2025-11-22

AI Update: NVIDIA’s Record Revenue, Elon’s Data Centers in Space & Gemini 3’s Insane Performance

The panel covers Nvidia's record $57B quarter and the Nvidia-Microsoft-Anthropic AI power block, then digs into Saudi Arabia's sovereign AI buildout for xAI and Elon Musk's plan to put 100 gigawatts/year of AI compute into orbit (eventually mined from the moon). They cover the US-China electricity gap, X-Energy's new TRISO nuclear fuel facility, Zipline's drone-delivery scale-up, and the humanoid/quadruped robot race (Unitree, Sunday Robotics, Clone Robotics, Figure). They close on epigenetic-reprogramming trials starting in 2026, Gemini 3 beating radiologists on a diagnostic benchmark, environmental remediation via drones, and why Massachusetts/MIT out-produces Silicon Valley in VC-backed IPOs.

▶ Watch on YouTube

Topics

Nvidia earnings and the race to redesign AI chips Compute ▶ 0:19

Nvidia posts $57B revenue (62% YoY) and 10GW of new AI infrastructure deals; panel debates whether Nvidia's pricing power and architecture lead are durable as TPUs, AMD, and custom ASICs emerge.

Nvidia-Microsoft-Anthropic AI power block Economy ▶ 10:22

Anthropic commits $30B to Microsoft Azure compute (powered by Nvidia chips) while Microsoft and Nvidia invest up to $15B into Anthropic, read as a hedge against antitrust risk and OpenAI dependence.

Nvidia's shift to full AI servers and the reversal of compute form factor Compute ▶ 18:26

Nvidia moving from chips to full AI servers reflects computing's form factor reversing after decades of miniaturization, as Moore's Law ends and AI superclusters become the dominant computer.

Saudi Arabia's sovereign AI buildout and xAI's data center Geopolitics ▶ 21:17

xAI becomes first customer of an Nvidia-backed Saudi data center under Saudi's Vision 2030 and ~$100B AI commitment; panel frames it as a sovereign-compute test bed, not the long-term compute capital.

Space-based AI data centers and Elon Musk's 100GW/year orbital plan Space ▶ 28:20

Elon Musk describes a path to 100 gigawatts/year of solar-powered AI satellites (a quarter of total US electricity), eventually mining the moon via O'Neill-style mass drivers for materials and solar power.

US-China electricity generation gap and the energy bottleneck for AI Energy ▶ 41:26

China is outpacing the US roughly 4x in new electricity generation (solar, wind, nuclear, gas); panel debates whether fusion, fission restarts, or solar can close the US energy gap before chip fabs outrun power supply.

Google's $40B Texas AI infrastructure investment Economy ▶ 48:19

Google commits $40B through 2027 to Texas cloud/AI infrastructure plus 6.2GW of new energy generation, with Texas framed as the only state building fast enough to host AI's exploding electricity demand.

Advanced nuclear fuel: X-Energy's TRISO pebble-bed reactors Energy ▶ 51:12

X-Energy breaks ground on the first domestic Cat 2 nuclear fuel facility, producing TRISO fuel pebbles for meltdown-resistant pebble-bed reactors decades after the design was first studied.

Drone delivery at scale: Zipline's growth and societal implications Robotics ▶ 57:32

Zipline is scaling to 20,000 autonomous aircraft/year, growing deliveries 15% week-over-week for 30 straight weeks; panel sees mature drone delivery leading to hyperlocal manufacturing and relocalized supply chains.

Humanoid and quadruped robots: the training-data race Robotics ▶ 1:09:00

Unitree, Sunday Robotics (teleoperation gloves), Clone Robotics (hydraulic tendons), Figure, and 1X compete on dexterity; panel agrees robotics' algorithms are largely solved and training data is now the bottleneck.

EV adoption outpacing official forecasts Transportation ▶ 1:17:47

IEA repeatedly underestimates EV adoption curves (e.g., predicting near-zero EVs in India through 2035); panel attributes this to simpler drivetrains (17 moving parts vs. 2,000) and coming robotaxi displacement.

Longevity breakthroughs and AI-driven medicine Longevity ▶ 1:19:21

David Sinclair's partial epigenetic reprogramming (Life Biosciences) enters human trials in Q1 2026 starting with eye disease; Anthropic is hiring biologists as Dario Amodei targets solving disease within 5 years, while Gemini 3 already beats radiology residents on a diagnostic benchmark.

Environmental remediation and rare-earth re-industrialization Other ▶ 1:25:08

Drone-based weather modification aims to refill the Great Salt Lake; a $1.4B government push (Vulcan/ReElement) seeks to rebuild domestic rare-earth supply chains that China currently dominates.

Boston/MIT vs Silicon Valley as startup geography Economy ▶ 1:30:42

Massachusetts leads the US in VC-backed IPO probability (4.1% vs. California's 2.3%); panel ties this to university density and MIT producing more unicorn founders than Stanford, despite California's larger IPO volume.

Predictions made

open Dave Blundin: Anthropic and Microsoft will eventually merge.

· due: unspecified · ▶ watch

“I'm predicting a merger between Anthropic and Microsoft here.”

Your call:

open Peter Diamandis: There will be a slight market correction in AI valuations as major AI companies expand into new industries, then growth resumes and goes vertical again, following a Gartner-hype-cycle pattern.

· due: unspecified near-term · ▶ watch

“I'll make a prediction. I think there'll be a slight correction... and then it just goes vertical again.”

Your call:

open Elon Musk (clip): Starship will become operational on a regular basis, enabling the space-based AI compute buildout to begin.

· due: 2027 · ▶ watch

“Starship has got to become up and operational on a regular basis... something I think we'll see in the next 12 to 18 months.”

Your call:

open Alex Wissner-Gross: Within 10 years, low hundreds to many hundreds of gigawatts of AI compute will be operating off the Earth's surface (in orbit).

· due: 2035 · ▶ watch

“I would be mildly, if not significantly, shocked if 10 years from now... we don't see low hundreds, if not many hundreds of gigawatts of AI compute not on the Earth's surface.”

Your call:

open Dave Blundin: A massive breakthrough in photonics (or a similar energy domain) will occur within the next couple of years, though the US still faces a difficult 5-to-7-year energy-constrained period before then.

· due: 2027-2032 · ▶ watch

“I'm predicting we'll see a massive breakthrough in photonics or some of those domains over the next couple of years.”

Your call:

open Dario Amodei (relayed by Alex Wissner-Gross): Disease, biology, and medicine will be effectively solved with the help of AI.

· due: 2030 · ▶ watch

“Daario has stated publicly that he expects disease... biology, and medicine to be solved by the end of the decade in the next 5 years.”

Your call:

open Alex Wissner-Gross: AI will effectively solve radiology diagnosis, surpassing board-certified radiologists.

· due: 2027 · ▶ watch

“I think we're going to see radiology get solved in the next year.”

Your call:

open Alex Wissner-Gross: Rare-earth and other key feedstock supply chains will eventually become hyperlocal, scavenged on-site by nanotech-enabled robots rather than shipped globally.

· due: 2035-2040 · ▶ watch

“I think ultimately that takes us to nanotech... robots that are able to scavenge raw resources from the immediate environment.”

Your call:

Numbers that matter

Nvidia Q3 revenue $57B, up 62% YoY Driven by AI compute demand; Nvidia also announced 10GW of new AI infrastructure deals and guided $65B revenue for next quarter.
Anthropic commits $30B to Microsoft Azure; Microsoft + Nvidia invest up to $15B into Anthropic Framed as formation of an AI power block rather than a simple product deal.
Elon Musk's plan: 100 gigawatts/year of solar-powered AI satellites in orbit US total average electricity load is ~460GW, so this represents roughly a quarter of total US electricity output, added annually.
Moon-mining mass-driver plan could scale to 100 terawatts/year of energy That is roughly five times the total current energy output of the entire Earth, per Elon Musk's estimate relayed by the panel.
First H100 launched to orbit weighs 50kg and is cooled using an aluminum-only radiator Panel notes 7% of Earth's crust is aluminum, making it an abundant cooling material; most of the mass is sensors that can be reduced.
Launch cost fell from ~$600M-$1B/Space Shuttle launch to $50-100M (SpaceX) to a projected $5-7M (Relativity Space) Roughly a 100x drop in launch cost, cited as evidence space-based compute economics are becoming viable.
China's year-over-year increase in electricity generation (~325 TWh) is about 4x the US increase (~80 TWh) China leads in new solar, wind, nuclear, and gas capacity, widening the US-China energy gap relevant to AI compute buildout.
Fusion timelines: first plants 2030-2035, mass production ~2040; reactivating idle fission plants takes 5+ years minimum Creates a projected 2030-2035 energy supply gap as AI chip fabs scale faster than new power generation can come online.
Google commits $40B through 2027 to Texas AI/cloud infrastructure, adding 6.2GW of new energy generation plus a $30M energy impact fund Texas framed as the state best positioned to build energy and data-center capacity fast enough for AI demand.
Zipline scaling to 20,000 autonomous delivery aircraft/year; ~15% week-over-week delivery growth sustained for 30 weeks; about 1 million commercial deliveries/year and 70-100 million autonomous miles flown; a delivery roughly every 30 seconds South San Francisco-based Zipline, founded 2014, started in Rwanda/Ghana delivering blood and medical supplies before scaling to US retail (Walmart, Chipotle).
$1.4 billion US government investment to build a domestic rare-earth supply chain (via ReElement/Vulcan) Responds to decades of rare-earth supply chain dependence on China despite the US having its own deposits.
Massachusetts' VC-backed companies have a 4.1% probability of an IPO within 10 years, vs. 2.3% for California and 1% for New York California still has about 10x more total IPOs than Massachusetts but only about 6x the population; data from Link Exponential Ventures, a Boston-based VC firm.

Worth digging into

🕳️ Space-based AI data center economics

Elon Musk's 100GW/year orbital compute plan plus moon mass-driver mining could scale energy to 5x Earth's total output, and the first H100 is already operating in orbit cooled by aluminum alone — a concrete near-term engineering story, not sci-fi.

🕳️ TRISO/pebble-bed nuclear fuel manufacturing

X-Energy's first domestic Cat 2 nuclear fuel facility just broke ground after decades of delay, and the panel claims the design literally cannot melt down.

🕳️ The robot training-data race

Alex Wissner-Gross frames training data, not algorithms, as the real bottleneck in robotics; competing approaches (Sunday Robotics' teleoperation gloves, Figure's palm cameras, 1X's teleoperation pipeline) are racing to solve it, with no clear winner yet.

🕳️ Anthropic's vertical integration into chips and data centers

Alex Wissner-Gross says Anthropic is the last major frontier lab without its own chip/data-center stack and is now moving into both, a structural shift reminiscent of Detroit's major automakers.

🕳️ Boston/MIT vs. Silicon Valley startup pipeline economics

Dave Blundin's claim that MIT produces more unicorn founders than Stanford, and that Massachusetts leads the US in VC-backed IPO probability, runs against the conventional wisdom that Silicon Valley is the center of the startup universe.