Summary

Autonomous hardware and scarce compute components were central technology winners. Friedberg called 2025 “the year of the robot,” citing Unitree’s $1,600 Go2 quadruped and $16,000 G1 humanoid; Baker added mainstream FSD adoption and picked high-bandwidth memory from SK Hynix and Micron as the best-performing asset. HBM, he said, is a larger share of GPU input costs than TSMC and has been sold out for two years.
Full-stack AI providers were cast as winners, while enterprise software and independent model labs face brutal economics. Baker argued that o3-style reasoning and test-time compute let a large company spend $1 million answering its most important question over six weeks, while cloud owners enjoy lower infrastructure costs than labs renting compute. Chamath called legacy SaaS the “software-industrial complex.” Jason argued Google’s Deep Research is already outperforming rivals and predicted OpenAI’s cited $57 billion valuation could be its peak, with a nonzero chance that court cases block the transfer of $157 billion in value from nonprofit to for-profit.
Chamath’s stablecoin call was that dollar-denominated stablecoins could quadruple or quintuple during 2025. He cited roughly 1.1 billion transactions and $8.5 trillion of second-quarter 2024 volume—more than twice Visa’s—and argued that removing 300 basis points of payment friction could be worth $1 trillion in the United States alone. Baker’s pushback was geopolitical: dollar rails are advantageous, but a stablecoin constellation replacing the dollar as reserve currency would be “very bad for America.”
The political forecasts favored fiscal restraint and a younger governing class while betting against Putin, neoconservatives and progressivism. Chamath saw austerity testing the case for fiscal conservatism; Friedberg contrasted Trump’s roughly 40–45-year-old cabinet with Biden’s nearly 60-year-old one. Baker predicted Europe’s rearmament would free American resources for the Pacific and induce Xi Jinping to distance China from Putin. Jason expected aggressive Trump rhetoric before negotiated deals.
The macro outlook was a barbell between an AI-driven growth boom and a low-probability banking rupture. Baker predicted at least one year above 5% real GDP growth within four years, powered by AI and deregulation; Chamath warned that 5% rates on roughly $70 trillion of aggregate Pax Americana debt can impose the dollar burden that 10% rates once did. His hedge was long CDS protection—a trade he expects to lose most of the time but that could return 100–1,000x in his scenarios; Baker said a genuine bank failure could produce 1,000–10,000x.
Friedberg’s contrarian social call was that accelerating growth could revive socialism rather than defeat it. DOGE cuts, contracting disruption and AI replacement of white-collar labor could leave large groups behind even as some billionaires become $100 billionaires and eventually trillionaires. Baker’s framing was that before AI makes money irrelevant, test-time compute means “money will matter more than it’s ever mattered before.”
A post-Lina-Khan M&A wave could consolidate autos, AI, robotics manufacturing and autonomous transportation. The panel treated Honda–Nissan as a warning for legacy OEMs caught between Tesla and Chinese producers, while Baker expected something significant at Intel and said independent frontier AI labs could go quiet as full-stack economics favor compute owners. Waymo’s San Francisco share reportedly reached 22% in 15 months—matching Lyft—making a financing, IPO or strategic transaction plausible alongside combinations involving Uber, DoorDash, Amazon, Tesla and drone-delivery operators.

Deep dive

1. Fiscal restraint and a younger political class take the wheel

Chamath’s political winner was fiscal conservatism: austerity must expose federal “waste, fraud and abuse” because the remaining alternative is entitlement cuts. He expects the result to spill into state elections and give fiscal conservatives their day in 2025.
Friedberg chose younger candidates, contrasting Trump cabinet picks averaging roughly 40–45 years with Biden’s cabinet at a little over 59, nearly 60. By late 2025, he expects new names to emerge for the midterms with resonant messages and less attachment to the aging political establishment.
Baker named Trump and centrism, then chose Gen X and elder millennials as his generational winner. He cited Elon Musk, David Sacks and Marco Rubio among Gen X appointments, and J. D. Vance and Vivek Ramaswamy among younger figures, arguing that this cohort will think more about its own children and produce a sea change.

2. Putin, neoconservatives and progressivism face different defeats

Baker’s political loser was Putin, who has lost roughly half a million people “and for what exactly?” European rearmament should let America transfer resources toward Japan, South Korea and the Pacific, complicating China’s Taiwan ambitions and giving Xi Jinping reason to decouple from a Russian client state.
Friedberg predicted a defeat for pro-war neoconservatives at the hands of J. D. Vance, Elon Musk and their allies. Jason’s pushback was that Trump may sound far more bellicose than expected, using a John Bolton-like threat posture to extract better deals before Friedberg ultimately proves right.
Chamath chose “progressivism.” He forecast Trudeau losing massively to Pierre Poilievre in Canada, the AfD winning in Germany, and Marine Le Pen likely winning in France if a deadlock leads to an election. His UK scenario relied on allegations that hundreds of thousands of girls were abused over decades by organizations of Pakistani Muslim men while prosecutions were avoided for fear of stoking Islamophobia. If that fallout comes to pass, he expects Labour to fall and Nigel Farage to win.
His retrospective claim was broader: 2024 ended the old Republican Party and turned it into a vessel for a MAGA coalition. That rendered the Koch family’s political machinery effectively worthless while elevating donors such as Miriam Adelson and Musk; Baker conceded that his view of money in politics partly depends on agreeing with the largest donor in the cycle.

3. Cheap robots move from demonstrations into deployment

Friedberg’s “year of the robot” call rested on price and programmability, not science fiction. Unitree’s Go2 costs about $1,600, exposes an API and carries lidar and intelligent guidance systems; his team placed an order to use one on test farms for imaging and data collection at a cost below $3,000.
Unitree’s roughly $16,000 G1 humanoid makes general-purpose hardware substantially more accessible. The military videos—with weapons mounted on quadrupeds—illustrated a field-soldier use case, while Friedberg highlighted scientific applications and farm data collection.
Friedberg argued that technologies take a long time to work and then arrive faster than expected. Baker included Tesla FSD in the category: he already prefers late-night Uber rides in Teslas because a tired driver feels less safe, and expects FSD’s improvement to compound at an accelerating rate.

4. Reasoning models make compute ownership a strategic moat

Baker’s mechanism starts with o3: reasoning plus test-time compute changes what a well-capitalized company can buy. A large enterprise able to spend $1 million letting AI think for six weeks about its most important problem gains an advantage that a smaller company cannot readily match.
Baker’s broader rule was that the ultimate AI winner will have the lowest infrastructure and compute costs. A lab renting capacity from Azure, AWS or Google pays a markup and is disadvantaged over time relative to the provider’s internal services; the full stack wins.
Jason grouped Google, Tesla and xAI among the beneficiaries, highlighting Elon Musk’s reported deployment of 100,000 GPUs in under 45 days. He reversed his prior optimism about Apple Intelligence—“it sucks”—and argued that Google’s search, YouTube, Gmail, Drive and Docs give Gemini unusually powerful context.
His best demonstration was Gemini Deep Research: for $20, it decomposed a question about bank insolvency, searched 162 sites and produced a cited report in about ten minutes. Jason compared the output with month-long work by Gartner, McKinsey or BCG and contrasted it with OpenAI’s $200-per-month o1 Pro, while admitting, “I don’t know how much of this is correct.”

5. Stablecoins challenge card economics without settling sovereignty

Chamath said 2024 delivered a critical decoupling: stablecoin usage kept rising independently of crypto volatility and moved into useful wholesale business functions. By the end of the second quarter of 2024, he cited about 1.1 billion transactions representing $8.5 trillion—more than twice Visa’s transaction volume over the same period.
That is his “point of no return”: dollar stablecoins could quadruple or quintuple by year-end 2025 and begin attacking the Visa–Mastercard duopoly. Removing roughly 300 basis points of friction from the global economy could, in his estimate, create $1 trillion of value in the United States alone.
Jason’s concrete example was his research product, which uses Stripe and incurs hundreds of thousands of dollars in transaction costs. Jeremy Allaire offered to rebuild the payment rails with U.S.-dollar stablecoins. Jason also raised concerns about stablecoins’ use in terrorism, sanctions evasion and human trafficking, while Baker warned that replacing the dollar as reserve currency would surrender America’s advantage of borrowing in—and controlling—the real value of its own currency.

6. Cost-plus incumbents and concentrated assets lose their cover

Baker’s cleanest loser was any government-service provider deriving more than 35% of revenue from the United States government. Under DOGE, “actually check the bill” becomes a threat to contractors accustomed to limited scrutiny.
Friedberg extended that logic to Boeing, Lockheed Martin and Raytheon, expecting traditional defense and aerospace providers to shift toward more technology-oriented, drone-driven systems. He said Palantir and Anduril would benefit, while large contractors face failures of scale and bureaucratic inefficiency.
Chamath separately warned that the Magnificent Seven’s index representation is approaching 40%, a concentration level that has historically preceded retrading. He still called them exceptional businesses, but even a 10% decline would erase a couple of trillion dollars—making them potentially the largest absolute-dollar losers without requiring broken fundamentals.
Sports produced a real disagreement. Chamath saw weakening NBA viewership, a less compelling product and shrinking advertising economics if pharmaceutical ads leave television; private-equity owners would apply DCF discipline rather than trophy-asset emotion. Baker saw institutional capital expanding the buyer base and Google, Amazon and Netflix happily acquiring sports rights. Personalized ads could make scarce live audiences more valuable even if pharmaceutical advertising retreats.

7. Jason doubles down on an OpenAI collapse call

In reviewing the prior OpenAI call, Jason noted that its value roughly doubled. He nevertheless made OpenAI his 2025 business loser and most contrarian call: Google is “kicking ass,” xAI is building infrastructure quickly, and Microsoft has source code and substantial computing infrastructure that could reduce its dependence on OpenAI.
Jason said OpenAI’s cited $57 billion valuation could be its peak. He also saw a nonzero chance that court cases involving the transfer of $157 billion in value from a nonprofit into a for-profit could derail the transition.
Baker’s pushback was that OpenAI is “a real business with real revenue, real scale, real growth, real technology,” distinct from a meme stock. Jason answered that its consumer and developer revenue could be challenged: developers prefer open-source models, route queries across multiple stacks and show little loyalty to OpenAI, Gemini or other closed services.

8. The deal drought breaks across autos, hardware and AI labs

Chamath treated the Honda–Nissan agreement, with Mitsubishi connected through Nissan’s alliance, as the opening signal for auto mega-mergers. Legacy OEMs face “melting iceberg” economics as Tesla leads in software and autonomy and Chinese manufacturers offer more competitive products.
Baker agreed but identified government intervention as the principal hedge: Volkswagen, Stellantis and other national champions employ too many people to be left entirely to market forces. Absent significant protectionism or government support, he expects them to lose their Chinese business and remain squeezed between Tesla and Chinese OEMs.
Friedberg forecast blockbuster capital raises for American robotic and autonomous-hardware manufacturing, potentially mixing private equity with government support. Baker anticipated a “tidal wave of M&A,” something consequential involving Intel, and independent frontier AI labs becoming quiet as rented compute prevents them from becoming the lowest-cost provider.

9. Waymo turns autonomy partnerships into an urgent chessboard

Jason envisaged combinations among Tesla, Uber, DoorDash, Amazon and Waymo as Lina Khan’s tenure ends. Tesla could buy Uber for roughly 10% of its own market capitalization, while Amazon could afford DoorDash; the strategic prize is a super-app spanning rides, food, parcels and commerce.
Baker supplied the adoption data: when Waymo opened broadly in San Francisco in August 2023, Uber and Lyft reportedly held 66% and 34%. Fifteen months later, Waymo had reached 22%—equal to Lyft—while Uber had fallen to 55%.
Waymo’s expansion into Los Angeles, Austin and other cities, plus a new hardware platform supposedly intended to reduce launch capex, could support a large financing, IPO, merger or acquisition. Users may currently find its routes slow or monotonous, but the panel’s repeated reaction after riding was: “That is the future.”
Baker added Zipline-style autonomous drones as the suburban wildcard; Jason cited Amazon’s Texas operations, where roughly 60,000 SKUs were reportedly eligible for backyard delivery in about 45 minutes. With ridesharing representing about 1.5% of U.S. rides and less than 1% globally, Jason expects the addressable share to approach 20%, leaving room for multiple global winners including Uber, DoorDash, Waymo, Amazon and BYD.

10. A bank failure is the tail risk; CDS is the asymmetric hedge

Chamath’s banking-crisis case begins with roughly $70 trillion of U.S. government, corporate and mortgage debt. At that scale, 5% rates impose a dollar burden comparable to 10% rates 25–30 years ago, when the debt base was far smaller; credit or mark-to-market losses could then become a reserve problem at a major bank.
He called the risk nontrivial and declined to identify the two banks he considered most exposed. Baker would not call failure likely either—“anything is possible”—but agreed that a bank event would make default protection explosively valuable.
Chamath characterized long CDS as a trade that goes to zero 92 times in 100; six of the other eight outcomes might return 10x, and the final two could return 100–1,000x. Baker said a true bank failure might produce 1,000x or 10,000x, though Chamath repeatedly emphasized that he wants the insurance to expire worthless.
At the opposite end of the barbell, Baker predicted at least one year above 5% real GDP growth within the next four years. AI productivity and deregulation could cause the economy to double roughly every 12 years at 5–6% growth, versus about 24 years at 3%.

11. AI abundance could intensify scarcity and revive socialism

Friedberg rejected the presumed post-election burial of socialism. Rapid progress creates concentrated winners and displaced industries; he invoked Argentina’s roughly 8% growth around Perón’s rise to show that aggregate expansion does not guarantee widely shared benefits.
DOGE cuts, reduced federal employment, reduced federal contracting and AI-driven employment disruption could create a “more difficult year” than Silicon Valley expects. Jason already sees well-qualified venture and technology workers unable to regain six-figure or mid-six-figure compensation because companies automate, outsource and “do more with less.”
Baker’s framing sharpened the paradox: people say money becomes meaningless under AGI or ASI, but first “money will matter more than it’s ever mattered before.” Companies and individuals able to purchase more test-time compute obtain an enormous advantage, so AI may amplify inequality for a meaningful period.
Friedberg said wokeism and progressivism would decline while socialism as government policy would rise. Jason asked why the United States lacks universal health care and pre-K and proposed state-level experiments, vouchers and competition. Chamath pushed back that federal funding had inflated education, health care and housing by distorting those markets.

12. HBM, Chinese technology and mega-cap AI define the asset debate

Baker selected high-bandwidth memory as 2025’s best-performing asset. HBM is a larger share of GPU input costs than TSMC, he said, and supplies Nvidia, AMD and Amazon Trainium chips; SK Hynix and Micron can make it today, with Samsung a potential third source if it “gets their act together.”
Friedberg chose Chinese technology stocks or ETFs, citing three possible catalysts: a U.S.–China market-access deal, massive low-cost electricity expansion—including a cited $137 billion hydroelectric project—and renewed Communist Party tolerance for entrepreneurship. Alibaba’s valuation and other beaten-down, mid-single-digit multiples made the asymmetry attractive.
Baker retained a “no China guideline” after an old financial fraud in which even Western-audited documents were allegedly altered locally before shipment. Still, he agreed that Trump and Xi both want a deal and that Chinese companies—with customers across Europe, Africa, South America and the broader Southern Hemisphere—could rally strongly if one materializes.
Jason took the other side of Chamath’s concentration warning and picked the Magnificent Seven. His premise was operating leverage: mega-caps deploy AI internally before selling it externally, suppress hiring and automate work, potentially producing earnings expansion that investors “will not be able to comprehend.”

13. Agents attack the software-industrial complex at its labor base

Baker’s worst-performing asset was enterprise application software, particularly in 2025’s second half when agents can perform online actions for users. Labs and cloud platforms have the models and compute; application vendors have customer data and relationships, but enterprises already maintain comparable relationships with AWS, Google and Microsoft.
Chamath named the target the “software-industrial complex”: large vendors wrap heuristics and business rules around CRUD databases, then defend them through “golf trips” and “steak dinners.” None of that equals product value when CEOs and CFOs pressure CIOs to justify spending.
His operating example was a 30-person engineering team doing the work of 300 people, potentially 3,000 the following year. A new vendor can answer a $100 incumbent bid with $10 and remain highly profitable. Chamath argued that incumbents make white-collar workers more efficient, whereas AI businesses may simply replace the worker.
Friedberg therefore “triple underlined” vertical SaaS as a loser through pricing compression and customer-built internal tools. Jason’s alternative short was legacy autos and real estate: indebted consumers cannot afford cars or mortgages, while Texas home values and rents had already declined for two consecutive years amid abundant construction.

14. Nuclear power, reasoning traces and exits become structural trends

Chamath’s canary was the Supplementary Leverage Ratio, an arcane bank rule governing whether Treasuries enter reserve calculations. With roughly $10 trillion maturing and perhaps $10–20 trillion of issuance needed over coming years, quiet rule changes would signal that the system is again “kicking the can down the road”; leaving the rules unchanged would suggest elected officials control policy.
Friedberg expected 2025 announcements for major U.S. nuclear-power buildouts, enabled by deregulation and the need to meet competitive electricity demand from China. His leading indicator was human capital: multiple smart people had left strong jobs to start nuclear companies in anticipation of policy opening.
Baker predicted AI would advance more per quarter in 2025 than per year in 2023 or 2024. Pretraining, test-time compute and reasoning now form three multiplicative scaling axes; synthetic “reasoning traces” teach models the internal monologue behind answers, but Ilya Sutskever’s warning was that reasoning systems are “inherently unpredictable.”
Baker also expects frontier labs to stop releasing their very best models to impede knowledge distillation—DeepSeek’s apparent belief that it was GPT-4 was his example. Jason’s capital-markets corollary was an “exit and DPI shower” as M&A and IPOs revive after Lina Khan’s tenure.

15. The finale makes the calls tradable, then looks off-world

The media picks ranged from Jason’s newsroom upheaval at The Washington Post, CNN and the Los Angeles Times to Chamath’s hoped-for release of JFK, Epstein, Diddy and other files. Friedberg chose AI-native games with dynamic plots and cheaper production; Baker’s unequivocal selection was season two of 1923.
Proposed Polymarket lines included whether Trump would deport more than 750,000 people in his first year, whether the Magnificent Seven’s S&P 500 representation would fall below 30%, and whether December 2025 federal debt would finish above or below $38 trillion. Azure versus AWS was suggested as a contest in absolute cloud-revenue gains.
Baker assigned at least a 25% chance that the U.S. government possesses knowledge of extraterrestrial life or proof of it; Chamath offered 20%. Baker said he did not know what the New Jersey drones were and suggested that, if they were not U.S. or Chinese government drones, the most likely explanation could be a drill after which hysteria caused people to misidentify aircraft. He cited pilot reports, sensor observations and recurring media investigations, wondering whether UFO attention clusters around technological phase shifts such as nuclear power and AI.
Friedberg challenged the premise that advanced intelligence would transport biological bodies through space. Once matter can be rearranged using abundant energy, a civilization might gather and affect information without moving biological organisms; physical craft resemble humanity’s current technological imagination more than an unavoidable feature of advanced life.

Jeff Clune - Agent AI Needs Darwin Teaching AI to See: A Technical Deep-Dive on Vision Language Models with Will Hardman of Veratai