3 Oct 2026EU Microsoft Teams | Video Turing Test | Mimetic Desire | Swarm Chasers
13 Aug 2026Lovable $400M | Corgi Investigation | Grok Bot | Investor Updates
Himanshu Dubey on OpenAI exits, Anthropic watermarks & Riemann
Guest Himanshu Dubey joins the studio to discuss Brad Lightcap's move into special projects at OpenAI and the wave of executive departures ahead of a potential IPO. He also explains Anthropic's AI watermark rollout in the context of the EU AI Act Article 50 and the 'Brussels effect,' and breaks down Anthropic's Riemann hypothesis project using 60 parallel agents with an unnamed model and heavy verification effort.
Azeem Azhar on China's AI Labs & the Open-Source Race
Azeem describes moving his OpenClaw agent from Claude Sonnet to DeepSeek to cut costs, but notes Sonnet's instruction-following still makes it noticeably sharper. He recounts his May trip to Chinese AI labs: the Minimax team's around-the-clock work culture, honesty about compute constraints, a culture where city mayors act as the real VCs (e.g., Wuhan for autonomous vehicles), and Z.ai's slides charting a path to ASI. He also reflects on how Chinese researchers' fluent 'scientific English' makes nuanced AGI/ASI discussions hard to translate.
Hyperscaler Capex, AI Revenues & SaaS Earnings
Azeem sizes the data-center buildout at $700-800B next year and $5-7T overall, noting hyperscaler AI revenues just exceeded annual depreciation ('from intensive care to high dependency') with revenues growing ~300% a year versus capex at 70-80%. He then explains why beaten-down SaaS stocks like Atlassian finally re-rated: analysts are lagging indicators, and MCP-driven per-transaction pricing eases 'SaaSpocalypse' fears (with a Datadog-style CFO cost-control caveat), while Lovable's $400M raise at a $13.3B valuation shows AI-native app builders disrupting traditional SaaS.
Unitree's IPO, China Robotics & the SpaceX Comparison
Azeem is skeptical of Unitree's 5,000x-oversubscribed consumer-driven IPO, preferring integrated AI-plus-hardware players like GalBot and Agibot, and arguing humanoids built separately from the underlying models are problematic. He compares the listing to a future SpaceX IPO, praising Elon's data-center relationships, giga-casting cost discipline and resilience, before the hosts thank him and plug his book 'Exponential: Order and Chaos in an Age of Accelerating Technology.'