Hard Look at AI's Massive Funding
This podcast episode covers major AI industry developments including Greg Brockman's claim that AI now writes 80% of OpenAI's code, Saudi Arabia's Humane launching an enterprise agentic OS on AWS, Morgan Stanley raising its hyperscaler capex forecast to $805 billion, and both OpenAI and Anthropic announcing private equity deals on the same day to accelerate enterprise AI adoption ahead of anticipated IPOs.
Summary
The episode opens with Greg Brockman's recent podcast claims that AI coding tools went from writing 20% to 80% of OpenAI's code in a single month — a 4x jump — and that OpenAI is 70-80% of the way to AGI, roughly two breakthroughs away. The host notes pushback from Yann LeCun on the AGI claim and Andrej Karpathy's more measured skepticism, pointing out that the metric measures tokens emitted rather than merge-ready commits a senior engineer would approve. The host shares his own vibe-coding experience, describing it as starting with a 'block of granite' that requires constant refinement, but argues the speed advantage is still transformative regardless of imperfections. He also plugs his personal vibe-coding project, AI Chat Daily, an automated news site that pulls podcast transcripts and scrapes external sources.
The episode then covers Saudi Arabia-backed Humane One launching on AWS as what it calls the first enterprise-grade agentic AI operating system, consisting of three layers: H2O (development SDK), Humane Code (agent orchestration), and Humane Fabric (data ingestion and governance). The host connects this to the broader trend of sovereign AI, referencing his interview with a Cohere co-founder about how countries and enterprises are seeking AI models they fully own and control on their own hardware.
Morgan Stanley's raised hyperscaler capex forecast of $805 billion for 2026 is discussed, up from $765 billion, covering Microsoft, Amazon, Google, Meta, and Oracle. The host highlights analyst projections that 2027 spending could reach $1.1 trillion — equivalent to all non-tech S&P 500 capex combined. David Sachs' claim that AI could account for 75% of US GDP growth draws both skepticism from economists like Tyler Cowen and supply-side concerns from Dylan Patel at SemiAnalysis about TSMC chip capacity. The host defends the ambition by invoking the SpaceX Starlink example, where expert critics were proven wrong by rapid technological execution.
The episode's centerpiece is the simultaneous announcement of OpenAI's $10 billion 'Deployment Company' joint venture and Anthropic's $1.5 billion deal with Blackstone, Goldman Sachs, Hellman & Friedman, and General Atlantic. Both strategies involve embedding developers inside portfolio companies of major PE firms to accelerate enterprise AI adoption — a Palantir-style approach. OpenAI's deal is more complex: it creates a standalone entity funded by TPG, Brookfield, Bain, Advent, and 15 others, with OpenAI contributing $1.5 billion. Notably, OpenAI is guaranteeing its PE partners a 17.5% annual return on investment. The host interprets both deals as primarily IPO preparation strategies designed to demonstrate broad enterprise revenue bases and diversified business structures.
Key Insights
- Greg Brockman claimed AI went from writing 20% to 80% of OpenAI's code in a single month, but Andrej Karpathy argues the meaningful metric is not tokens emitted but what fraction of that code would actually be approved as merge-ready commits by senior engineers.
- OpenAI is guaranteeing its private equity partners in the Deployment Company a 17.5% annual return on investment — a structured financial commitment the host notes is modeled after OpenAI's earlier revenue-sharing arrangement with Microsoft.
- Both OpenAI and Anthropic are partnering with private equity firms not just for capital but to gain instant access to thousands of mid-market portfolio companies, bypassing lengthy enterprise sales cycles by having PE owners mandate adoption across their holdings.
- Morgan Stanley projects that 2027 hyperscaler capex could reach $1.1 trillion — equal to the combined capex of all non-tech S&P 500 companies — while Dylan Patel at SemiAnalysis argues TSMC's current build-out cannot support that level of chip demand.
- The host argues that the sovereign AI category, where countries and enterprises own and operate their own AI models on private hardware with no vendor dependency, is becoming a major business segment — citing Cohere raising over $1 billion specifically to serve this market.
Topics
Full transcript available for MurmurCast members
Sign Up to Access