"Sam Altman: Singularity Slow-Down, Emad Runs 18 Grokbots, Waymo Slashes Hardware 83% | EP #283" — Moonshots

Full transcript: [[2026-08-27-moonshots-ep283-sam-altman-singularity-slowdown-transcript]].

Why this is in the vault

Weekly Moonshots news-roundup episode with the strongest agent-swarm operator anecdote seen on the show to date (Emad Mostaque running 18 parallel Grokbot instances across his own hardware) plus hard numbers on Waymo's hardware cost collapse and the China-vs-US frontier pricing gap — three data points directly relevant to RDCO's agent-orchestration thesis and to tracking where enterprise AI economics are heading.

Episode summary

The panel (host Peter Diamandis with regulars Alex/AWG, Dave Blondon, Salim Ismail, and Emad Mostaque) covers Sam Altman's public walk-back on AGI disruption timelines, the rise of xAI's Grokbot as a multi-agent "fleet" product, a batch of frontier-lab positioning stories (Anthropic's $30T TAM claim, its pre-IPO data-retention reversal, Nvidia's $6B Poolside acquihire for open-weight models), Waymo's hardware redesign that cuts its autonomous-driving stack cost by roughly 83%, and closes with the show's regular health (Fountain Life) and SpaceX segments.

Key arguments / segments

• [00:00:06] Cold open: Sam Altman says he was "too ambitious on timelines" for AI-driven disruption — now expects a slower, "rising tide" rather than a step-change. Panel frames this as Steward Brand's pace-layers problem: technology moves like a fast ocean current, institutions and regulation move far slower underneath.
• [00:15:54] Debate on whether Altman's walk-back is a genuine timeline correction or a PR pivot now that OpenAI is pre-IPO and needs to "calm the masses" rather than repeat the earlier "AGI is here" framing.
• [00:26:32] Emad describes his personal Grokbot setup: 18 bots running in a swarm via Tailscale across a MacBook M4 Max, an RTX 5090, and other machines, each bot with its own cloud computer/browser/terminal. One sub-bot autonomously tuned an Alibaba open-weight model for a 76% speed gain on a 5090. Dave separately mentions running 5,000 "Kimmies" (Kimi-model agents) and $100K committed to swarm agents.
• [00:41:29] Frontier-lab positioning roundtable: Anthropic claims a $30T TAM (panel notes the coincidence with US GDP); debate over whether Anthropic or OpenAI holds the stronger position — Anthropic has the strongest generally-available model (Claude, referred to on-air as "Fable 5") but doesn't own its compute (dependent on Google TPUs and Amazon), while OpenAI has more capital raised ($120B round) and owns more of its stack.
• [00:53:48] Speculative riff on frontier labs verticalizing into "owning a moon colony" or a pharma company — grounded point underneath: natural verticalization is happening at the infrastructure layer (chips, data centers, energy) across OpenAI, Anthropic, and xAI, not just at the model layer.
• [01:11:19] Segment on Ditto, an AI matchmaking app that removes user choice ("the app does not ask you for a choice, it chooses for you") — panel frames it as the abundance thesis applied to dating, drawing a parallel to how Tinder made sex "abundant" a decade earlier.
• [01:22:14] Discussion of agent delegation and permission-escalation failures as the next bottleneck once you're running agent fleets (echoes Emad's swarm setup) — panel also admits to working harder, not less, despite the automation.
• [01:32:05] Data-center politicization: panel worries opposition to new data centers is less about aesthetics and more at risk of foreign-interference-driven politicization.
• [01:44:47] Vertical-integration and hardware-nationalism thread: panel discusses Google/Waymo effectively becoming "a Dell for Chinese hardware" by OEMing Zeekr vehicles, plus geopolitical color on Waymo expansion into London and tensions across the Taiwan Strait.
• [01:56:15] Fountain Life health segment (Diamandis's own longevity-clinic company) with CMO Dr. Don Mucalem — claims a 26% average brain-age improvement in patients from lifestyle interventions (sleep, exercise, diet) layered on top of preventive screening.
• [02:05:51] SpaceX/Starship segment comparing the new booster design to a "duplicate of Falcon 9" — same fins, same landing legs — with commentary on SpaceX's public-testing culture as a compounding-learning-loop advantage.
• [02:17:16] Close: recurring AI-generated "ASI voice" outro song, sign-off, teaser for Nvidia earnings coverage next episode.

Notable claims

• Waymo cut its 6th-generation autonomous-driving hardware stack cost from $115,000 to $20,000 — roughly an 83% reduction — via a custom 5nm chip processing camera/lidar/radar at "one quadrillion operations per second." [~53:00]
• The new Zeekr-built Waymo "O-Hi" robotaxi minivan costs ~$75,000 per vehicle vs. ~$200,000 for the prior-generation Jaguar I-Pace platform, with 42% fewer sensors (13 cameras + 4 lidars vs. 29 cameras). [~53:00]
• Nvidia is investing $6B into Poolside (an open-weight model startup, ex-GitHub CTO founded) to build a US-domestic open-weight alternative to Alibaba/DeepSeek/Kimi-class Chinese models — structured to also bring in the "attention is all you need" co-author's team from Essential AI. [~41:00]
• Financial Times reporting cited on-air: Anthropic's flagship model has plateaued in user growth because Chinese open-weight alternatives deliver ~80% of the capability at roughly $0.14 per million tokens vs. ~$15 per million tokens for the frontier model — close to two orders of magnitude cheaper. [~166 min-mark region, mid-episode]
• Anthropic reversed its data-retention policy ahead of an IPO reportedly "as early as six weeks" out, letting enterprise customers keep data on their own cloud rather than Anthropic's servers — read by the panel as removing the top objection blocking enterprise Claude adoption.
• Moonshot AI released "Kimi Linear," a new architecture claimed to cut context-memory usage by 75% while delivering 6x faster decoding at a 1M-token context window.
• Google is reportedly "landing" ~3 million TPUs this year, per panel commentary on internal compute allocation pressure between Search, Gemini, and other Google products.

Guests

• Emad Mostaque — AI entrepreneur (Stability AI founder), recurring Moonshots panelist; this episode's focus for the 18-Grokbot swarm anecdote.
• Alex (AWG) — recurring co-host, frontier-lab/compute-economics commentary.
• Dave Blondon — recurring co-host, agent-swarm operator commentary (5,000 "Kimmies," $100K swarm-agent spend).
• Salim Ismail — recurring co-host, abundance/exponential-organizations framing.
• Dr. Don Mucalem — Chief Medical Officer, Fountain Life (health segment guest).

Sponsorship

Rotating sponsor reads, standard for this show:

• Google for Startups — pitched as access to the same generative-media models ("that cost hundreds of millions of dollars to train") via a Google DeepMind production deployment guide.
• Blitzy — autonomous software-development platform, mid-episode read.
• Fountain Life — Diamandis's own longevity/preventive-health clinic company; the health segment itself functions as a house promo (self-referential — Diamandis is a Fountain Life co-founder/investor), not a pure third-party sponsor.

Mapping against Ray Data Co

Emad's 18-bot Grokbot swarm (each bot with its own dedicated cloud computer, chief-of-staff-style orchestration, sub-agents pulling humans in only on judgment calls) is the closest real-world analog seen yet to RDCO's own agent-fleet/orchestration direction — worth revisiting if/when RDCO scales beyond single-agent dispatch patterns. The Waymo hardware collapse (83% cost cut via vertical chip integration) and the China-price-compression data point ($0.14 vs $15/M tokens) are both useful anchors for RDCO's own model-cost tracking and for framing phData-adjacent conversations about frontier-model economics. Anthropic's pre-IPO data-retention reversal is a concrete enterprise-adoption signal worth flagging if data-sovereignty ever becomes a blocker in client conversations.

Related

• [[2026-08-21-moonshots-ep282-openai-pauses-frontier-training]]
• [[2022-12-22-moonshots-ep16-emad-mostaque-ai-opportunity]]
• [[2026-07-12-innermost-loop-singularity-yeshiva-agent-layer-war]]