The Trillion Dollar AI Reset: Advertising & Gaming is Next w/ Salim Ismail & Dave Blundin | EP #170
Peter Diamandis runs the weekly WTF-just-happened news roundup with co-hosts Salim Ismail and Dave Blundin, covering AI's spread into classrooms, government lawmaking (UAE), and OpenAI's reversal on its for-profit conversion. The panel digs into AI-native startups like Blitzy writing millions of lines of code overnight, David Sacks' and Eric Schmidt's million-x compute and energy projections, and Mastercard's agentic-payments push that they argue will gut search advertising. They cover Nvidia's Gretel acquisition for synthetic data, the billionaire-funded epigenetic reprogramming wave (New Limit, Altos Labs, Retro), AI's convergence with gaming (Roblox, Mr. Beast's AI dubbing), Amazon's Kuiper satellites versus Starlink, SpaceX's Starship launch-rate request, a first real-world quantum teleportation test, a superconductivity research clue, and Michael Saylor's MicroStrategy as the live case study for Bitcoin as a corporate treasury asset.
AI disruption of K-12 and higher educationOther▶ 3:10
Peter cites China's and Estonia's mandatory AI-in-schools policies and a new US executive order; the panel argues schools are underreacting, AI already teaches better than most TAs and professors, and learning with AI is 2-4x faster than traditional methods.
AI writing government law and policy (UAE)Geopolitics▶ 8:02
UAE's project to use AI to write laws claims a 50% cost cut and 70% faster lawmaking; Salim argues UAE is underestimating the gains (closer to 5x), and the panel sketches a three-step path from AI navigating existing law to AI rewriting legislative structure from first principles.
OpenAI keeps a nonprofit-controlled, profit-capped ($100 return per $1 invested) structure instead of converting fully for-profit, after pressure from Elon Musk and state authorities; the panel reads it as risk management by Sam Altman more than a real change of direction.
AI-native startups and the changing value of a technical co-founderAI▶ 17:02
Dave describes Blitzy, a Link Ventures portfolio company that writes 3 million lines of code in a single night and jumped from a $2-3M to an $8M deal run-rate in a week; the panel discusses how AI shifts startup value from grind-it-out technical execution toward visionary product judgment and finding monetizable data.
Compute-bound AI research and the million-x scaling claimCompute▶ 20:49
Citing Zuckerberg, the panel notes AI research is now compute-bound rather than idea-bound; a David Sacks clip argues algorithms, chips, and data-center scale are each compounding 3-4x/year, multiplying to roughly a millionfold capability increase within four years, which Dave argues actually understates the algorithm/inference-time dimension.
An Eric Schmidt clip projects data centers will need 29 additional gigawatts of power by 2027 and 67 more by 2030; the panel argues solar has been cheaper than running fossil plants since 2019, so market economics (not policy) will push most new AI power toward renewables and remote siting (SMRs, stranded wind/solar).
Agentic commerce and the coming collapse of search advertisingAI▶ 38:04
Mastercard's new Agent Pay lets AI agents transact on a user's behalf; the panel argues this, combined with AI chat replacing search, will crush traditional advertising (Google's $300B ad business included) in favor of a trusted personal agent that shops and pays for you.
Crypto payment rails for cross-border tradeCrypto/Web3▶ 42:54
Salim argues Lightning-style crypto payments are about to scale fast, including for large cross-border transactions where countries restrict capital outflows (illustrated by GE once being paid for jet engines in frozen chickens to dodge Chinese capital controls).
Nvidia acquires Gretel for synthetic dataAI▶ 44:54
Nvidia's acquisition of synthetic-data company Gretel is framed as a defensive moat play: math and code are 'unbounded' domains because synthetic training data is nearly infinite and self-verifiable, while domains like biotech remain synthetic-data-constrained.
Epigenetic reprogramming and the billionaire longevity investment waveLongevity▶ 46:54
Kleiner Perkins led a $130M round into New Limit (Coinbase's Brian Armstrong and Blake Byers); the panel explains epigenetic reprogramming (resetting which genes are on/off rather than changing the genome itself) and notes Bezos/Milner (Altos Labs) and Altman (Retro) are independently betting on the same mechanism, enabled by AI's ability to process cellular data at scale.
AI's convergence with gaming and creator mediaOther▶ 50:50
Roblox now reaches 43% of Netflix's daily users (97.8M DAU, 7.2B hours); the panel expects AI-generated content and AI interaction inside games like Roblox/Minecraft to broaden gaming's appeal beyond young boys, citing Mr. Beast's AI-dubbed, lip-synced localization into every language as a preview of AI-driven global content.
The new space race: Kuiper, Starlink, and Starship's launch cadenceSpace▶ 53:42
Amazon launched its first 27 Project Kuiper satellites (on an Atlas V, not Bezos's own Blue Origin rocket) to compete with Starlink's roughly 8,000-9,000; separately SpaceX asked the FAA to raise Starship launch permits from 5 to 25 per year, with Peter arguing Starship is the only vehicle that will realistically get humans back to the Moon and on to Mars.
Quantum teleportation and a superconductivity research clueCompute▶ 1:00:53
A BBC report describes the first real-world quantum teleportation test over 30km of standard fiber carrying live internet traffic; separately, researchers found superconductivity varies at the atomic scale within iron-based materials, a clue toward practical room-temperature superconductors that could eliminate transmission loss in long-distance power lines.
Bitcoin as a sovereign and corporate reserve assetCrypto/Web3▶ 1:05:04
A Brian Armstrong clip argues Bitcoin is a better store of value than gold; the panel points to Michael Saylor's MicroStrategy (about 500,000 BTC, 2.8% of circulating supply, worth $52.8B against a $105.4B market cap) as proof that corporate treasuries and governments are being financially irresponsible by not holding it.
Predictions made
openDave Blundin: A highly intelligent large AI model (2-5 trillion parameters) paired with engaging, sympathetic voice interfaces will be out in the world.
“predicts that 95% of code will be AI generated by 2030”
Your call:
openDavid Sacks (clip): Compounding gains in algorithms, chips, and data-center scale (each roughly 3-4x per year) will multiply into about a millionfold increase in AI capability.
openDave Blundin: Actual AI scaling will outperform David Sacks' million-x-in-four-years figure, because both the chip and algorithm dimensions are being understated.
“the chips will outperform what he just said and the algorithms will way outperform what he just said”
Your call:
openDave Blundin: Fast, highly engaging conversational AI voice interfaces will be merged with frontier-scale (2-10 trillion parameter) reasoning models into a single seamless brainstorming partner.
“All of that will be stitched together within the next 3 months”
Your call:
openEric Schmidt (clip): AI data centers will require an additional 29 gigawatts of power beyond current demand, growing to 67 more gigawatts beyond that.
“Starship will get us back to the moon... Starship will get us to Mars”
Your call:
Numbers that matter
One computer science class can raise a student's future wages by about 8%Cited from an open letter where over 250 CEOs urged states to make AI and computer science a K-12 graduation requirement.
UAE's AI-lawmaking project targets a 50% cut in government cost and a 70% speed-up in lawmakingOfficial project goal; Salim Ismail argues the real speed-up will be closer to 5x.
OpenAI's capped-profit structure limits investor returns to $100 for every $1 investedPart of the nonprofit-controlled structure OpenAI kept instead of converting fully for-profit.
OpenAI last raised capital at a $300 billion valuationReferenced while discussing the scrapped for-profit conversion.
Up to 30% of one company's code is already written by AI; predicted to reach 95% by 2030Cited from Microsoft CEO Satya Nadella's public comments.
Blitzy writes about 3 million lines of code in a single nightLink Ventures-backed AI coding startup, using heavy inference-time compute.
David Sacks estimates AI algorithms, chips, and data-center scale are each improving 3-4x per year, compounding to roughly 10x every 2 yearsBasis for his millionfold-capability-in-four-years claim.
Solar passed fossil fuels on build cost in 2016, and by 2019 became cheaper to build-and-run than to merely operate an existing fossil fuel plantCited as why market economics alone now favor renewables for new AI data-center power.
Revenue per session from an AI chat dialogue is roughly 1/10th to 1/100th of a Google searchCited from Richard Socher's internal testing, discussed as a threat to Google's $300B ad business.
Roblox reaches up to 43% of Netflix's user base, with 97.8 million average daily active users and 7.2 billion hoursCited to illustrate gaming's rapid growth relative to traditional streaming media.
The first real-world quantum teleportation test succeeded over 30 kilometers of standard fiber optic cable carrying live internet trafficBBC-reported breakthrough discussed as a precursor to instantaneous, delay-free communication networks.
MicroStrategy holds about 500,000 Bitcoin (2.8% of all circulating Bitcoin), worth $52.8 billion, against a $105.4 billion market capCited as the live case study for corporate Bitcoin treasury strategy.
Worth digging into
🕳️ AI-written lawmaking at national scale (UAE)
UAE's project to have AI draft and deconflict law could become the template every government eventually needs, since legislation can no longer keep pace with AI-driven change on its own.
🕳️ The billionaire epigenetic reprogramming investment wave
Bezos/Milner (Altos Labs), Altman (Retro Biosciences), and Armstrong/Byers (New Limit) are independently betting large sums on the same age-reversal mechanism, suggesting real scientific conviction behind it.
🕳️ Debugging AI-generated code at Blitzy's scale
Writing 3 million lines of code overnight raises an unanswered question about how reliably that code can be verified and maintained at scale.
🕳️ Quantum teleportation over live commercial fiber
A successful 30km test on standard fiber carrying real internet traffic suggests this isn't purely lab-bound anymore, with implications for finance and satellite networking.
🕳️ MicroStrategy's Bitcoin balance-sheet strategy as a live case study
A company trading at roughly 2x the value of its Bitcoin holdings, run by a non-crypto-native CEO using leverage to keep buying, is a real-time experiment every CFO is implicitly watching.