty. As software ceases to be a moat, Angelopoulos expects businesses to fine-tune open models on their own data to defend network-effect and data moats — wanting to own their stack for cost, security and supply-chain reasons rather than hand data to third parties.
Fake AI job applicants are a growing threat. Arena has interviewed candidates who pass technical rounds and sit in front of engineers but turn out not to exist — AI-generated personas he suspects are corporate espionage or nation-state attacks. Arena is considering fully in-person onboarding, echoing moves by firms such as Figma.
A brutal neolab shakeout is coming. With at least 75 neolabs raising multi-billion-dollar rounds on names alone, Angelopoulos expects two-thirds to be worth nothing or sold for parts, and warns that the "next round is a pit" for those without a sustainable revenue model.
Anthropic likely IPOs first. He predicts Anthropic — generating free cash flow and better prepared — could go public as soon as October, ahead of OpenAI, and that the eventual visibility of its inference margins will exert downward pricing pressure across the market.
Read more: 20VC
