Elon's $60B Cursor Bet, Claude kills SaaS, and OpenAI's Mass Departures | EP #249
The Moonshots panel opens with Peter's night at the Breakthrough Prize before diving into a wave of AI-industry shocks: Anthropic's Claude Design tanking Figma and Adobe stock, an internal Anthropic estimate that entry-level engineers get replaced within three months, Elon's tweeted Grok 4.4-through-7 roadmap, xAI's new text-to-speech product, three senior OpenAI executives departing amid an IPO push, OpenAI's compute-heavy new image model, and SpaceX's negotiated right to buy Cursor for $60 billion. The group frames nearly all of it as different fronts in a single race to dominate AI code generation and recursive self-improvement. Later segments cover a chart showing AI data-center capex dwarfing Apollo and the Manhattan Project as a share of GDP, US manufacturing's reshoring rebound, San Francisco's market cap surpassing all of China, UAP/UFO disclosure news tied to missing scientists, China's solar and humanoid-robot dominance, collapsing trust in mass media, Apple's CEO transition, and the Iran war's cascading effects on oil, helium, natural gas, and fertilizer supply chains. The episode closes with the panel responding directly to listener pushback on entrepreneurship, data-center water use, and accusations of an out-of-touch, "ivory tower" tone.
Anthropic's Claude Design (built on Opus 4.7) immediately dropped Figma stock 10% and Adobe about 2%. The panel frames this as an 'unhobbling' of capabilities frontier models already had, arguing many SaaS verticals (legal research, BI, medical documentation, financial modeling, HR) are one unhobbling away from being cratered by the base models themselves.
Following Dario Amodei's longstanding claim that AI will wipe out 50% of entry-level white-collar jobs in 1-5 years, an internal (small-sample) Anthropic employee survey estimated entry-level software engineers and researchers will be replaced by Anthropic's internal coding model 'Mythos' within three months, prompting debate about apprenticeship pipelines and career ladders.
Grok's parameter-count roadmap vs. intelligence densityCompute▶ 26:52
Elon Musk tweeted a Grok roadmap: 4.4 (1T params) in early April, then 4.5 (1.5T), 4.8, 4.9, with Grok 5 pegged as AGI and Grok 6/7 as ASI. Alex Wissner-Gross argues raw parameter bragging is a bad sign versus racing toward 'intelligence density' (compression/distillation), comparing it to the end of Moore's-law-era gigahertz scaling.
xAI launches Grok text-to-speech vs ElevenLabsAI▶ 34:02
xAI shipped a standalone Grok speech API at a 5% phone-call error rate versus ElevenLabs' 12%, priced at 10 cents/hour across 25 languages. Panel debates whether this is Elon's usual vertical-integration playbook aimed at dominating media generation via X, and concludes ElevenLabs isn't seriously threatened given the enormous unpenetrated voice/call-center market.
Three senior OpenAI leaders (Kevin Weel - VP of Science, Bill Peebles - head of Sora, Srinivas Narayanan - CTO of B2B) departed on April 17th. The panel ties this to OpenAI's recent $120B+ fundraise, its push toward an IPO, and a narrowing focus on the innermost loop of recursive self-improvement (code generation) at the expense of broader research.
OpenAI released a new image model with 99% text accuracy, tool use, web search, and multi-image generation, topping the arena.ai text-to-image leaderboard. The panel questions why OpenAI would ship such a compute-intensive consumer feature mid-race with Anthropic on code generation, speculating about algorithmic efficiency gains or images as an intermediate representation for coding.
SpaceX negotiates right to buy Cursor for $60BCompute▶ 57:09
SpaceX secured an option to acquire coding-tool company Cursor for $60 billion (or pay a $10B walkaway fee). The panel reads this as xAI/Grok admitting its code-gen isn't competitive with Claude, SpaceX buying 'Anthropic-level' user interaction data via Cursor, and SpaceX's Colossus GPU cluster starting to open to third-party tenants like an early Dyson-swarm cloud.
AI data-center capex vs historic megaprojectsEconomy▶ 1:09:19
A chart shows data-center capex approaching $1 trillion this year (projected $6T over 6 years), outpacing the Apollo program, Manhattan Project, and Interstate Highway System as a share of GDP by roughly 5x, driven entirely by private capital rather than government funding.
US manufacturing reshoring reboundEconomy▶ 1:20:00
US manufacturing saw 17 quarters of contraction (tariffs, COVID supply-chain collapse, chip shortages, 2008-crisis hangover) followed by 16 quarters of acceleration, credited to the CHIPS Act, Inflation Reduction Act, infrastructure spending, and AI-driven robotic reshoring.
Space race: Blue Origin's New Glenn and the Artemis Moon contractSpace▶ 1:32:39
Blue Origin's third New Glenn launch reused its booster successfully but lost an AST satellite to the wrong orbit. Panel discusses SpaceX vs Blue Origin competing for NASA's Artemis 4 lunar lander contract, Jared Isaacman's revised Artemis roadmap, and Relativity Space as a dark-horse booster competitor.
UAP disclosure and missing scientistsGeopolitics▶ 1:33:31
Trump announced accelerating UFO/UAP file declassification; separately, ten scientists with classified nuclear/aerospace access have gone missing or turned up dead, which Congress is now treating as potentially connected. Alex Wissner-Gross argues that if decades-long alleged reverse-engineering cover-ups are real, it constitutes a crime against humanity that set back progress by a century.
Iran war and the Strait of Hormuz supply-chain shockGeopolitics▶ 2:02:51
The US and Israel's strikes on Iranian nuclear facilities led to the Strait of Hormuz closing, threatening roughly a quarter of world oil supply and hitting Korea and Japan hardest. Panel flags cascading shortages in helium (chip fabrication), Taiwan's natural gas reserves (semiconductors), global fertilizer, and jet fuel.
Predictions made
openElon Musk (clip): Grok will follow a roadmap of 4.4 (1 trillion parameters, early April), then 4.5 (1.5 trillion), 4.8, 4.9, with Grok 5 achieving AGI and Grok 6/7 achieving ASI and 'ASI 2.'
“Grok 4.4 will be twice the size of 1 trillion parameters of 4.3... Grok 5 he predicts to be AGI, Grok 6 ASI, Grok 7 ASI 2.”
Your call:
openAlex Wissner-Gross: A new, well-funded frontier AI lab will emerge from OpenAI's current executive exodus, similar to how the original 'Anthropic schism' produced Anthropic.
“UAP disclosure in connection with the president's executive order is scheduled to conclude no later than January of 2027.”
Your call:
openAlex Wissner-Gross: If humanity is alone in the galaxy, AI and space advances will let civilization convert the entire Milky Way to computronium within a few years; any non-human intelligence wanting to prevent that will need to make a 'cameo appearance' soon.
“If there were a time for any non-human intelligence elsewhere in our Milky Way galaxy to make a cameo appearance... I think they're going to have to make a cameo appearance sometime in the next few years.”
Your call:
openDave Blundin: Annual spend on AI data centers as a share of GDP will approach 100% once physical goods/robotics built in service of AI are counted.
“I think over time it should get higher and higher because... everything becomes computronium and everything should be in service of building that.”
Your call:
openAlex Wissner-Gross: AI data-center spending as a share of GDP will peak at roughly a quarter to a third of GDP, then decline as automation (robots, nano-assemblers) creates offsetting cost efficiencies.
“Still still expected this month, possibly... Still still expected based on everything I'm hearing. Still expected eminently.”
Your call:
openAlex Wissner-Gross: New Apple CEO John Ternus will steer the company toward breakthrough hardware devices and a new device-oriented AI strategy, analogous to Microsoft's Ballmer-to-Satya turnaround.
“I would expect a new device-oriented AI strategy from John Turnus.”
Your call:
openDave Blundin: If major AI labs devote even 10% of their effort to a deliberate roadmap for displaced workers (starting with software engineers, then legal and other professional services), the transition disruption will largely be resolved.
Compute by end of year: xAI ~2GW, OpenAI 1.2GW, Meta ~1GW, Anthropic+AWS combined ~1GWCompute capacity comparison across frontier labs
Grok TTS: 5% error rate on phone calls vs ElevenLabs' 12%; priced at 10 cents/hour across 25 languagesxAI's new speech API vs incumbent ElevenLabs (valued at $6B)
Three senior OpenAI leaders departed April 17th: Kevin Weel (VP of Science), Bill Peebles (head of Sora), Srinivas Narayanan (CTO of B2B)OpenAI executive exodus
OpenAI's recent fundraising round exceeded $120 billionCited as context for the IPO push and departures
Mark Chen was offered ~$1 billion by Meta and turned it downIllustrating OpenAI's extreme compensation dynamics
Sam Altman gave nearly every OpenAI employee a $1 million spot bonusResponse to the Mark Chen retention episode
Each 0.1% equity stake in OpenAI's new funding round equals roughly $1 billionIllustrating scale of unvested equity departing execs forfeit
GPT Image-2 scored 1,512 Elo on arena.ai's text-to-image leaderboard, ranked #1, ahead of Nano Banana 2Benchmark for OpenAI's new image model
ChatGPT Images 2.0 reports 99% text accuracyHeadline capability claim in OpenAI's launch video
SpaceX negotiated the right to buy Cursor for $60 billion, or pay a $10 billion walkaway feeSpaceX-Cursor acquisition option
SpaceX's Colossus cluster represents roughly a million H100-equivalent GPUsScale of SpaceX's compute buildout
Worth digging into
🕳️ SpaceX's $60B option to buy Cursor
Signals xAI may be resetting its code-gen strategy entirely and SpaceX is turning its Colossus GPU cluster into a third-party cloud, an early real-world version of the panel's recurring 'Dyson swarm' thesis.
A very aggressive, specific timeline sourced from a small internal Anthropic employee survey, cited alongside Dario Amodei's broader 50%-of-entry-level-jobs claim.
🕳️ Age of Disclosure documentary's UAP/NHI allegations
Alex Wissner-Gross calls the alleged 80+ year reverse-engineering cover-up a potential 'crime against humanity' that set back civilization roughly a century, and cites a live executive order with a January 2027 disclosure deadline.
🕳️ 10 missing or dead scientists with classified access
Raised alongside UAP disclosure as possibly connected; Rep. Ogles' tweet claims 'just knowing it exists makes you a target,' and Congress is reportedly now treating the deaths as linked rather than coincidental.
🕳️ Strait of Hormuz closure's cascading shortages
Panel flags multiple under-the-radar supply-chain chokepoints (Qatari helium for chip fabs, Taiwan's 11-day gas reserve, global fertilizer, European jet fuel) stemming from the Iran war that could hit semiconductors and food production hardest.
🕳️ Intelligence density as the real AI race
Alex Wissner-Gross argues the meaningful competitive axis isn't parameter count but compressing capability via distillation/quantization, a contrarian read on Elon's raw-parameter Grok roadmap.