Descriptions:
In this episode of 20VC, host Harry Stebbings and investor Jason Calacanis cover three major AI industry stories from the week. The episode opens with Google DeepMind losing two landmark researchers within 48 hours: Noam Shazeer — co-author of the original attention paper and recently re-acquired by Google via the Character AI acquisition — and John Jumper, Nobel Prize winner and AlphaFold co-creator, both departing for Anthropic. The hosts debate whether this signals a structural talent environment problem at Google or simply the magnetism of research-focused settings where top scientists can work on whatever they want, backed by packages reportedly in the $500M–$2B range.
The second story centers on DeepSeek’s reported $7.4 billion Series A at a $50 billion valuation, with the notable catch that only Chinese entities receive voting rights. This triggers a broader argument about whether Chinese open-source AI releases are genuinely open or a form of subsidized geopolitical competition — with Calacanis arguing that “open source is a bit of a fake because China is paying for all the training.”
The third topic frames itself as the $725 billion question Wall Street is now asking: who will actually pay for AI? The hosts dissect OpenAI and Anthropic’s three-tier pricing reality — free users (subsidized), pro-sumer plans like the $200/month Claude Max offering $10,000 of inference (heavily subsidized for power users), and enterprise contracts (reportedly 40–70% gross margin on inference). The central argument is that open-source models are attacking the lucrative enterprise tier, and that by 2027 CIOs will shift from experimental token-maxing budgets to demanding measurable ROI.
📺 Source: 20VC with Harry Stebbings · Published June 25, 2026
🏷️ Format: Podcast
