All-In PodcastWhy OpenAI's Code Red signals AI is already fragmented
Altman halted all side projects at OpenAI just to defend ChatGPT; Gemini and Claude carve separate niches while the New York Times targets Sacks.
CHAPTERS
- 0:00 – 1:46
Besties reunite, Sam Altman’s ‘Code Red,’ and the competitive threat to ChatGPT
Jason opens with the news that Sam Altman has declared a “code red” at OpenAI, urging staff to drop side projects and refocus on the core ChatGPT experience. The group frames this as a response to fast-improving competitors (Gemini, Grok, Claude) and shifting market dynamics between consumer and enterprise.
- •Altman memo: stop “side quests” (ads, etc.) and optimize ChatGPT core
- •Competitors cited as closing the gap or outperforming in key areas
- •OpenAI’s mix: consumer-heavy versus rivals’ enterprise momentum
- •Charts discussed: market share/traffic trends and infrastructure deals vs revenue
- 1:46 – 4:45
Chamath’s strategic vs tactical read: early market, distribution wins, focus matters
Chamath argues it’s too early to pick winners at the model layer, even if the silicon layer is more settled. He stresses that distribution is a decisive advantage and that “code red” is a useful forcing function to re-focus a fast-growing org.
- •AI market is still early; hard to call winners beyond compute/silicon
- •Distribution advantages favor Google, Meta, and still OpenAI (large MAUs)
- •‘Code red’ as an organizational refocusing tool as companies bloat
- •Analogy to Facebook vs Myspace: better product can overtake incumbents
- 4:45 – 6:40
Friedberg on Google’s ‘Project Canada’: war rooms, existential threats, and market-share decline
Friedberg recounts Google’s historical “code red” response to Microsoft, including war-room cadence and recruiting tactics. He connects the pattern to AI, showing how OpenAI’s early dominance naturally invited share loss as Google’s flywheel ramps.
- •Google’s internal code red: ‘Project Canada’ vs Microsoft
- •Threat-driven focus as a proven innovation catalyst (moonshot analogy)
- •OpenAI traffic share down from near-monopoly as Gemini rises (~14–15%)
- •Google’s data/distribution flywheel as a structural advantage
- 6:40 – 9:44
Sacks’ ecosystem map: five major players, specialization, and a ‘Goldilocks’ competitive market
Sacks praises Altman for publicly acknowledging the threat and using urgency to drive execution. He outlines each major AI player’s strengths and argues that specialization and leapfrogging create healthy competition rather than monopoly consolidation.
- •Credit to CEOs who admit problems despite negative PR risk
- •Player strengths: OpenAI (consumer), Google (search integration), Anthropic (enterprise/coding), xAI (current events + X), Meta (deep pockets)
- •Leaderboards show rapid leapfrogging and benchmark churn
- •Competition reduces monopoly power; ‘Goldilocks’ scenario for consumers
- 9:44 – 10:33
US vs China and why competition is America’s advantage
Prompted on China, Sacks contends that open competition is a strength of the U.S. innovation system. The group frames the AI race as a continuing horse race where competitive pressure accelerates progress.
- •Competition as the engine of U.S. technological progress
- •China remains formidable but tends to anoint national champions
- •AI leadership framed as an ongoing race, not a settled outcome
- •Optimism that U.S. market dynamics can sustain leadership
- 10:33 – 17:56
‘ChatGPT vs the world’: market-share erosion, founder rivalries, and the coming price war
Jason argues OpenAI’s market share is declining faster as distribution and bundling kick in. He predicts Google/Meta will commoditize consumer chat by making top models free, collapsing OpenAI’s subscription economics, while Chamath frames multi-model usage as normal consumer behavior.
- •Jason’s forecast: OpenAI share could fall below 50% in 12–24 months
- •Specialization across modalities (image, research, coding) drives multi-tool usage
- •Prediction: Google/Meta will subsidize AI to protect ad businesses and deny OpenAI oxygen
- •Chamath: big-tech cash makes subsidization rational; a third of a massive market is still enormous
- 17:56 – 19:20
Deal-making, capital momentum, and NVIDIA ‘options not deals’ skepticism
The discussion turns to whether OpenAI’s infrastructure partnerships are overstated and how deal-making can backfire with ecosystem partners. They debate incentives—raising momentum to raise huge capital—and speculate on NVIDIA reallocating investment to other labs.
- •Jason claims many OpenAI “deals” were options, not firm commitments
- •Potential partner backlash: AMD/NVIDIA competitive sensitivities
- •Chamath: deal flow can manufacture momentum needed for mega-fundraising
- •Speculation: NVIDIA may reduce/decline OpenAI participation and diversify exposure
- 19:20 – 21:24
Friedberg’s ‘not just LLMs’: multimodal complexity, video architectures, and the interface shift
Friedberg argues the real battleground won’t be a single text chatbot but a broader set of AI fronts—especially non-text modalities where differentiation is larger. He predicts the chat interface will look primitive in hindsight as agents and new applications dominate.
- •LLMs may commoditize; differentiation expands in multimodal/non-LLM systems
- •Video generation involves multiple interacting model types and architectures
- •Analogy: Omaha vs Hold ’Em—complexity widens skill gaps and winners
- •Future usage shifts from chat/search to agents: booking, media, real workflows
- 21:24 – 27:25
Google’s comeback: risk tolerance, leadership changes, and OpenAI’s ‘incumbent’ posture
They revisit how quickly narratives shift—Google went from “doomed” to highly competitive after organizational focus and a willingness to ship. Friedberg criticizes OpenAI for becoming overly defensive (safety/politeness/hedging), which he says degraded product usefulness compared to Gemini.
- •Reversal of sentiment: recent “eulogies for Google” look premature
- •Key factors: leadership focus (e.g., Demis oversight) and board permission to take risk
- •Friedberg: OpenAI’s defensive posture harmed voice UX and willingness to provide data/numbers
- •OpenAI as media target created pressure; Google benefited from being ignored
- 27:25 – 30:13
Sponsor/venue recap and pivot to NYT profile of David Sacks
Jason briefly recaps a Vegas/F1 weekend with partners and hospitality, then transitions to a New York Times article portraying Sacks as conflicted in his government role. The group frames the piece as a biased “hit job” that tried to imply self-dealing and boost the podcast’s business.
- •Event recap: Venetian hosting; partners Oracle/OKX/NYSE; F1 weekend
- •NYT headline claims: Sacks benefiting self/friends; emphasis on investment count
- •Explanation of SGE role constraints (limited days, no Senate confirmation)
- •Group sets up response: story seen as biased and under-evidenced
- 30:13 – 33:02
Sacks responds: loss of NYT credibility, alleged activism, and ‘coordinated response’ denial
Sacks says the backlash was organic and reflects diminished fear of NYT influence. He argues the paper functions as political activists laundering viewpoints through anonymous sources and claims the article failed to substantiate its own headline.
- •Support from across Silicon Valley interpreted as organic, not coordinated
- •Claim: NYT has lost mystique; targets no longer fear speaking up
- •Sacks accuses NYT of presenting activist narratives as objective truth
- •Reporting process described: repeated fact checks and shifting accusations
- 33:02 – 48:08
Ethics mechanics: disclosures, divestments, blind trust limits, and the ‘449 AI ties’ framing
Friedberg and Sacks walk through the ethics process and why a blind trust wasn’t feasible, emphasizing divestment instead. Sacks argues the NYT cited investments he had already disclosed publicly and downplays the “ties to AI” language as overly broad and misleading.
- •OGE ethics review and disclosures: positions were disclosed and assessed
- •Sacks: divested ~99% of potentially conflicting positions per ethics letter
- •Blind trust not workable due to minor children beneficiary rules
- •Sold LP interests/private holdings at steep discounts (~50%) to avoid conflicts
- 48:08 – 51:24
Example of alleged fabrication: the ‘Jensen Huang dinner’ that never happened
Sacks cites a specific fact-check excerpt claiming he had dinner with Jensen Huang where export-control arguments were discussed, which he says was fabricated. The group argues that removing one false detail doesn’t repair a narrative built on unreliable sourcing.
- •NYT fact-check process: sentence-by-sentence approval requests under time pressure
- •Sacks: no such dinner occurred; schedules checked for all parties
- •Argument that source credibility collapses if key scene is fabricated
- •Theme: insinuation via friendships and influence without hard proof
- 51:24 – 1:14:16
‘New poverty line’ debate: childcare shocks, benefit cliffs, and the politics of affordability/socialism
The group examines a viral claim that the poverty line should be far higher once childcare and modern expenses are considered. Chamath refines the claim with national vs high-cost-area comparisons and highlights a ‘stagnation zone’ where benefit cliffs weaken incentives; Friedberg broadens it into a critique of expanding taxation and a ‘spiral of socialism,’ prompting debate about solutions (housing, healthcare, education).
- •Viral claim: official poverty line ($31k for family of four) vs proposed ~$140k (high-cost area)
- •Chamath: use MIT Living Wage Calculator; example city shows ~$93k needs; childcare/housing are key drivers
- •Benefit cliffs: limited ‘Death Valley’ zone where marginal gains can be muted
- •Friedberg: taxes/spending cause a deficit spiral; cites business exits and wealth-tax examples; Jason counters with policy focus on housing/healthcare/education
