2025-07-10

AI Is Making More Millionaires Than Anything in History w/ Salim Ismail & Dave Blundin | EP #181

Peter Diamandis, Dave Blundin, and Salim Ismail run their weekly WTF-just-happened-in-tech roundup: 36 AI unicorns minted in half a year with time-to-revenue collapsing 4x, and Link Exponential Ventures' Mercor bet illustrating just how fast young founders are compounding value. They debate where the smart money should go (chips vs. power vs. software layer), dig into Apple's reported move to power Siri with Anthropic or OpenAI, and revisit Salim's innovator's-dilemma thesis using Walmart's repeated failed attempts to beat Amazon internally. A chunk of the episode covers the AI talent war and eye-popping comp packages, a sharp disagreement over Roman Yampolskiy's 'we're cooked' superintelligence framing, and Salim pushing back on Vinod Khosla's 80%-of-jobs-by-2030 prediction. They close on breakthrough science (Neuralink's BCI roadmap, the Chai-2 molecular design model), humanoid robots, and Elon Musk's new 'America Party.'

▶ Watch on YouTube

Topics

Unicorn boom and collapsing time-to-revenue Economy ▶ 6:08
Dave Blundin walks through 36 new unicorns in half a year and data showing median time to $1M and $5M in annual revenue has compressed roughly 4x since 2020, calling it 'the opportunity of a lifetime.'
Vibe coding and vibe valuations Economy ▶ 12:49
Base-layer AI coding startups (Cursor, Lovable, Bolt) are hitting huge ARR with tiny teams, and investors are opening bids at $9-30B on Gen AI startups with little traditional revenue evidence — dubbed 'vibe valuations.'
Compute scale and the power bottleneck Compute ▶ 36:17
XAI's cluster already runs 340,000 Nvidia GPUs with a goal of 1 million by year-end; the panel argues the real constraint is no longer chips but electrical power, illustrated by XAI buying an overseas gas turbine plant to skip permitting.
Nvidia overtakes Apple in market cap Economy ▶ 42:47
Nvidia hits a $3.92T market cap, passing Apple, prompting a discussion of where investors should put money to ride the AI buildout curve (chips, power, real estate, or the software layer).
Corporate innovator's dilemma and edge disruption AI ▶ 28:02
Salim Ismail argues big companies structurally cannot disrupt themselves because they're optimized for efficiency and predictability; disruptive bets must be built at the edge (like Steve Jobs did with the Mac team) and only reintegrated after they hit critical mass, illustrated by Walmart's repeated failed attempts to beat Amazon.
Apple evaluating Anthropic or OpenAI for Siri AI ▶ 25:49
A Bloomberg clip reports Apple is considering powering the next Siri with Claude (Anthropic) or ChatGPT (OpenAI) instead of in-house models; insiders reportedly favor Anthropic, though Anthropic wants pricing that scales with revenue.
AI talent war and mega comp packages AI ▶ 54:37
Meta, OpenAI, and others are handing out $100M-plus and even billion-dollar offers for top AI researchers; Zuckerberg builds a 'Superintelligence Labs' dream team from OpenAI/Anthropic/DeepMind poaches while OpenAI's research chief calls it a break-in.
Defining and debating superintelligence safety AI ▶ 1:03:44
Prompted by a Roman Yampolskiy/Joe Rogan clip warning that superintelligence is an unstoppable adversarial force, Dave Blundin argues self-improvement can be bounded to algorithmic/hardware efficiency and fully logged, while Salim argues current systems aren't self-aware and panic is overstated.
AI and the jobs debate Economy ▶ 1:20:24
AI has already displaced 94,000 tech workers in H1 2025; Vinod Khosla predicts AI replaces 80% of jobs by 2030, but Salim disagrees, arguing (via Erik Brynjolfsson) that jobs decompose into dozens of tasks and only some get automated.
Product-market-fit collapse from LLM wrappers Economy ▶ 1:24:41
Chegg lost 90% of its market cap to ChatGPT; the panel reviews a list of companies at high disruption risk (Reddit, Quora, Wikipedia, Wolfram Alpha, Canva, banks, insurers) and argues regulation is the only thing currently protecting banks and insurers.
Neuralink's brain-computer interface roadmap Biotech ▶ 1:30:10
A Neuralink roadmap clip lays out speech-cortex decoding next quarter, tripled electrode counts and blind-sight trials in 2026, multi-implant capability in 2027, and 25,000+ channel implants with AI integration demos by 2028.
AI-driven molecular design breakthrough (Chai-2) Biotech ▶ 1:36:12
A Chai-2 promo clip claims the model can place atoms in 3D like 'Photoshop for molecules,' solving in hours what took a lab 3-4 years and $5-10M, with experimental validation in 2 weeks.
Humanoid robots and Elon Musk's America Party Robotics ▶ 1:38:34
Beijing hosts the first humanoid robot games and Agility Robotics will ride in Amazon delivery vans for last-100-feet delivery; the episode closes noting Elon Musk's proposed 'America Party' and the panel's concern over US debt-to-GDP nearing historical collapse thresholds.

Predictions made

open Dave Blundin: The rate of new AI unicorn creation keeps accelerating for at least a couple more years before AGI changes the picture entirely.
EP #? · · due: 2027 · ▶ watch
“we have at least a couple years of really good very rapid expansion of that number”
Your call:
open Salim Ismail: Hypergrowth AI startups will go through a wave of founder attrition and angst once they stabilize and lose the buzz that got them there.
EP #? · · due: unspecified · ▶ watch
“they're going to go through a lot of conions as they lose the buzz that got them there... you'll see a lot of angst, a lot of founders leaving”
Your call:
open Sam Altman (clip): GPT-5 will be released this summer.
EP #? · · due: 2025-09 · ▶ watch
“Probably sometime this summer.”
Your call:
open Elon Musk: XAI will scale its GPU cluster to a million GPUs.
EP #? · · due: 2025-12-31 · ▶ watch
“their goal what Elon has announced is a million GPUs by December 31st”
Your call:
open Peter Diamandis: The AI buildout will require roughly 100 gigawatts of new power capacity.
EP #? · · due: 2029 · ▶ watch
“we need a 100 gawatts by 2029”
Your call:
open Peter Diamandis: Global AI infrastructure spending, currently about $1 billion a day, will roughly triple to $1 trillion a year.
EP #? · · due: 2030 · ▶ watch
“we expect by 2030 it will triple to a trillion dollars a year”
Your call:
open Dave Blundin: AI models will be able to autonomously find and fix bugs in their own training code.
EP #? · · due: 2026 · ▶ watch
“that's certainly going to be the truth within a year”
Your call:
open Vinod Khosla: AI will replace 80% of jobs.
EP #? · · due: 2030 · ▶ watch
“AI will replace 80% of jobs by 2030”
Your call:
open Vinod Khosla: Humanoid robots will hit their 'ChatGPT moment' and become available for about $300/month.
EP #? · · due: 2027-2028 · ▶ watch
“humanoid robots are going to hit their chat GPT moment. In 2 to 3 years, they're available for you at 300 bucks a month”
Your call:
open Vinod Khosla: By 2040 people will work out of passion rather than necessity in an era of abundance.
EP #? · · due: 2040 · ▶ watch
“by 2040 people will work out of passion, not necessarily necessity in an era of abundance”
Your call:
open Salim Ismail: Retail banking will be substantially disrupted by decentralized finance.
EP #? · · due: 2028 · ▶ watch
“I would give it three years for retail banking”
Your call:
open Elon Musk (Neuralink roadmap, clip): Neuralink will triple electrode counts to 3,000 and run its first blind-sight participant trial in 2026, reach 10,000 channels with multiple simultaneous implants in 2027, and hit 25,000+ channels per implant with AI-integration demos by 2028.
EP #? · · due: 2028 · ▶ watch
“in 2028, our goal is to get to more than 25,000 channels per implant, have multiple of these, have ability to access any part of the brain”
Your call:
open Ray Kurzweil: High-bandwidth brain-computer interfaces will arrive.
EP #? · · due: mid-2030s · ▶ watch
“predicted by the mid 2030s we would have high bandwidth BCI”
Your call:

Numbers that matter

Worth digging into

🕳️ The GPT-4.5 training run bug
Dave Blundin claims a single code bug (possibly in PyTorch) silently wasted a large fraction of a multi-hundred-million-dollar training run, offered as the reason GPT-4.5 underwhelmed and why AI labs pay top researchers so much.
🕳️ Mercor's valuation trajectory
Going from a ~$30M seed to a rumored $8-10B term sheet in roughly two years is cited as one of the fastest value climbs in the current AI cycle and a case study for Link's investment thesis.
🕳️ Chai-2's molecular design claims
The clip claims Chai-2 solved in hours/weeks a protein-design problem that took a lab 3-4 years and $5-10 million, which would be a landmark result if verified.
🕳️ Walmart's four internal attempts to beat Amazon
Salim's story is his core evidentiary case for why disruptive innovation must happen at the edge of a big company, not the core — worth verifying against primary sources.
🕳️ XAI's overseas power plant purchase
Buying a fully built gas turbine plant abroad to skip US permitting is cited as a template for 'first principles' infrastructure speed, relevant to the broader AI power bottleneck story.