Tech Experts Break Down the Incoming AI-Crypto Collision That Will Redefine Global Power
Summary
Washington’s crypto turn could remake the financial plumbing beneath a multitrillion-dollar asset base. The White House plan, GENIUS Act, 401(k) order, and SEC position on liquid staking collectively move policy from “regulation by enforcement” toward dollar-backed stablecoins, instant settlement, tokenized collateral, and yield-bearing crypto. Eric Pulier calls it potentially “the most significant economic legislation and changes that we’ve seen in our lifetimes,” while Dave Blundin warns that programmable securitization also makes almost anything shortable—and therefore easier to weaponize.
Tokenization’s larger unlock is not crypto exposure but programmable ownership. A $20 million apartment could become 1 million $20 tokens; a $30 million house could become verifiable collateral in seconds; and loyalty points, equities, real estate, gold, and local tokens could carry access rights, rewards, and yield. Eric Pulier says the cited $600 billion of tokenized real-world assets by 2030 is far too small; Peter Diamandis contrasts it with roughly $120 trillion of real estate, $100 trillion of equities, $13 trillion of Treasuries, and $12 trillion of gold and precious metals: “It’s all coming.”
Giving AI agents access to crypto’s native payment layer is the episode’s central acceleration thesis. Agents will research, select, purchase, and settle without SWIFT, three-day delays, or $2 transaction fees; websites and brands will increasingly optimize for machine decision-makers rather than human persuasion. Pulier’s blunt framing is that “websites will be for AIs, not for humans,” and he expects an “explosion in the economy” when AI and crypto work together.
The chip race is becoming industrial policy, geopolitical leverage, and a concentrated Intel turnaround bet. Blundin says AI remains gated by fabrication capacity, with TSMC holding 66% of advanced-chip share and the administration proposing a 15% toll on approved prior-generation Nvidia and AMD exports to China. Intel’s 1.8-nanometer yields, underfunded 1.4-nanometer program, and paused Ohio construction make state support likely; Leopold Aschenbrenner’s hedge fund reportedly put 46% of its disclosed holdings into Intel, embodying the call that America “absolutely desperately” needs the company.
Retirement access and liquid staking could pull substantial new demand into crypto, but execution is not automatic. The $8 trillion-$12 trillion 401(k) market still needs SEC guidance, while one informal comparison across four AI engines put Bitcoin at $1 million between 2028 and 2031—an anecdotal model consensus, not evidence. Liquid staking could let tradable receipts and wrapped or tokenized assets earn DeFi yield, but Blundin invokes the Big Short era’s $60 trillion of synthetic instruments atop a $20 trillion housing market: “It’s not going to be trivial to make this real.”
AI is simultaneously commoditizing intelligence, reviewing regulation, and threatening to make unaided public investors “exit liquidity.” Grok 4 went free under pressure from GPT-5 and rivals, while an AI-led federal review targets 200,000 regulations and a 50% reduction by 2026; HUD reportedly removed 1,083 rules in two weeks versus an estimated 3.6 million human hours for the broader review. In markets, a six-month ChatGPT micro-cap experiment returned 23.8%, but Dave Blundin’s warning is harsher: tiny differences in institutional AI quality could create an overwhelming information advantage.
Physical-world abundance still bottlenecks on power, even as the panel disputes whether energy decides the AI race. Helion targets a 50-megawatt Microsoft-linked plant by the end of 2028, Commonwealth Fusion has a 200-megawatt Google agreement for the early 2030s, and China added $2.1 billion to fusion. Diamandis repeats that AI is “electron limited,” but Blundin argues chips and algorithms remain the decisive 10x-to-100x multipliers because AI buyers can outbid residential and industrial users for electricity by 5x or 10x.
Deep dive
1. Intelligence is racing toward free while agents become the web’s real audience
Dave Blundin reads xAI making Grok 4 free—after it had cost him a couple hundred dollars per month—as competitive pressure from GPT-5 and Google’s rapid product releases. His base case is a “demonetization race to the bottom of cost,” with only the heaviest models retaining a surcharge.
Salim Ismail sees a workable funnel rather than pure charity: free intelligence becomes addictive, perhaps 10% of users upgrade, and premium plans at roughly $200-$250 per month fund the broad free tier. His caveat is monetization quality—embedding advertising inside an assistant would be “a disaster.”
Pulier pushes the model beyond chat interfaces: agents will visit services to solve a user’s actual problem, not consume human-oriented marketing. “Websites will be for AIs, not for humans,” meaning machine-readable evidence, terms, inventory, and transaction capability eventually matter more than emotionally persuasive landing pages.
2. Chip fabrication is the real AI choke point—and government is entering the deal room
Blundin connects several apparently separate headlines through one constraint: frontier chips are “sold out for years into the future,” so fabrication—not the latest GPT or Grok release—is gating the US-China AI race. China is building domestic capacity while America tries to reverse decades of manufacturing concentration in Taiwan.
The administration’s proposed bargain, as the panel describes it, permits Nvidia and AMD to sell prior-generation chips into China but takes a 15% toll to help finance US capacity. Blundin calls that an excellent short-term business deal and “an incredibly slippery slope precedent”: direct federal negotiation may solve a strategic emergency while normalizing dangerous political intervention.
Taiwan’s leverage is its “silicon shield.” TSMC’s cited 66% share of advanced chips gives Washington a concrete reason to deter invasion; moving fabs onto US soil strengthens American resilience but could weaken the strategic rationale for defending Taiwan. Samsung and Intel are the only meaningful alternative manufacturers identified by the panel.
Intel is the immediate stress test. Its 1.8-nanometer process is working with low yields, but funding that recovery is cannibalizing the 1.4-nanometer, or 14-angstrom, generation and construction in Ohio has paused. Blundin predicts Intel will be revived—possibly with $40 billion of government money—while CEO Lip-Bu Tan’s Chinese semiconductor investments could force a “graceful exit” despite his technical stature.
3. Open-source AI has become economic statecraft
Z.AI’s GLM-4.5, backed with $400 million from Saudi fund Prosperity7, is presented as a startling new Chinese contender. The panel had not verified its benchmark claims, so its “best-performing open-source model” status remains provisional; what mattered was an unfamiliar team apparently leapfrogging known names such as Qwen and DeepSeek.
Salim values models that can run locally and spread globally, but resists reducing every development to a US-China binary. Blundin offers the harder geopolitical interpretation: lightweight Chinese models target phones, while free high-quality weights may also “dump” capability into the market and weaken the financing logic of US foundation-model companies.
OpenAI’s own open-source release is interpreted as pressure against both Meta and Chinese weights. Blundin speculates Washington may not want US startups building on foreign weights that could conceal unwanted behavior; Pulier calls the resulting competition “the biggest Trojan horse,” as old debates about open models unleashing “the pits of hell” disappear beneath the race to distribute capability.
The broader direction looks quasi-nationalized. Salim compares China to one corporation with companies operating like apps; America is now brokering chip deals and treating AI like the Manhattan Project, space race, or nuclear arms race. Blundin’s conclusion: “We’re moving to a war footing.”
4. Stablecoins supply the financial layer the internet never had
The White House plan aims to make America “the crypto capital of the world,” clarify SEC-versus-CFTC authority, modernize mining, staking, and digital-asset taxes, stop banking discrimination, and shift away from regulation by enforcement. Pulier’s applause is bipartisan: clarity brings innovators and bank accounts back onshore after years of forcing experiments abroad.
Under the panel’s description of the GENIUS Act, an issuer can take a dollar and provide a payment token while retaining AML and KYC obligations. That creates new rails where a consumer-to-merchant payment can settle instantly, rather than traversing banks, taking days, and leaving intermediaries to profit from fees and float.
Diamandis frames tokenization as the missing financial layer of the internet: the original network exchanged information, video, and data but could not natively exchange value. Blundin’s requirement is categorical—“You can’t use the SWIFT network and three-day settlement and $2 transaction fees” in an economy increasingly operated by AI.
5. Tokenized collateral could release trillions of dormant value
Pulier defines a real-world asset token as a digital representation of real estate, gold, dollars, stocks, or anything else with value, including fractional claims. Diamandis’s specimen is a $20 million Central Park West apartment divided into 1 million $20 tokens, letting a buyer acquire 100 tokens rather than the whole property.
Diamandis contrasts that with a friend trying to borrow a few million dollars against a $30 million house despite personally knowing the bank’s head. Ownership, liens, encumbrances, and collateral still require cumbersome verification; on-chain ownership and smart contracts could establish the same facts in seconds, providing “no delivery risk,” programmable control, fractional liquidity, and global 24/7 access.
Salim highlights the dollar-defense embedded in the first phase: approved stablecoins must be backed by dollars, T-bills, or recognized banking collateral. That preserves demand for Treasuries and “buys the US dollar a few years” while still delivering USDC-like efficiency, rather than allowing tokenized assets immediately to displace the dollar as settlement collateral.
The panel rejects $600 billion of tokenized assets by 2030 as a meaningful ceiling. Diamandis cites roughly $120 trillion in real estate, $100 trillion in equities, $13 trillion in Treasuries, and $12 trillion in gold and precious metals; Pulier suspects the modest forecast exists partly to reassure incumbent banks that disruption will be contained.
6. Programmable ownership merges finance, loyalty, and autonomous commerce
Pulier’s “Pandora’s box of innovation” begins when money carries instructions. A yogurt-shop reward becomes useful if it interoperates with restaurants and airlines, or moves into a DeFi-enabled token that earns yield; stablecoins could support “lifestyle alliances” across companies rather than isolated points programs customers ignore.
Tokenized equities could turn shareholders into recognized communities. Diamandis contrasts Yankees fans, who demand improvement after a bad game, with US Steel holders who can sell in seconds; a company that knows its owners could reward three-year holders, waive stadium parking for major Coca-Cola investors, or deliver access and benefits directly into wallets.
Pulier sees credit unions as potential community hubs after local newspapers lost their connective role. Because he estimates 80% of monetary transactions occur within one’s town, locally issued tokens could bind commerce to shared values and local participation rather than functioning solely as speculative assets.
Pulier’s diabetic-dog example captures the AI collision: an agent knows the animal’s condition and the owner’s budget, researches the medical evidence, selects food, and pays. Brands must then prove suitability to software instead of buying a heartwarming commercial—“AI appeal, not the human appeal.”
7. Retirement access and staking expand demand—and the surface area for failure
Trump’s executive order opens a path for alternative assets, including crypto and real estate, inside a 401(k) market estimated at $8 trillion-$12 trillion, but it does not implement access immediately. Diamandis thinks flows could approach the better part of $1 trillion, joking that standard 1%-2% crypto guidance clashes with his own roughly 80% exposure.
Salim cites an informal prompt given to Grok, ChatGPT, Gemini, and Perplexity: acting as financial analysts, all placed Bitcoin at $1 million within a relatively narrow 2028-2031 window. The panel treats that convergence enthusiastically, though it remains generated forecasting rather than a substantiated market model.
Pulier explains liquid staking as locking assets in a staking pool while receiving a tradable receipt—“a bearer bond of sorts”—that can participate in DeFi and generate yield. The SEC’s stated treatment removes securities uncertainty; BlackRock has already applied, and yield-bearing ETFs may follow, though whether issuers or ETF owners receive that yield remains unknown.
Blundin supplies the necessary dissent: tokenized apartments and stadiums can also be shorted. During the Big Short period, he says markets manufactured $60 trillion of synthetic instruments atop only $20 trillion of US housing, much of it landing in pension funds. Clear rules invite ethical participants; unclear rules select for “the sleaziest people on the planet.”
8. AI could finally reverse the regulatory ratchet
A DOGE AI project is reviewing 200,000 federal regulations with a stated goal of eliminating 50% by 2026. HUD reportedly removed 1,083 rules in two weeks, while a human review of the full stack was estimated at 3.6 million labor hours; Diamandis calls AI-driven deregulation a route to trillions in savings.
Pulier proposes “MVR”—minimum viable regulation—as the smallest constraint set needed while policymakers observe a technology and learn. A panelist’s European counterexample is a French identity-card process where a woman spent months assembling ancestral proof, only to learn that her birth certificate had expired after three months and she needed a new document proving she was born.
Blundin describes government as a one-way ratchet: every war or crisis expands spending, taxes, and rulebooks, but the stack never contracts afterward. AI creates perhaps the first serious chance in nearly 250 years to move downward, which becomes essential as exponential technologies generate new regulatory questions faster than humans can simplify old ones.
9. AI is reallocating capital—and widening the gap between institutions and everyone else
A six-month ChatGPT micro-cap trading experiment reportedly returned 23.8%, close to 50% annualized if extrapolated. Blundin expects AI-assisted trading to become standard, noting that Two Sigma, D. E. Shaw, and similar firms already hire exceptional technical talent into systems that operate largely outside public view.
Blundin’s warning is that even small differences in model or simulation quality may dominate returns. Retail investors can ask future ChatGPT versions for help, but institutions with superior systems and information pipelines may effectively rename the public “exit liquidity” because “you’re not going to be able to compete with people who actually know what’s going on.”
Diamandis and Blundin defend AI-assisted allocation as socially useful: capital determines which biotech, infrastructure, and frontier projects live. Blundin criticizes oversized ETFs as “a sucker fish” dragging on the market because indiscriminate flows replace judgments such as “this is garbage,” “this is real,” or “this is going to change humanity.”
Leopold Aschenbrenner’s Situational Awareness LP makes that thesis concrete. After graduating from Columbia at 19, working at OpenAI, and starting a billion-dollar hedge fund, his disclosed holdings reportedly placed 46% in Intel; Blundin’s speculative reconstruction suggests around $100 million in calls could pay roughly $1.5 billion if Intel doubles.
10. SpaceX is turning reusable launch into lunar infrastructure
Diamandis corrects an earlier SpaceX valuation from $210 billion to $400 billion and credits it with more than 95% of US launches and over 50% globally. Falcon 9 boosters have been recovered about 450 times, with one booster flying 29 missions—evidence that reusability is operational rather than aspirational.
Starship flight 10, then scheduled for August 22 using Booster 16 and Ship 37, was to test heat-shield upgrades and deploy dummy Starlink satellites after a pad explosion caused delay. Diamandis expects Starlink, SpaceX’s profitable engine for funding Starship and Mars, might be spun out within 12 months, while SpaceX itself stays private to protect the Mars mission from public-market pressure.
Starship is roughly Saturn V-sized, produces about 2.5 times its thrust, and is designed for full reuse and vertical lunar landing. The stated target is the Moon around 2027-2028, while an uncrewed Mars mission in the next couple of years and a possible Optimus payload are also discussed; Earth-orbit refueling and methane made from Martian CO₂ and water support the return architecture.
Lunar infrastructure clusters around the south-pole ice deposits preserved inside permanently shadowed craters. Nuclear power could mine that ice into drinking water, oxygen, hydrogen, and rocket fuel, supporting bases and data centers when solar is unavailable; the Moon remains about a three-day trip.
11. An interstellar visitor exposes the boundary between evidence and speculation
The panel discusses a third detected interstellar object approaching closest in October. Harvard astrophysicist Avi Loeb’s intelligent-probe hypothesis rests on its unusual path near the ecliptic and inner planets; astronomers dispute that inference, and an ordinary cometary tail produced by sublimating ice would be an important natural explanation as the body nears the Sun.
Diamandis favors redirecting a mission to intercept it, while another speaker’s pushback is wonderfully practical: if its natural origin is “unbelievably” unlikely, “do I really want to go up there and intercept it?” The exchange preserves the uncertainty—an anomalous trajectory is not proof of design.
On the Fermi paradox, Diamandis relays Bruce Damer’s argument that Earth may be unusual because liquid water persisted for more than 4 billion years, allowing life time to evolve. Another panelist offers the opposing speculation: advanced visitors may already be present but undetectable, observing under a “prime directive,” with renewed UAP testimony making the question harder to dismiss culturally.
12. Autonomy moves from cars into aviation and reshapes location value
Elon Musk’s stated target is millions of fully autonomous Teslas without human oversight in the second half of 2026, with service in dozens of cities covering 50% of the US population by year-end. Blundin says mechanics were ready years ago; neural networks that learn from experience are the control-system breakthrough.
Blundin separates two learning problems: fleets already improve driving from accumulated examples, while systems autonomously improving their own learning algorithms are “kind of imminent,” perhaps within a year. The former alone may unlock vehicles, household robots, and robotic dentistry; the panel repeats Emad Mostaque’s provocation that human cognitive input could become negative or legally prohibited in safety-critical systems.
Tesla’s forthcoming FSD model was described as using 10x the compute with materially improved video-compression loss and a possible public release by the end of the next month. A $16.5 billion minimum Samsung chip agreement secures capacity for cars, Dojo, and data-center research while TSMC is sold out; Blundin expects Tesla eventually to negotiate for Intel capacity too.
Archer’s four-passenger Midnight aircraft completed an Abu Dhabi test flight and was named air-taxi provider for the 2028 Los Angeles Olympics, linking SoFi Stadium, LAX, and Santa Monica. With an approximately 100-mile range but mostly 10-mile missions and an UberX-like cost objective, the second-order trade is land: islands, lakes, and remote sites gain value when flying cars and satellite internet remove access constraints.
13. Humanoid robots split into labor, companionship, and spectacle
Figure 02’s Helix system autonomously loaded laundry without teleoperation, though the video ran at roughly 2x speed. Blundin is unconcerned about speed because robots can work 24/7; folding laundry is the harder threshold, spanning perception and manipulation skills that generalize across “tens of thousands of tasks” in homes and factories.
Figure, Tesla, and 1X are vertically integrating even basic components such as motors because a mature supply chain does not yet exist. Tesla’s Optimus generation 3 slipped to 2026 after producing only hundreds against a 5,000-unit aspiration, pausing production in June, freezing component orders, and changing leadership—Pulier treats that as normal organizational turbulence around an exceptionally hard problem.
Real Botics’ Melody represents the opposite branch: replaceable faces, multiple languages, and AI optimized for companionship rather than physical work. Pulier and the hosts see plausible elder-care benefits but a larger danger of sex robots, plummeting birth rates, and people retreating from difficult human relationships; “we’re fueled by human connection,” Pulier argues, so deployment needs a deliberate counterbalance.
An underground San Francisco fight between Unitree G1 robots divides the panel. Pulier argues it should not be allowed and could set the field back; Blundin wants the mechatronics and giant robots, while Salim compares it with Formula 1, esports, and future Olympic virtual sports, arguing national teams and competition could accelerate hardware and software. The unresolved question lands perfectly: “What could possibly go wrong?”
14. Fusion is investable before it works because AI creates demand for power
The energy thesis begins with Eric Schmidt’s formulation that AI is “not chip limited, it’s electron limited.” Salim asks whether a six-month AGI or ASI lead could become permanent if America falls behind on generation; Blundin disagrees that power decides the race, calling fabs and algorithms the 10x-to-100x multipliers and arguing AI can outbid other users for electricity by 5x or 10x.
Helion’s magneto-inertial Orion project is described as a 50-megawatt plant beginning construction with a Microsoft contract and an end-of-2028 operating goal. Pulier suggests new physics or AI could help solve fusion’s magnetic-containment mathematics, while the site’s concrete, generators, grid connection, and other infrastructure retain value if the heat source ultimately becomes a small modular fission reactor instead.
China added $2.1 billion to fusion and, according to the panel, has the longest sustained fusion reaction. Salim contrasts roughly 10 terawatts of Chinese generation with a US level around 4 terawatts. His slogan is that China pursues “build baby build,” combining solar and other sources through long-term state planning.
Commonwealth Fusion Systems has a 200-megawatt Google purchase agreement for a Virginia plant expected in the early 2030s; Google will take half its power, while the company cites more than 1,000 employees and $2 billion invested. Data centers remove traditional grid-demand risk, and existing coal plants already possess land, containment, and transmission—but investors hesitate on solar because successful fusion could strand panels and storage.