In this FSD episode, titled The Next Phase of the AI Trade, Ram records on the drive and opens on what he calls the next phase of the AI trade: the application and beneficiary layer. He points to Instinct, a general consumer app that raised $350 million at a $2 billion valuation, and his own note-taking app of choice, Granola, as early signs that value is moving up the stack. The last three years, he argues, were about infrastructure and frontier labs, a chapter that is closing as Anthropic heads public at a $2 trillion valuation with OpenAI to follow, leaving sixty-plus revenue-less unicorns chasing a boat that has already sailed. He traces how AI's usefulness has evolved in steps: GPT delivered novelty, Claude delivered actual work product and deliverables, which he considers more transformational than GPT, and now autonomous agents promise to sit on every customer for compliance, account management, and cross-sell. The obvious "AI assistant" startups, he warns, are about to be rug-pulled by Google Spark, which is why he prefers de-risked market leaders to chasing the next Poolside.
On positioning, Ram pushes back on Jordi Visser's move out of NVIDIA into longevity and Eli Lilly, arguing you are supposed to own NVIDIA precisely when it is unloved, complacent, and still posting strong numbers, especially once its forward valuation drops below the S&P 500. He explains the power of free cash flow yield as a stock-selection metric and why the perceived "round trip financing" in AI is actually senior secured positioning, not first-loss equity risk. On humanoids he is skeptical of the hype and would rather play the theme through Shield AI, a private drone company serving the U.S. government and Taiwan that is rumored to raise at a $20 billion valuation this fall, than pay up for Figure, Prometheus, or Tesla. The simplest humanoid bet, he says, is semiconductors and memory, and he adds Micron today after buying Western Digital yesterday. He frames NVIDIA's monster day and its re-guide from 40% to 70% growth on a $5 trillion company as water sloshing through the supply chain, where the most operating leverage sits in semis, memory, and edge computing. He also shares a health tip on clearing a lingering cough and notes that spending on GLPs and peptides now exceeds OpenAI and Anthropic combined.
The heart of the episode is his answer to the bubble fear. People pattern-match this AI build-out to the dot-com fiber glut of Exodus Communications and Global Crossing, or to the debt-fueled housing boom he watched firsthand meeting cab drivers who owned six houses. The difference, he argues, is that the market for intelligence always clears. Drawing on Say's Law, Jevons Paradox, and comparative advantage, he explains that intelligence is a general-purpose good that adds commercial value, so as its cost falls demand rises and it always gets consumed and generates a yield, unlike dark fiber or fertilizer. This is also why he believes AI tokens are becoming a form of money, since a token should price to the marginal value of intelligence at any moment, and in a competitive market to its marginal cost, though gross margins remain high for now, something he expects to scrutinize when the Anthropic S1 drops the day after Labor Day. He sees no glut today, only labor hoarding, H-1B constraints, and Google losing top engineers, and concludes you simply cannot get enough intelligence. He closes on the human side, debating AI use with his wife and using it to explain evolution and Humboldt to his kids, the September seasonal head-fake that fools people into recession fears, why it is feast or famine to be a VC right now, and why there has never been a better time to be a founder who maps each industry for AI beneficiaries, the way Brad Jacobs built and sold company after company.
(00:00) The next phase of the AI trade: the application layer
(01:35) Why the frontier-lab chapter is closing: Anthropic at $2T
(02:23) GPT to Claude to agents: how AI's value evolved
(03:37) AI agents on every customer, the phase still coming
(04:36) The beneficiary layer: Instinct, Granola, and consumer apps
(05:30) Consumer vs B2B SaaS, and why Lumida is consumer
(06:04) Vertical apps: Harvey, Google legal, and DocuSign
(06:49) Free advice: cut stock comp to lift your share price
(07:15) Why Jordi Visser is wrong to rotate out of NVIDIA
(08:12) Free cash flow yield and the "round trip" financing myth
(09:28) How to actually play humanoids: Shield AI, not Figure or Tesla
(10:56) The simplest humanoid bet: semis and memory (Micron, WDC)
(11:35) NVIDIA's monster day and the "bathtub" rotation
(12:27) From 40% to 70% growth: where the operating leverage is
(13:23) Health tip: fixing a lingering cough, plus Amgen and Repatha
(14:45) Why this is NOT a CapEx glut: dot-com, dark fiber, housing
(15:39) The key difference: intelligence always clears
(16:38) Say's Law, Jevons Paradox, and comparative advantage
(18:48) Tokens as money: pricing the marginal value of intelligence
(19:26) The Anthropic S1 drops the day after Labor Day
(20:06) No glut: labor hoarding, H-1B, and Google's brain drain
(21:14) Raising kids with AI, and the Humboldt story
(23:29) Consumer discretionary and the September head-fake
(24:42) Why it's feast or famine to be a VC right now
(26:03) Founders: map every industry for AI winners (Brad Jacobs)
About the show: Non-Consensus Investing is Ram Ahluwalia's running commentary on markets, where he shares how he's actually positioning capital and talks through the ideas most investors are missing. Real-time analysis, specific names, and a bias toward what's overlooked rather than what's crowded.