Emad Mostaque

Peter Diamandis × Salim Ismail × Dave Blundin × Alexander Wissner-Gross × Emad Mostaque

Memory and reserved compute—not model quality alone—have become AI’s binding bottleneck, with an NVL72 order moving from $3.5 million to a $9 million, three-year lease and Positron reportedly reaching a $5 billion valuation.Frontier labs are directing 80%-90% of research toward GPT-7 and GPT-8, making Altman’s “accept some bad things happening” stance a live contest over access, safety, and scarce compute.

Peter Diamandis × Salim Ismail × Dave Blundin × Dr. Alexander Wissner-Gross × Emad Mostaque

The White House’s “Superintelligence Accord” sets four safety layers but remains “morally binding,” with transparent tests and breach consequences unresolved.Anthropic’s proposed IPO combines $4.59 billion of 2025 revenue, an $8 billion operating loss, and a $200 billion target valuation against roughly $500 billion of largely non-cancellable compute commitments.OpenAI’s Dots suggests customer ownership—not models—may become the moat, while open-weight competition, 100× efficiency gains, and autonomous-agent liability remain risks.

Peter Diamandis × Salim Ismail × Dave Blundin × Alexander Wissner-Gross × Emad Mostaque

AI progress is becoming a measurable cost collapse: Astra beat o3’s ARC-AGI-1 result for $20 versus $500,000, while OpenAI claimed a 10,000-agent Navier–Stokes breakthrough.Data delivered 12× compute-efficiency gains versus 3.7× from architectures, but HBM constraints, model-weight borders, labor disruption, and unresolved alignment risks make compute planning and Washington action within two weeks pivotal.

Peter Diamandis × Salim Ismail × Dave Blundin × Alexander Wissner-Gross × Emad Mostaque

OpenAI’s RL pause drew safety-versus-marketing disagreement, while recursive self-improvement may command more revenue per token than enterprise codegen, keeping frontier models internal.Memory prices rose 500% in 12 months as AI demand grows 200% versus 20% supply; Unitree hit 12.66 m/s and Zipline targets over 1 million daily deliveries, while etched-weight supply chains, regulation and Uber’s aggregator position remain risks.

Peter Diamandis × Salim Ismail × Dave Blundin × Alexander Wissner-Gross × Emad Mostaque

NVIDIA is moving from chip vendor to financing architect, partnering with Apollo, BlackRock, Blackstone, Brookfield and KKR to mobilize over $500B for customer compute.The scalable AI-factory model faces stranded-asset risk from unpredictable depreciation, while xAI’s Grok 4.6 uses Cursor reasoning traces and NVIDIA GPUs, with Emad Mostaque saying Grok 4.7 could reach #1 within weeks.

Peter Diamandis × Salim Ismail × Dave Blundin × Alexander Wissner-Gross × Emad Mostaque

SpaceX guided to $100B+ annual recurring revenue by December and moved its $1T target to 2030, with “a nonzero chance” of 2029.Terafab’s initial $16.8B investment and confirmed free-electron laser could onshore the semiconductor stack and challenge ASML.Astra’s 249-page math manuscript, produced for an estimated $2,000 compute, sharpens the capability-cost race.

Peter Diamandis × Salim Ismail × Dave Blundin × Alexander Wissner-Gross × Emad Mostaque

Kimi K3 puts a 2.8-trillion-parameter Chinese open-weight model on the frontier, ranking first in frontend coding and six other domains while reaching the third point on Artificial Analysis’s cost-performance frontier.Its recognizable transformer architecture, constrained-chip optimization, and expected July 27 weight release could pressure closed-model margins and valuations, while enterprise value shifts toward proprietary data, verification, support, and model-swapping infrastructure.

Peter Diamandis × Emad Mostaque × Dave Blundin × Alexander Wissner-Gross

Washington is placing frontier-model releases inside a customer-by-customer approval loop, reportedly limiting Anthropic’s Mythos 5 to 100 companies and OpenAI’s GPT-5.6 variants to 20 amid cybersecurity and recursive self-improvement concerns.Harnesses can lift older or open models above restricted systems—GLM 5.2 reportedly topped Frontier SWE after orchestration—making prompt logging, licensing, trusted-model certification, and geopolitical AI blocs more likely than controls focused only on weights.

Peter Diamandis × Salim Ismail × Dave Blundin × Dr. Alexander Wissner-Gross × Emad Mostaque

Recursive AI self-improvement and GPT-5.4’s reported 38% FrontierMath Tier 4 score point to reasoning and computer-use capability advancing beyond conventional chatbot benchmarks.Agents are becoming a new customer class as synthetic experimentation, AutoResearch, and faster inference expand the addressable market while compute, memory, electricity, security, and scarce attention remain bottlenecks.Whether a handful of frontier labs consolidate the value—or open models and tiny systems redistribute it—remains the key competitive question.

Peter Diamandis × Emad Mostaque × Salim Ismail × Dave Blundin × Alexander Wissner-Gross

Google’s Titans and MIRAS use surprise-based memory, potentially easing transformer context costs beyond 2 million tokens while visual chain-of-thought adds 3% to 6% reasoning gains.Chinese-affiliated first authors rose from 9% to 30% at ICLR as the U.S. share fell from 52% to 36%, making open weights a distribution and integration advantage.Weekly leapfrogging, falling token prices and rising agent consumption intensify compute needs, while Anthropic’s possible 2026 IPO and trust—not benchmarks alone—could shape capital and adoption.