Meta’s 15 Billion Dollar AI Bet & The Race To ASI w/ Salim Ismail & Dave Blundin
Peter Diamandis convenes his regular moonshot mates, Salim Ismail and Dave Blundin, for a rapid-fire tour of the biggest AI stories of early summer 2025. They dig into the blurring line between AGI and ASI (sparked by Elon Musk's 'this year or next' prediction), Ilya Sutskever's $32B SSI valuation, and Meta's winner-take-all talent war including $100M signing bonuses and its 49%-stake acquisition of Scale AI. The panel covers government AI adoption (AI.gov, tech execs becoming Army lieutenant colonels), warns that AI is already eroding critical thinking per an MIT study, and surveys robotics and robotaxis (Tesla vs Waymo) alongside a growing backlash against self-driving tech. The back half pivots to the US-China energy race, where China's nuclear and solar buildout dwarfs the US, and closes with crypto notes on Bitcoin price targets and Circle's stablecoin IPO enabling AI agent micropayments.
AGI vs ASI definitions and Elon's ASI timelineAI▶ 6:09
The panel reacts to an Elon Musk clip predicting digital superintelligence this year or next, debates how blurry AGI/ASI definitions really are, and lands on Dave's preferred framing of tracking capability 'swim lanes' (math/code vs biology) instead of chasing a single finish line.
Winner-take-all ASI and nationalization riskAI▶ 12:01
Prompted by a Jeff Clune clip arguing 'the first ASI may be the last ASI,' the group debates whether a government would nationalize or heavily control the first company to build a superintelligence, with Dave betting on a winner-take-all dynamic and Salim pushing back.
Ilya Sutskever's SSI and its $32B valuationAI▶ 17:38
Dave floats a theory that SSI's investors (a16z, Sequoia, DST, Alphabet, Nvidia) are betting purely on Ilya's promise of a 'safe' self-improving ASI strategy, with no product to show, because being first and safe would make the bet unbeatable.
Meta's AI talent war and the Scale AI deal structureAI▶ 22:11
The panel unpacks reports of Meta offering $100M signing bonuses to poach OpenAI staff after Llama 4 underperformed, and Dave explains why Meta structured its $14.8B Scale AI investment as a 49% non-voting stake to dodge antitrust review and financial consolidation rules.
What makes winning AI startup teams (Fred Wilson rule)Economy▶ 39:42
Dave shares the 'Fred Wilson rule' (fund technical best-friend founding teams of three or more) and the group discusses why bonded teams (Israeli military service, MIT's brutal problem sets) resist being poached, unlike strangers thrown together.
The Trump administration's AI.gov project, led by a Tesla engineer, aims to bring AI into GSA, DOT, DOE, FAA and FDA; the panel argues ideas must come from compartmentalized private-sector AI modules rather than the government itself.
Big tech execs appointed Army lieutenant colonelsGeopolitics▶ 54:49
Palantir, Meta and OpenAI executives were appointed lieutenant colonels in a new US Army unit without traditional training; the panel largely defends the move as a fast way to get elite AI talent into the military despite social media backlash.
A DeepMind weather model trained on 5,000 cyclones over 45 years improved 5-day track predictions by 140km, illustrating the 'bitter lesson' that data-plus-compute beats decades of hand-built physics models; the panel wants an X Prize for earthquake prediction next.
Mattel/OpenAI smart toys and AI's effect on kidsAI▶ 1:03:07
Mattel and OpenAI's partnership to bring GPT-5 into Barbie, Hot Wheels and American Girl dolls sparks debate over the upside (sparking imagination, helping neurodiverse kids) versus the risk of kids preferring AI companionship to real friends.
AI's effect on critical thinking (MIT study)AI▶ 1:06:01
A Time/MIT study found ChatGPT users couldn't recall what they had just 'written,' which the panel compares to over-reliance on GPS; Salim is 80% unnerved after watching his 13-year-old lose all memory of an AI-written essay, but also pushes for teaching critical thinking directly.
AI job automation, reskilling, and the Luddite parallelEconomy▶ 1:13:26
A Stanford survey of which jobs AI will likely automate first (bookkeepers, payroll clerks, data entry) leads the panel to draw an explicit parallel to the 19th-century Luddite revolt, warning of real backlash once job displacement becomes visible in the next 2-3 years.
Robotics, robotaxis, and the Tesla-vs-Waymo betRobotics▶ 1:17:10
Amazon testing humanoid delivery robots and Tesla's cheap camera-only robotaxi launch in Austin set up a Tesla-vs-Waymo debate: Dave bets Tesla wins 70/30 on its consumer-funded fleet model despite Waymo's superior LiDAR-based sensor stack.
US vs China energy race (nuclear and solar)Energy▶ 1:40:06
The panel is alarmed that the US has licensed only 2 new nuclear reactors this century versus China's one every 52 months, and that China alone deployed solar capacity equal to half of total US energy output in 2024, calling short election cycles America's structural 'Achilles heel.'
Energy storage breakthroughs for solarEnergy▶ 1:46:35
Solar's intermittency problem (lithium storage costs ~5x the panels) could be solved by Bill Gross's gravitational storage towers or a future AI-discovered reversible chemical battery, which the group floats as a trillion-dollar opportunity.
Predictions made
openElon Musk (clip): Digital superintelligence (AI smarter than any human at any task) will arrive this year or by the end of next year.
“Solar power in China will exceed all sources of electricity combined in the US in three to four years”
Your call:
Numbers that matter
Meta worth $1.8 trillion with $70 billion in cash on handUsed to explain how Meta can afford $100M AI signing bonuses and acquisition offers.
SSI raised $6 billion at a $32 billion valuation with no public productInvestors include a16z, Sequoia, DST Global, Alphabet, Nvidia, Lightspeed.
Meta reportedly offering $100 million signing bonuses to poach OpenAI talentBonuses described as having no vesting cliffs, per Sam Altman's comments.
Meta paid $14.8 billion for a 49% non-voting stake in Scale AIStructured to dodge antitrust review and financial consolidation thresholds.
Cursor hit $500 million ARR in under three years, the fastest SaaS growth on recordCompared against Anthropic, Uber, and OpenAI's own growth curves.
Top-quartile AI startup: $3.1M raised pre-Series A leading to $8.7M ARR and a Series A in 5 monthsFrom an Andreessen Horowitz chart on capital-efficient AI app growth.
DeepMind cyclone model improved 5-day storm-track prediction by 140km, trained on 5,000 cyclones over 45 yearsCyclones have caused an estimated $1.4 trillion in economic losses over 50 years.
Stanford survey of 1,500 workers/AI experts: 69.4% want AI to focus on high-value work, 46.6% want it to take repetitive junk workUsed to identify which jobs (bookkeepers, payroll clerks, data entry) are seen as most automatable.
US has 94 total nuclear reactors and added only 2 this century; China has 58 reactors and is building one every 52 monthsUS nuclear licensing alone takes 10 to 12 years.
China deployed ~700GW of manufactured solar panels and ~250GW of peak solar capacity in 2024 alone, vs total US energy capacity of 1.2TWFramed as China building the equivalent of half of total US energy output in solar in a single year.
Covering 0.1% of Earth (~150,000 km², the size of South Dakota) with 20%-efficient solar panels would generate 200,000 TWh/year, exceeding global demandIllustrates how little land area would be needed to meet world energy needs with solar.
Circle's IPO priced at $31/share and peaked at $300/shareFramed as proof the IPO/acquisition market has reopened for tech and crypto-adjacent companies.
A 2011 three-day BlackBerry outage correlated with a 40% drop in the Abu Dhabi car-accident rateCited as evidence that phone distraction, not driver skill, is the dominant cause of modern car accidents.
Worth digging into
🕳️ Ilya Sutskever's SSI safety strategy
SSI raised $6B at a $32B valuation with no public product, purely on Ilya's promise of a verifiably 'safe' self-improving ASI approach.
Dave calls this deal structure (dodging Hart-Scott-Rodino review and consolidation rules) 'the deal structure of the future' for Big Tech AI acqui-hires.
🕳️ Bill Gross's gravitational energy storage (Energy Vault)
Pitched as a working, contracted alternative to lithium batteries for solving solar's intermittency problem at data-center scale.
🕳️ China's 2030 nuclear and solar buildout pace
China is adding solar capacity equal to roughly half of total US energy output in a single year and building nuclear reactors every 52 months versus a 10-12 year US licensing timeline.
🕳️ The MIT/Time critical-thinking study on ChatGPT use
Cited as hard evidence that AI writing assistance erodes recall and comprehension, but the panel disagrees sharply on how serious this is.
🕳️ Circle's stablecoin rails enabling AI agent micropayments
The panel frames Circle/USDC as the 'missing financial layer' that could shift AI pricing from flat subscriptions to usage-based microtransactions.