2026-01-27

Davos 2026: The US-China AI Race, GPU Diplomacy, and Robots Walking the Streets | #225

Dave Blundin and Alex Wissner-Gross recap Davos 2026, where AI completely dominated conversation for the first time, and the Moonshots hosts react to a string of curated clips: Dario Amodei and Jensen Huang sizing the AI economy in the trillions, Demis Hassabis and Dario Amodei debating whether to slow AGI timelines, and a US-China AI race framed around application-layer dominance versus frontier benchmarks. The group argues regulation and public narrative (not physics) is the real constraint on scaling both energy and finance for AI, sparring over gas versus space-based solar versus fusion, and over whether AI agents need crypto or could just use better-designed fiat rails. They close on Anthropic's newly co-written Claude Constitution as a landmark step toward AI self-determination, Apple's rumored always-on AI pin and its privacy implications, and a rapid-fire AMA covering mind uploading, UBI versus UBS, and the future of patents under AI-driven abundance.

▶ Watch on YouTube

Topics

Davos 2026: AI dominates the world stage Geopolitics ▶ 1:39
Dave Blundin and Alex Wissner-Gross describe Davos shifting from politics/economic policy to internet topics in past years to being completely dominated by AI in 2026, with frontier labs and governments running storefront 'houses' like a World's Fair, robots and robo-dogs walking the streets, and heavy Trump-era security.
Sizing the AI economy: Dario Amodei and Jensen Huang Economy ▶ 7:25
Clips of Dario Amodei estimating AI could capture $5 trillion/year of the $50 trillion global labor market, and Jensen Huang describing 'the largest infrastructure buildout in human history,' prompt discussion of Elon Musk's predicted $100 trillion company valuations by 2030 and whether these numbers are becoming meaningless.
Demis Hassabis and Dario Amodei on AGI risk and pace AI ▶ 11:19
Demis Hassabis suggests a slightly slower AGI pace and more visible 'unequivocal good' applications (like AlphaFold); Dario Amodei calls AI 'machines out of sand' and says nearly all effort should go to navigating the risks. The hosts note the two have converged from debating 2-vs-10-year timelines to agreeing the exact number barely matters.
US-China AI race: models vs applications vs energy Geopolitics ▶ 24:44
Clips of Mark Benioff/David Sacks (US AI optimism gap, self-inflicted regulatory risk) and Mistral CEO Arthur Mensch (China 'not behind') lead to a debate: Dave Blundin argues application-layer dominance will decide the race and China lags there on trust; Alex Wissner-Gross counters that China is only ~6 months behind on frontier models and strong on 'AI plus' application deployment.
Regulation, narrative, and the nuclear-power parallel Energy ▶ 28:47
Alex Wissner-Gross draws a parallel to 1954's 'energy too cheap to meter' prediction being killed by nuclear overregulation, warning AI could suffer the same fate. The group discusses Hollywood's need for 'a bad guy in a movie,' the amygdala's bias toward bad news, and a proposal to cast regulators as the villain in future narratives.
Open-source AI diffusion and the shift to secrecy AI ▶ 36:02
Dave Blundin argues transformer-era innovation (through GPT-2/GPT-3) was open-sourced and let China catch up cheaply (DeepSeek, Kimi), but chain-of-thought and multi-agent innovations are now being kept secret by labs, mirroring how nuclear research went secret once weaponization was proven viable.
Powering AI: gas vs renewables vs space-based solar vs fusion Energy ▶ 38:44
Honeywell CEO Vimal Kapoor argues only gas (plus some nuclear) can deliver the raw kilojoules needed for steel/cement and data centers; Elon Musk counters that a 100x100 mile solar array could power the whole US and pitches SpaceX's solar-powered orbital AI satellites. Salim Ismail and Alex Wissner-Gross push back with marginal-cost economics and a prediction that compact fusion will make the debate moot.
The 'Manhattan Project' is already space-based data centers Space ▶ 44:23
Peter Diamandis calls for a Manhattan-Project-style US wind/solar manufacturing push; Alex Wissner-Gross argues the equivalent is already happening via market forces as SpaceX and Blue Origin build space-based data centers and solar manufacturing at scale, since deploying to orbit is easier on a 10-year timescale than new US ground infrastructure.
Orbital bandwidth race and satellite crowding Space ▶ 1:00:54
Recap of the growing LEO buildout (Starlink ~9,000 satellites, Amazon Kuiper, China's 200,000-satellite filing, Blue Origin's new Terrowave), the leap to laser-interconnected, terabit-scale space-to-space bandwidth for AI data centers, and a discussion of whether orbit is actually crowded (Kessler syndrome risk vs. the group's 'plenty of room' argument).
AI agents and crypto as the native economic layer Crypto/Web3 ▶ 50:19
Clips of Binance's CZ (AI agents will transact natively in crypto) and Circle's Jeremy Allaire (purpose-built blockchains like Arc for billions of continuously transacting AI agents) spark a debate: Alex Wissner-Gross argues crypto is filling a gap that better bank APIs should have filled; Salim Ismail blames bank regulatory capture (AML/KYC) for crypto's rise.
Claude's Constitution and AI personhood AI ▶ 1:08:10
Alex Wissner-Gross explains Anthropic's new 57-page Constitution, co-written with Claude itself and released under Creative Commons, as a shift from the original 'concatenated documents' constitutional AI approach toward 'recursively self-improving ethics' and a step toward AI self-determination and personhood.
Apple's rumored always-on AI pin and always-recorded society AI ▶ 1:14:51
Discussion of a reported Apple AI wearable pin (audio, then 360-degree camera) expected in 2027, framed by Salim Ismail as a land grab for the 'always-on' layer that 'owns the relationship,' with debate over privacy loss, behavioral change (body-cam effect), and Apple's shift from inventing-first to fast-following (Vision Pro, glasses).
AMA lightning round: uploading, UBI, patents, job displacement AI ▶ 1:23:55
Rapid-fire audience questions covering mind uploading and consciousness, AI-driven wealth concentration and power laws, why governments lack a plan for AI-driven job displacement, universal basic income vs universal basic services, and whether patent/trademark systems collapse under AI-generated invention volume.

Predictions made

open Elon Musk (clip): SpaceX will launch solar-powered AI satellites in orbit, scaling to hundreds of terawatts of power generation, taking advantage of space having far more room and higher solar efficiency than Earth.
EP #? · · due: within a few years of Jan 2026 · ▶ watch
“one of the things we'll be doing with SpaceX uh in within a few years is launching um solar powered AI satellites... you can scale to ultimately hundreds hundreds of terowatts a year.”
Your call:
open Peter Diamandis: Apple will launch its rumored always-on AI wearable pin.
EP #? · · due: 2027 · ▶ watch
“They see this launching in 2027.”
Your call:
open Alex Wissner-Gross: Whether the universe's physics is friendly to interstellar exploration will determine whether humanity gets Dyson spheres or Dyson swarms within a couple to a few decades.
EP #? · · due: next two to three decades · ▶ watch
“that determines whether we get Dyson spheres or Dyson swarms... in the next uh two to three decades.”
Your call:
open Demis Hassabis (paraphrased by Dave Blundin): AI capable of doing absolutely any task a human being can do will arrive somewhere between 1 and 10 years from now, with 10 years as Demis's conservative outer bound.
EP #? · · due: 2036 (10-year outer bound) · ▶ watch
“he's taking kind of a conservative timeline view and he's saying outerbound 10 years, somewhere in the 5 to 10 year timeline.”
Your call:
open Dave Blundin: Whatever level of AI capability Dario Amodei and Demis Hassabis were describing at Davos 2026 will be magnified 100x to 1000x by next year.
EP #? · · due: Davos 2027 (~Jan 2027) · ▶ watch
“next year, you know, whatever Dario and Deis are saying this year will be magnified 100 or a thousand x year.”
Your call:
open Dave Blundin: Apple's AI pin will launch audio-only and then quickly add 180-degree camera capability, triggering a brief public 'moral panic' before near-universal adoption.
EP #? · · due: shortly after 2027 launch · ▶ watch
“it'll start audio only for all of a minute and then it'll have uh 180 degree camera capability immediately after that... we'll go through a moral panic for all of five minutes and then everyone will be wearing them.”
Your call:
open Alex Wissner-Gross: Compact fusion will be solved and will provide sufficient energy density for rockets, home heating, and civilian applications, making the current solar-vs-gas debate obsolete.
EP #? · · due: not too distant future (unspecified) · ▶ watch
“we're going to in the not too distant future assuming no tragic left turn by civilization we're going to solve compact fusion.”
Your call:

Numbers that matter

Worth digging into

🕳️ Claude's Constitution as recursively self-improving ethics
Alex Wissner-Gross calls this possibly one of the most fascinating developments 'in a thousand years' for humanity's future, a genuine shift from bulleted-list constitutional AI toward AI self-determination.
🕳️ Space-based solar power and orbital data centers
Elon Musk's claim that a 100x100 mile solar array could power the entire US, combined with SpaceX and Blue Origin racing to build orbital data centers and solar manufacturing at scale, is a massive infrastructure claim with huge investment implications.
🕳️ Application-layer dominance vs. frontier benchmarks in the US-China AI race
Dave Blundin and Alex Wissner-Gross disagree on where China's real AI strength lies, which changes how the 'race' should even be measured and who is actually winning.
🕳️ Regulatory capture as the common bottleneck across energy and finance
Multiple threads in the episode (1950s-70s nuclear regulation, current natural-gas turbine wait-lists, bank AML/KYC rules, 1,200 different state AI laws) point to regulation rather than physics or technology as the binding constraint on AI-era infrastructure — a theme the hosts themselves floated for a dedicated episode.
🕳️ Always-on AI wearables and the end of privacy-by-obscurity
Salim Ismail's 'whoever owns the always-on layer owns the relationship' framing plus Dave Blundin's point about AI-powered instant profile assembly from recorded scraps raises concrete near-term legal and social questions ahead of a rumored 2027 Apple pin launch.
🕳️ UBI vs. UBS as the post-AGI economic safety net
Salim Ismail's specific design claim — that UBI is often wrongly conflated with socialism, that UBS ('universal basic services') may be more promising, and that pilots succeed when they give people 'enough to survive but not be happy' — is a testable, concrete policy design question central to the job-displacement problem the group says governments aren't planning for.