All-In PodcastWhy Cursor chose xAI's compute over Anthropic and OpenAI
Competing foundation model vendors threatened Cursor's independence; SpaceX structured a $60B acquisition or $10B breakup fee to lock in xAI compute alignment.
CHAPTERS
- 0:00 – 2:30
Bestie cold open: SPLC jokes, DC trip, and Sacks’ White House meeting
The episode opens with banter about Jason being “SPLC-adjacent,” followed by David Sacks explaining why the show started late: he was in Washington, DC and met with the President. Sacks contrasts media portrayals with his own experience and frames the moment as pivotal for US competitiveness in AI.
- •Running jokes about “SPLC files” and Jason’s notoriety
- •Sacks recounts a White House meeting and praises the President’s demeanor
- •Media portrayal vs. personal experience and why AI policy timing matters
- •Discussion of AI regulation: targeted rules vs. blanket fear-based slowdown
- 2:30 – 4:51
AI policy and the White House: data centers, Anthropic, and ‘winning’ the AI race
The group zooms in on AI strategy—especially energy and data centers—as national leverage. They discuss the President’s comments about Anthropic, the tension between politics and procurement, and an approach focused on enabling buildout rather than restricting it.
- •AI infrastructure as the bottleneck: power generation and data centers
- •Sacks on Biden-era DEI/censorship concerns vs. current posture
- •Anthropic relationship: product quality vs. perceived political alignment
- •National competitiveness frame: enable growth, avoid ‘halt progress’ approaches
- 4:51 – 10:21
SpaceX–Cursor mega-deal: structure, economics, and compute as leverage
Jason introduces the reported SpaceX–Cursor deal: either a $60B acquisition by 2026 or a $10B collaboration payment. The besties unpack Cursor’s growth, SpaceX’s IPO narrative, and why access to Colossus-scale GPU compute could catapult Cursor/xAI in coding.
- •Deal terms: $60B buyout option vs. $10B ‘breakup fee’ collaboration payment
- •Cursor metrics: strong run-rate growth expectations and category leadership in AI coding
- •SpaceX IPO storytelling: revenue multiples and accretive software/services angle
- •Compute scarcity as leverage: xAI/Colossus GPUs as the strategic unlock
- 10:21 – 14:53
Why Cursor fits xAI: vertical integration pressure, IDE UX edge, and enterprise token costs
Sacks and Friedberg explain why Cursor needed a new foundation-model alliance as OpenAI and Anthropic vertically integrate into coding tools. Friedberg argues Cursor’s IDE layer and developer UX are key, while Sacks highlights the coming enterprise push to optimize token spending and route tasks across models.
- •Cursor’s risk: dependence on foundation model providers that now compete (Claude Code/Codex)
- •Cursor’s advantage: IDE/application layer UX that developers prefer
- •Enterprise reality: token bills rising fast; efficiency and model-routing middleware becomes vital
- •Agents create software sprawl; IDE-centric workflows help rein in and engineer systems
- 14:53 – 18:28
Cyber as the next ‘white-hot’ frontier: Mythos, dedicated security models, and arms-race dynamics
Sacks flags cybersecurity as the most intense subdomain of AI coding. He argues frontier-scale cyber models (e.g., “Mythos”) are expensive and compute-constrained, pushing the market toward purpose-built cyber models that can be deployed commercially for defense (and potentially offense).
- •Cybersecurity becomes central as frontier models enable new attack/defense capabilities
- •Very large models create serving constraints; commercialization requires cheaper specialized models
- •CISOs and IT departments expected to accelerate adoption due to AI-driven threat concerns
- •Near-term race: dedicated cyber models optimized for cost/performance
- 18:28 – 27:21
SaaS debt bomb: Medallia/Thoma Bravo, collapsing predictability, and agent-driven deflation
The conversation shifts to Thoma Bravo potentially handing Medallia to creditors after rising debt service costs. Friedberg argues agent-built internal tools are eroding new sales and retention, breaking the predictability PE underwriting depends on; he connects it to broader AI-driven deflation and Fed policy debates.
- •Medallia case: debt service costs spike; equity impairment and creditor pain
- •AI agents as SaaS substitutes: easier/cheaper to spin up internal workflows than buy vertical SaaS
- •Key metric risk: net revenue retention becomes unstable, undermining leveraged buyout models
- •Deflation lens: AI lowers operating costs, forcing revenue compression in legacy software segments
- 27:21 – 35:37
Is this a generational SaaS bargain—or a falling knife? Valuation collapse and founder ‘maneuverability’
Sacks weighs PE opportunity versus structural risk: public SaaS multiples look historically cheap, but cash flows may no longer be durable. The group debates whether leaders like Salesforce can adapt (headless/agent integrations) and whether founder-led companies will navigate the transition better than manager-run incumbents.
- •Valuations reset: quality SaaS trading at fractions of prior ARR multiples
- •Bear case: unpredictability from churn and seat compression breaks debt-financing assumptions
- •Salesforce as a potential winner: platform scale, data gravity, and headless/agent strategy
- •Founder-led resilience: founders ‘burn the boats’ and make harder pivots faster
- 35:37 – 46:18
Debt makes companies brittle: venture debt warnings and the ‘optional flexibility’ premium
They broaden from PE leverage to venture debt, calling it a common fragility trap for founders. The besties argue debt reduces strategic flexibility, adds covenants, and can lead to punitive renegotiation—while free cash flow and equity financing provide maneuverability in turbulent markets.
- •Debt vs. equity: debt imposes fixed payments and covenants that limit pivots
- •Venture debt critique: founders treat it like VC capital and underestimate downside
- •Banks can’t tolerate ‘zeros’ like VCs; they will extract harsh terms when scared
- •Free cash flow as strategic optionality during disruption
- 46:18 – 49:20
Apple succession: John Ternus takes over—what Apple must do next in an AI-first world
Jason tees up Tim Cook’s transition and the reported successor John Ternus, framing it as a shift back toward product leadership. Friedberg and Sacks argue Apple’s core challenge is building an AI-native assistant layer across devices, while balancing privacy, model choice, and platform openness.
- •Leadership change: Ternus as long-tenured hardware/product executive
- •AI imperative: Siri replacement/upgrade and a ubiquitous personal assistant layer
- •Platform strategy: potential ‘choose your model provider’ approach vs. single-vendor lock-in
- •Wearables and new device categories as the next battleground for Apple’s ecosystem
- 49:20 – 1:00:32
Tim Cook’s legacy: stewardship, services mix, buybacks, and the ‘Disney succession’ analogy
Sacks and Chamath defend Cook’s tenure as exceptionally effective stewardship—massive market cap growth, services expansion, and disciplined capital return via buybacks. Sacks compares Apple’s moment to Disney after Walt: the next CEO must avoid a stagnant “post-founder funk” and find a revitalizing path.
- •Cook’s performance: revenue scale-up, services mix, and strong brand/low-scandal era
- •Capital allocation: unprecedented buybacks and shareholder-friendly cash returns
- •Innovation debate: many product releases, but fewer iPhone-scale breakthroughs
- •Succession framing: Apple now needs its ‘Eisner/Iger’ revitalization moment
- 1:00:32 – 1:06:58
SPLC indictment: alleged wire fraud, informants, and ‘arsonist firefighters’ incentives
The episode pivots to the Southern Poverty Law Center’s indictment on wire fraud and money laundering allegations tied to hidden accounts and paid informants. The besties argue incentives may push NGOs to amplify or even manufacture the problem they fundraise against, while noting the legal case remains allegations until adjudicated.
- •Indictment summary: alleged hidden accounts and payments linked to infiltrating hate groups
- •Charlottesville-era fundraising spike and questions about incentives and transparency
- •Debate on ‘informants’ vs. ‘fomenting’: concealment from donors as a red flag
- •Wider critique: cancellation culture and expansive ‘hate-adjacent’ labeling practices
- 1:06:58 – 1:19:03
NGOs and 501(c)(3) drift: why nonprofit incentives break feedback loops
Chamath, Sacks, and Friedberg generalize the SPLC story into a critique of nonprofit structures. They argue NGOs can prioritize fundraising and job preservation over mission outcomes, and call for tighter definitions, audits, and removal of tax-advantaged status for politicized or non-exempt activities.
- •Nonprofit incentive problem: fundraising replaces mission success as the key KPI
- •IRS ‘exempt purpose’ definition vs. modern nonprofit behavior and scope creep
- •Calls for transparency: auditing, donor accountability, and reducing tax-privileged abuse
- •‘No one declares victory’: missions expand rather than conclude, moving goalposts over time
- 1:19:03 – 1:30:41
Science Corner: colon cancer spike in young people linked to epigenomic signal from picloram pesticide
Friedberg spotlights a new paper analyzing tumor epigenomics in under-50 vs. over-70 colorectal cancer patients and finding a strong association signal with the herbicide picloram. He explains persistence in the environment, the lack of modern EPA review, and why RNA/gene-expression tools can reveal long-term carcinogenic risk pathways.
- •Trend: ~80% rise in young-onset colorectal cancer over two decades
- •Method: epigenomic/gene expression comparison to infer environmental triggers
- •Key signal: picloram (Dow-developed herbicide) associated with cancer signature and geography
- •Policy implication: modernize chemical safety review using epigenomics, not just acute toxicity
