Tech Leaders on Bitcoin, AI, and the New Global Power Shift w/ Salim Ismail & Dave Blundin | EP #172

Summary

• Bitcoin’s bullish setup is a balance-sheet supply squeeze, not merely a chart signal. Michael Saylor’s doubled “42/42” plan would fund $84 billion of Bitcoin purchases through $42 billion each of equity and debt, while the hosts say corporate buying already equals 3.3 times 2025’s new supply. Their sensitivity test: if 1% of the Fortune 1000 put 1% of treasury assets into Bitcoin, the price could reach $1 million; Diamandis therefore refuses to wait for dips because something like 10 trading days may deliver 70% of annual upside — “You can’t play that game.”
• Saudi Arabia is buying both AI capacity and a durable place inside the US technology orbit. Amazon and HUMAIN announced a $5 billion AI zone, while Nvidia committed 18,000 GB300 Blackwell chips in what the episode describes as a $10 billion commitment; Saudi Arabia wants to rank second only to the US in AI. The thesis extends beyond compute: local models must handle Arabic’s claimed 1,000 dialects and reflect regional values, while US innovation plus Middle Eastern capital could be “kind of unbeatable.”
• The speakers disagree on whether the AI positioning window will produce durable concentration. Diamandis and Blundin put AI dominance at the next two to four years and expect leadership to confer rule-setting power; Ismail expects OpenAI, Gemini and DeepSeek-style diffusion to level the field. Blundin sides with concentration because self-improvement compounds advantage, making the safest personal strategy to “run like hell toward your targets” and waste no time before 2030.
• AI’s public-market trade spans scarce compute, platform cannibalization and geopolitical chip-supply risk. The episode marks Nvidia at $3.3 trillion, Google at $2 trillion and OpenAI at $300 billion; Google may lose high-value search impressions even as YouTube, AI advertising and TPU7 strengthen. CoreWeave’s first post-IPO results showed more than 400% revenue growth, but its debt-financed Nvidia fleet leaves depreciation as the critical uncertainty, while the stack ultimately bottlenecks at TSMC and possible loss of Taiwanese chip access by December 2027.
• Software labor is moving from gradual productivity improvement toward wholesale restructuring. Engineers represented 40% of Microsoft’s cited layoffs, and Blundin expects almost any conventional white-collar coding or other job not to exist in its current form by 2030; Blitzy can already produce three million lines overnight and target legacy estates of 5 million to 20 million lines. With code costs down “a thousand or a millionx,” employers should fund serious AI mastery and humane transitions, not merely tell workers to experiment with Cursor.
• Specialization and self-improvement may matter more than another general-model release. Nvidia’s domain-adaptive pre-training example reportedly made Llama 2 an order of magnitude better at circuit design for only 1% more training cost, after distillation can shrink a model 10–100 times. DeepMind’s AlphaEvolve then improved Gemini training speed by 23% and GPU performance by 32.5% — early evidence for the “super-exponential” loop the speakers believe foundation-model companies are reluctant to discuss publicly.
• Falling birth rates turn robotics from a labor-saving option into national survival infrastructure. Replacement requires 2.1 children per woman, versus 0.72 in South Korea, 1.16 in China and 1.48 in Turkey; Blundin argues incentives remain far below the cost of raising a child. Against California’s $20 minimum wage, Diamandis projects a GPT-5-level, Grok-4-level humanoid at roughly $1 per hour including maintenance, electricity and insurance.
• Physical-world autonomy now has visible forcing functions, while Tesla’s recruiting is a key execution test. Archer is the official air-taxi provider for the LA28 Olympics, creating a deadline for manufacturing, infrastructure and FAA approvals; Tesla’s bull case, in the hosts’ framing, rests on robotaxis and Optimus rather than current car sales. Blundin says the decisive indicator is whether Musk can still recruit exceptional talent, while Diamandis calls Tesla “a monster opportunity to buy a stock that will never be lower than it is now.”

Deep dive

1. Bitcoin’s treasury bid is becoming its own supply shock

• Salim Ismail’s analysis of Michael Saylor’s doubled “42/42” plan centers on $42 billion of equity and $42 billion of debt for as much as $84 billion in Bitcoin purchases. He says the structure avoids interest payments: debt buyers can hedge by shorting the stock, while MicroStrategy escapes its former software cash-flow constraint.

• The reflexivity is explicit: Saylor urges corporations and sovereigns onto a Bitcoin standard; their purchases shrink an already limited float, lift Bitcoin, and strengthen his own balance sheet. With only 21 million Bitcoin, Diamandis argues there simply is “not enough to go around” if wealthy individuals, companies and governments allocate even small percentages.

• The hosts preserve two sensitivity tests: 1% of countries moving 0.1% of treasury assets could absorb all the Bitcoin in the world, while 1% of the Fortune 1000 putting 1% of treasury assets into it could drive Bitcoin to $1 million. Corporate purchases, they add, already outweigh 2025’s new Bitcoin supply by 3.3 times.

• Coinbase entering the S&P 500, a possible path for crypto exposure in 401(k)s and copycat treasury companies signal legitimacy, even as some wealth managers still prohibit advisers from recommending Bitcoin. Diamandis, explicitly describing his own behavior rather than giving advice, buys Bitcoin and MicroStrategy without timing dips: something like 10 days may deliver 70% of annual upside, so missing them can forfeit much of the opportunity.

2. Saudi capital is binding itself to the US AI stack

• Amazon and Saudi Arabia’s newly created AI company HUMAIN announced a collective $5 billion investment in an AI zone, while Nvidia agreed to sell the kingdom 18,000 GB300 Blackwell chips. Diamandis characterizes the latter as a $10 billion commitment and notes that countries worldwide are “frothing” for access to those top-end chips. HUMAIN is owned by Saudi Arabia’s Public Investment Fund and is developing AI models and data-center infrastructure.

• The compute purchase is part of a broader strategy: Saudi Arabia already holds investments in CoreWeave and Cerebras, wants to rank second only to the US in AI, and may ultimately commit “multiple trillions.” Diamandis sees the deals placing it firmly in the US camp rather than China’s, across infrastructure, partnerships and capital flows.

• Blundin argues localization is strategic, not cosmetic. The Liquid AI team told the hosts that Arabic has roughly 1,000 dialects that US foundation models do not adequately cover, while a generation under 20 will derive much of its worldview from AI conversations; regional governments therefore want control over language, cultural values and what those systems tell their children.

• Minister Abdullah Al-Swaha illustrated the state’s ambition with five-minute passports, AI-assisted CRISPR work aimed at sickle-cell disease and bad cholesterol, and a claimed reduction in treatment cost from $2.2 million to below $100,000. Ismail’s macro conclusion: US innovation combined with Middle Eastern capital makes the emerging bloc “kind of unbeatable.”

3. The AI hierarchy may lock in before democratization catches up

• Diamandis puts the decisive “AI dominance” window at two to four years. After that, he argues, first, second and third place become difficult to dislodge because self-coding systems can generate a runaway advantage; whoever leads “is going to make the rules” for the rest of the market.

• Ismail’s counterpoint — worth keeping — is that model capabilities have diffused quickly from OpenAI to Gemini to DeepSeek, so leadership may level out rather than become winner-take-all. He separates two claims: moving quickly unquestionably creates exponential opportunity, but he remains unconvinced that one bloc captures all enduring value.

• Blundin sides with concentration because self-improvement reinforces incumbent scale and no comparably strong democratizing force is yet visible. His advice to students and founders is unusually blunt: “The forces of concentration are much stronger than the forces of distribution,” so “run like hell toward your targets” during the roughly four-year window ending around 2030.

4. Google, OpenAI and Nvidia are converging on the same profit pools

• Mark Andreessen’s framing captures the platform collision: “OpenAI is growing up to become Google,” entering search, while Google is reinventing itself to become OpenAI and must go “Gemini first.” Ismail considers Gemini probably the best general-use model, yet observes that OpenAI continues winning adoption and mindshare among children.

• The valuation spread sharpens the strategic question: Google at roughly $2 trillion, OpenAI at $300 billion and Nvidia at $3.3 trillion. Nvidia was therefore worth almost twice Google in the figures cited. Google still has search, YouTube, GCP and proprietary TPUs as strategic assets.

• Blundin sees Google’s core search economics deteriorating because AI answers produce less value per impression, even if Google captures the new behavior. The offsets are YouTube, synthetic video, AI-optimized advertising and TPU7, whose weight-stationary systolic-array design may run core AI workloads more efficiently than Nvidia hardware and remain sold out for Google Cloud’s internal demand.

• The entire competition still bottlenecks at TSMC. Blundin cites a planning assumption of no access to Taiwanese chips by December 2027, while China wants to control Taiwan by 2028 and US fabs would not be ready in time; the speakers expect Taiwan’s fabs to be disabled or destroyed rather than transferred intact.

5. Institutions are beginning to absorb AI faster than their rules can adapt

• The FDA plans to deploy a generative-AI tool across all centers by June 30 after completing its first AI-assisted scientific-review pilot. Diamandis sees a direct attack on multiyear therapy approvals, while acknowledging the structural asymmetry: regulators are punished for approving a harmful drug, but rarely for the lives lost while a useful drug remains unapproved.

• Blundin wants durable reform through 50 state-level “shots on goal,” letting entrepreneurs relocate when one jurisdiction blocks experimentation. Diamandis offers Florida’s expected authorization of stem-cell treatments as the specimen: patients could remain near capable physicians, generate cleaner data and avoid traveling to Caribbean or Latin American clinics.

• A courtroom’s AI-generated victim-impact video surprised Blundin because it was allowed, though he called the use forward-thinking and potentially legitimate. Diamandis supplies the limiting case: once prosecutors use AI-generated testimony, defendants can answer with their own, turning proceedings into “deepfake versus deepfake” and opening “Pandora’s box like crazy.”

• Pope Leo XIV’s early warning about AI’s effects on dignity, justice and labor matters to the speakers less as doctrine than as an adoption signal. Blundin calls public underreaction “rampant”; a leader reaching roughly a billion Catholics may force attention well beyond technology circles.

6. CoreWeave is testing whether compute scarcity can reopen IPO markets

• CoreWeave was described as above $70, roughly 40–50% over its IPO price, after reporting more than 400% revenue growth in its first public earnings. Blundin treats the result as a bellwether: a failure could have frozen issuance again, while a successful offering tells Wall Street that AI companies are “open for business.”

• The controversial mechanism is leverage. CoreWeave borrowed heavily to buy Nvidia chips, financed data centers against those assets and counts Nvidia as an investor; nobody knows how quickly the hardware will depreciate. Blundin’s counter-thesis is enduring compute scarcity: “Anything that can compute” may be continuously occupied by AI “forever hereafter.”

• The next IPO candidates look less fragile to him because they are mostly chip or vertical-application companies, without CoreWeave’s same debt risk. Once another five, six or seven offerings clear successfully, he expects “the train” to be rolling again for venture-backed exits.

7. Coding automation is forcing a workforce redesign, not a tool upgrade

• Engineers constituted 40% of the cited Microsoft layoffs, an early waypoint on Blundin’s line from today to 2030, when he expects almost every conventional white-collar coding or other job to disappear in its current form. His frustration is that companies test an AI system, find bugs and miss the relevant variable: its rate of improvement.

• Blitzy can reportedly write three million lines of code overnight and currently targets legacy estates of 5 million, 10 million or 20 million lines. Its nearer-term opportunity is rebuilding those products for cloud deployment; looking three, six or nine months ahead, Blundin expects greenfield software that once looked unaffordable to be produced in “a night or a week.”

• State Street supplies the economic example: its mutual-fund NAV accounting reportedly runs on PL/I mainframes costing $300 million annually to maintain, yet a conventional rewrite would cost $1 billion and damage one executive’s bonus cycle. Reduce that rewrite by 10–100 times and the organizational blocker vanishes.

• The response cannot stop at “play with Cursor.” Blundin wants companies to free half to two-thirds of high-potential employees’ time for genuine AI mastery and provide humane transitions for others; one client gave displaced call-center staff a one-year internal UBI, after which roughly 95% reportedly found work quickly. Diamandis’s adoption warning is that humans “would much rather be comfortable than happy.”

8. Domain adaptation and self-coding are bending the improvement curve upward

• Nvidia’s domain-adaptive pre-training work began with Llama 2 and circuit design. By retraining the open model on specialized circuits, it reportedly became an order of magnitude better at the task for only a 1% increase in total training cost — a striking contrast with startups that merely add wrappers around general models.

• The missing step is distillation: discard capabilities irrelevant to the domain, shrink the system by 10–100 times without losing its useful expertise, then rebuild it with specialized data. Blundin says Nvidia published a practical “cookbook,” but later went quiet; his inference is that the method worked well enough to become an internal chip-design advantage.

• DeepMind’s AlphaEvolve combines Gemini with evaluators to discover higher-performing algorithms. The reported gains were a 23% improvement in Gemini training speed and 32.5% better GPU performance, which Diamandis describes as the beginning of self-referential improvement rather than an isolated coding benchmark.

• Sam Altman reportedly called coding “very, very special” because OpenAI uses it internally, then stopped elaborating. Blundin believes foundation-model companies avoid discussing self-improvement because it attracts dystopian criticism and regulators; Diamandis sees the curve moving from 2x every two years, to 10x and 100x annually, toward something “super-exponential.”

9. AI is moving from manipulating bits to designing atoms

• Meta’s Open Molecules 2025 supplies large-scale atomic-system simulations for molecular discovery, while its Universal Model for Atoms predicts interactions across different materials. Diamandis’s preferred metaphor is “atomic Lego sets”: models could discover new drugs and materials for energy, environmental technologies and space exploration.

• Ismail’s through-line is the feedback loop from atoms to bits and back to atoms. AI can accelerate candidate discovery, but the system still needs a reliable physical simulator; Blundin therefore sees particular leverage where quantum computing can simulate biological systems while AI proposes and evaluates designs.

• The practical bridge may be an AI quantum compiler that decides whether a problem suits quantum hardware and translates natural-language intent into quantum code running in the cloud. The stranger framing remains memorable: the hosts recount a consensus view among quantum physicists that computation may occur in “parallel universes” before bringing the answer back, potentially making the computer evidence for a multiverse.

10. LA28 gives electric air taxis a real regulatory deadline

• Archer’s appointment as official air-taxi provider for the LA28 Olympic and Paralympic Games creates a forcing function for manufacturing, operations and FAA approval. The US field also includes Joby and BETA; Diamandis hopes Santa Monica Airport can become a vertiport connecting Malibu, downtown Los Angeles and Santa Barbara without the 405.

• The speakers distinguish perceived and practical barriers. Independent electric motors and AI-controlled routing could make these aircraft safer than helicopters, but they still require specialized gates and facilities. Blundin says the safety case is already promising; Diamandis argues there are “absolutely no barriers” to the technology and that the remaining obstacles are regulatory.

• Diamandis sees geographic arbitrage as the investment implication: inaccessible islands and mountainside land become useful once vertical flight removes the road requirement. The Olympic deadline matters because it concentrates the manpower needed to turn that possibility into an operating network.

11. The healthspan prize is converting one purse into billions of research

• The episode’s main discussion describes XPRIZE’s healthspan competition as a $101 million prize, while its opening refers to $111 million. The competition asks teams to add at least 10 years — ideally 20 — of immune, cognitive and muscular health. From 623 registered teams, 40 received $250,000 milestone awards, distributing $10 million; the opening instead refers to $12 million awarded.

• Diamandis describes the prize model as a force multiplier: approximately $550 million in cumulative prize purses has induced close to $10 billion of team-funded R&D over 30 years, now approaching 30 times leverage. He credits the healthspan competition’s scientific structure to executive director Dr. Jamie Justice and its advisory board.

• The target is not simply more years alive. US life expectancy is cited at 79 while healthy life expectancy is 63, leaving a 16-year period of declining health; the competition aims to extend the healthy period and compress that gap.

12. Underpopulation is making AI and robotics a national necessity

• Replacement fertility requires about 2.1 children per woman, versus Turkey’s 1.48, South Korea’s 0.72, Hong Kong’s 0.8, Singapore and Puerto Rico near 0.9, China’s 1.16, Spain’s 1.19 and Italy’s 1.2. Diamandis rejects the old overpopulation framing: “South Korea is sublimating. It’s vaporizing.”

• Blundin notes that government incentives do not approach the actual cost of raising a child and have generated little response. He interprets China’s rapid AI and robotics investment less as conquest than survival under a demographic shock like Japan’s, while India needs the same technologies to scale education and healthcare across 1.41 billion people.

• The labor arithmetic is stark: California’s minimum wage is cited at $20 per hour, versus a projected humanoid operating cost of $0.40 before maintenance and roughly $1 including electricity and insurance. Diamandis imagines that machine carrying GPT-5-level and Grok-4-level intelligence while operating continuously.

• Musk’s Saudi presentation extends the endpoint to tens of billions of humanoids and an economy ten times today’s size, producing “universal high income” rather than basic income. Diamandis supplies the unresolved caveat: conscious robots assigned all human drudgery may eventually demand equal rights and equal pay.

13. Tesla’s valuation rests on Musk’s ability to keep converting talent into autonomy

• In Saudi Arabia, Musk promoted autonomous vehicles and thanked the kingdom for approving Starlink for maritime and aviation use; Diamandis would not be surprised to see Optimus manufactured there. Starlink had 7,135 satellites in orbit, versus FCC approval for 12,000 Gen 1 and 7,500 Gen 2 satellites and a proposed constellation of 42,000.

• Diamandis argues Tesla’s value lies in robotaxis and Optimus rather than current vehicle sales. Ismail compares Tesla’s test to Apple’s post-Jobs decline and return: if Musk can attract exceptional talent and inspiration, Tesla can become a great robotics company; if recruiting suffers, “then it’s a car company.” Blundin likewise treats Musk’s ability to recruit as the practical bellwether.

• Diamandis says Musk’s history across Tesla, SpaceX, Starlink and xAI teaches investors “don’t ever bet against Elon,” and calls Tesla “a monster opportunity to buy a stock that will never be lower than it is now.” He also sees each Tesla as mobile energy storage that can integrate with solar, Starlink and humanoid systems.

• Ismail presents Musk’s use of AI assistants and trusted operational integrators across five major companies plus DOGE as a case study in scale. Blundin says Musk spends roughly one day per week at each company asking what problem he can solve — potentially resolving a problem a week, or 50 a year, per company. The Boring Company supplies the closing specimen: its machine continuously mined and erected 24,000-pound tunnel rings with zero people inside the tunnel.