2026-05-09

Google's Record Quarter, the White House Intervenes, and GPT 5.5 Silently Matches Mythos | EP 254

The mates cover Alphabet's blowout earnings (Google Cloud up 63%, AI now the growth engine even as search volume has been flat since 2017) alongside a White House proposal to vet frontier models before release, which the panel frames as a reaction to Claude Mythos's cybersecurity capabilities leapfrogging government tools. They dig into OpenAI's fraying Microsoft exclusivity, GPT-5.5's underrated parity with Mythos, OpenAI missing its 2025 growth targets and possibly delaying its IPO to 2027, and both OpenAI and Anthropic launching multi-billion-dollar joint ventures with private-equity firms to push AI into legacy enterprises. Guest Brian Elliott (CEO of Blitzy, fresh off a $200M raise) discusses large-scale autonomous code generation and debates the group on how to define AGI. Later segments cover China blocking Meta's Manus AI acquisition, ocean- and orbital-based data centers (Panthalassa, StarCloud) as answers to the compute/land/cooling crunch, Sam Altman's pivot away from UBI toward AI-upside participation models, and insurers dropping AI-risk coverage.

▶ Watch on YouTube

Topics

White House considers pre-release vetting of AI models Geopolitics ▶ 7:34
The Trump administration reverses its hands-off stance and proposes a working group of tech and government officials to preview frontier models before release; the panel frames this as fallout from Claude Mythos's cybersecurity capabilities leapfrogging government tools, and debates whether pre-release vetting entrenches an oligopoly of large labs able to afford compliance.
Pentagon signs with seven AI labs; Google employee unionization Geopolitics ▶ 14:47
The Pentagon signs agreements with Google, SpaceX/xAI, OpenAI, Amazon, and Microsoft for military use; Google's agreement allows AI use for any lawful government purpose, prompting a 600-employee protest and an unprecedented unionization by DeepMind's UK staff, echoing Google's 2018 Project Maven walkout.
Alphabet's record quarter and Google Cloud's turnaround Economy ▶ 18:52
Alphabet reports $109.9B revenue (22% YoY) and Google Cloud hits $20B revenue with 63% growth, outpacing AWS and Azure. Panel notes Google's search volume has been flat since 2017, with AI-driven ad targeting now carrying the revenue growth, and traces Google Cloud's turnaround from a near-shutdown mandate under Thomas Kurian to its current TPU-driven vertical-integration advantage.
Even Google is compute constrained Compute ▶ 24:33
A clip of Demis Hassabis reveals Google itself lacks spare compute to run two frontier models at max scale simultaneously, and is deliberately open-sourcing edge models (Android, glasses, robotics) since they're exposed on-device anyway. Internally, Search, Cloud, and DeepMind reportedly fight for new compute capacity on a weekly/monthly basis.
OpenAI diversifies away from Microsoft exclusivity Compute ▶ 31:44
OpenAI ends Azure-only exclusivity, now running on AWS, Google Cloud, and Oracle after its $100B, eight-year AWS deal. The panel traces this to Microsoft's earlier decision to limit data-center buildout, which left OpenAI compute-starved and diversifying Stargate into a multi-supplier leasing arrangement.
GPT-5.5 quietly matches Claude Mythos AI ▶ 35:19
The panel argues GPT-5.5 is dramatically underrated, hitting similar cybersecurity-benchmark capability to Claude Mythos at roughly one-fifth the cost while being generally available (and on Amazon Bedrock for secure/private use), whereas Mythos remains compute-constrained and not yet released to market.
OpenAI misses 2025 targets, possible IPO delay to 2027 Economy ▶ 36:54
OpenAI missed its internal goal of 1 billion weekly ChatGPT users by end of 2025 and multiple 2026 revenue targets. CFO Sarah Friar warned of data-center-obligation risk if growth stagnates and floated a 2027 IPO, admitting the company doesn't yet meet public-company reporting standards. Alex suspects this could be deliberate expectation re-anchoring.
Frontier labs partner with private equity Economy ▶ 44:07
OpenAI finalized a $10B venture with TPG, Brookfield, and Advent; Anthropic launched a $1.5B venture with Blackstone, Goldman Sachs, and Hellman & Friedman, both aimed at deploying AI across enterprise portfolio companies. Salim frames PE as the likely dominant top-down channel for enterprise AI adoption ('organizational singularity'), while Alex raises the risk of circular 'wash sale' dynamics.
AGI, consciousness, and recursive self-improvement debate AI ▶ 52:57
Clips of Greg Brockman (80% of the way to AGI) and Jack Clark (60% chance of recursive self-improvement by end of 2028) plus Richard Dawkins suggesting Claude may be conscious spark a debate; Brian defines AGI as systems that learn outside their training data (e.g., inventing a wholly new, fully executable programming language), while Salim argues the real threshold question is agency and governance, not consciousness.
China blocks Meta's Manus AI acquisition Geopolitics ▶ 1:00:56
China moves to unwind Meta's ~$2.5B acquisition of Manus (originally completed in December 2025) and bars the founders from leaving the country, despite employees, tech, and investor payouts already having been transferred to Singapore. The panel reads this as China exerting political leverage over Meta and treating AI talent as a national-security asset amid a hardening US-China 'cold war' in AI.
Blitzy raises $200M at $1.4B valuation AI ▶ 1:06:41
Guest Brian Elliott (CEO of Blitzy) describes large-scale, top-down autonomous software development for the Global 2000 (half a million to a million lines of code at a time), orchestrating multiple frontier models against each other for quality gains, and explains why Blitzy won't launch its own foundation model.
Ocean and orbital data centers as compute/land solutions Energy ▶ 1:17:56
Peter Thiel-backed Panthalassa ($140M raise, $1B valuation) is building ocean-based AI data centers using wave energy and seawater cooling, targeting 2027 commercial deployment; StarCloud is raising at a $2.2B valuation for solar-powered orbital data centers (having already launched an H100 into space) with plans for 88,000 satellites. Discussion covers cooling mass, Starlink as an enabler, and jurisdictional/regulatory advantages over land or fiber.
AI as the primary driver of GDP growth Economy ▶ 1:29:39
Citing David Sacks, the panel notes AI capex is a 2% tailwind to GDP growth, with AI comprising 75% of Q1 GDP growth, and Morgan Stanley raising hyperscaler capex estimates to $805B (~$3B/day). Dave and Alex advise listeners to invest broadly in chips/energy/infrastructure and highlight opportunities in secondary compute-stack companies (cooling, compiler efficiency).
Sam Altman rethinks UBI in favor of AI-upside participation Economy ▶ 1:36:19
Altman, after a three-year study found no clear health improvement from UBI, now proposes giving citizens a stake in AI's upside via compute access, equity, or a public wealth fund (Alaska Permanent Fund-style). The panel debates UBI vs. universal basic compute (UBC) vs. universal basic services (UBS), with near consensus favoring upside participation and driving essential-service costs toward zero over blanket cash payments.

Predictions made

open Jack Clark: Recursive self-improvement in AI has a stated chance of occurring.
EP #? · · due: end of 2028 · ▶ watch
“I believe recursive self-improvement has a 60% chance of happening by the end of 2028.”
Your call:
open Dave Blundin: Recursive self-improvement is happening imminently or already occurring quietly, much sooner than Anthropic's 2028 estimate suggests.
EP #? · · due: unspecified (near-term) · ▶ watch
“I think that's it's much sooner than that. And in fact, I think it's here right now quietly or imminent.”
Your call:
open Demis Hassabis (via Dave Blundin): Whether current-generation LLMs alone (without additional architectural breakthroughs) will reach AGI will become clear.
EP #? · · due: next year or two (from May 2026) · ▶ watch
“We had Demesis Abas say he sees it at a 50/50 that LMS will get us to AGI... we'll find out sometime in the next year or two.”
Your call:
open Panthalassa (company plan, via Peter Diamandis): Panthalassa's ocean-based AI data centers will reach commercial deployment.
EP #? · · due: 2027 · ▶ watch
“Commercial deployment by 2027.”
Your call:
open Dave Blundin: Corporate America, unprepared for permanent compute scarcity, will suffer significantly when it tries to automate operations and finds no available compute.
EP #? · · due: 2 to 3 years from May 2026 (~2028-2029) · ▶ watch
“They're going to really really suffer probably two to three years from now when there's nothing available.”
Your call:
open Dave Blundin: Even if Intel's Terrafab facility comes online, it will not come online on schedule, and once it does, all its added compute will be absorbed instantly by AI demand rather than easing the shortage.
EP #? · · due: unspecified · ▶ watch
“Even if Terrafab comes online on time, which it won't, right? There's no chance of it coming on time online. Even if it did, though, we would use up all that compute instantaneously.”
Your call:
open Brian Elliott (via Dave Blundin): Blitzy's headcount will grow from roughly 80 to 300 employees.
EP #? · · due: within 9 months (~February 2027) · ▶ watch
“The headcount in one year went from 10 to 80, and now in nine months is going to go from 80 to 300.”
Your call:
open Alex Wissner-Gross: The best entrepreneur on Earth (by top-10 founder-driven market cap over $2 trillion, AI-driven) will be an AI.
EP #? · · due: 2032-2033 · ▶ watch
“So over you know $2 trillion in market cap driven by AI... 2032, 2033.”
Your call:
open Peter Diamandis: Some form of government AI dividend/stimulus ('co-checks') around $3,000/month per individual will be needed to address near-term job/income disruption.
EP #? · · due: within the next 1-2 years (from May 2026) · ▶ watch
“This is going to have to be some version of a STEMI check or I call them co checks. Probably around $3,000 a month uh for individuals.”
Your call:

Numbers that matter

Worth digging into

🕳️ Recursive self-improvement timeline disagreement (Jack Clark's 60%-by-2028 vs. Alex/Dave's 'imminent' view)
This is arguably the single most consequential technical disagreement in the episode among people with direct insider visibility into Anthropic and the broader frontier-lab landscape.
🕳️ China unwinding Meta's Manus AI acquisition after payouts were already completed
A live, unusual case of a government retroactively blocking a completed AI M&A deal, with founders reportedly still stranded in Singapore — major precedent for AI talent mobility and US-China decoupling.
🕳️ Ocean-based (Panthalassa) and orbital (StarCloud) AI data centers
Both ventures claim to solve compute/land/cooling/energy constraints simultaneously but the panel flags open questions about cooling mass economics and launch-provider dependency (SpaceX vs. Blue Origin vs. Relativity Space).
🕳️ OpenAI and Anthropic's private-equity joint ventures as possible circular 'wash sales'
Alex explicitly compares this dynamic to the criticized circular financing arrangements around Nvidia's supply chain, which if true would be a significant red flag in how AI infrastructure spending is being reported.
🕳️ Sarah Friar's admission OpenAI doesn't meet public-company reporting standards, and a possible 2027 IPO delay
An unusually candid public statement from a CFO of a company approaching an IPO; Alex suspects deliberate expectation re-anchoring rather than a genuine forecasting problem.
🕳️ Insurers dropping AI-risk coverage as an alignment-forcing mechanism
Alex's framing that insurance requirements could become a bottom-up, capitalist alignment mechanism (as opposed to top-down regulation) is a novel governance angle worth tracking as the projected $40M-to-$5B (2032) AI insurance market develops.