Library/Andrej Karpathy Joins Anthropic | SpaceX Files S1: How Does it Trade | Cerebras Smashes Day 1

20VC with Harry Stebbings

Andrej Karpathy Joins Anthropic | SpaceX Files S1: How Does it Trade | Cerebras Smashes Day 1

Key Themes
AI fundraisingtoken economicssoftware disruptioncompute scarcityIPO speculationAI infrastructurepublic market sentimentAI backlash
1h 27mMay 21, 2026
Summary

A wide-ranging AI and software market conversation spanning Anthropic, token economics, AI infrastructure, software disruption, and SpaceX’s rumored IPO

This episode moves from blockbuster AI fundraising and talent moves into the economics of tokens, the resilience and vulnerability of public software companies, and the current boom in AI infrastructure. The hosts repeatedly return to a central idea: compute scarcity and capital intensity are shaping the whole AI stack, while public-market excitement is increasingly concentrated in a small set of AI-linked stories. Later segments widen into the SpaceX IPO rumor mill, OpenAI token grants, and the political backlash now building around layoffs and AI messaging.

1
Frontier AI is becoming a capital-intensity story

The Anthropic discussion makes clear that the biggest AI players are competing not just on models and talent, but on the ability to fund enormous compute and infrastructure needs. That shifts the industry from a pure software narrative toward a balance-sheet and scale narrative.

2
Token costs are real, but the market may still be early

The Salesforce example shows that enterprise AI spend can already be large in absolute dollars, yet still modest relative to core engineering budgets. That suggests there is meaningful commercial activity today, but not necessarily the fully saturated demand curve implied by the most aggressive revenue projections.

3
Incumbent software is not uniformly breaking yet

The episode repeatedly contrasts AI-native excitement with the more mixed reality for public software companies. Figma appears to be adapting and even benefiting from AI workflows, while other names like Wix and Squarespace are portrayed as facing real structural pressure, showing how uneven the disruption is.

4
Compute scarcity is the key watch item for AI infrastructure

The hardware and infrastructure segment argues that these businesses remain powerful only so long as compute stays scarce. If supply catches up, pricing power could evaporate quickly, making capacity growth, permitting, and data-center buildout crucial variables to monitor.

5
Narrative and sentiment can dominate near-term pricing

The SpaceX and Cerebras segments both point to a market that is rewarding excitement, novelty, and AI adjacency. That can create powerful short-term price moves even when the available information is incomplete or the underlying fundamentals are still hard to assess.

6
AI messaging now carries political risk

The closing discussion highlights a growing backlash: after years of warning that AI could displace jobs, leaders now face criticism when layoffs are framed as efficiency gains. That makes communication strategy and credibility increasingly important for companies rolling out AI-driven restructuring.

Select any chapter text to Deep Dive with AI
01Anthropic's $30B Raise and Karpathy's Move Signal a Capital-Intensive AI Race

The opening segment centers on Anthropic’s reported $30 billion financing at a valuation above $900 billion and Andrej Karpathy’s move to the company. The panel frames the round as rational if viewed through ARR multiples, but emphasizes that the real driver is the enormous compute and capex burden of competing in frontier AI. Anthropic is contrasted with OpenAI as a more straightforward, less dramatic capital raiser that values speed and certainty.

Anthropic is reportedly raising $30 billion at above a $900 billion valuation.
Andrej Karpathy’s move to Anthropic is treated as a major signal.
The valuation is framed as more understandable on an ARR-multiple basis.
The real need is massive compute and capex funding.
Anthropic’s financing is contrasted with OpenAI’s more complex structure.
The panel suggests capital speed matters as much as price in the AI race.
02The True Cost of AI Tokens

This chapter examines how much companies are really spending on AI tokens, using Salesforce’s reported Anthropic spend as a benchmark. The speakers argue that token costs are already meaningful but still modest relative to engineering payroll, and that the market may be underestimating how much labor spend would need to shift to justify extreme AI revenue projections. They also discuss why some software companies may have more time than expected before AI meaningfully disrupts their growth.

Salesforce’s token spend is used as a benchmark for enterprise AI usage.
Current token costs are meaningful but still small relative to engineering payroll.
Large AI revenue projections would require a major share of labor opex.
The bull case depends on AI absorbing more salary spend than many expect.
The bear case is that usage stays lower and efficiency keeps tokens cheap.
Some incumbent software companies may have more time before disruption shows up in revenue.
03Figma Strength vs. Wix and Squarespace Under AI Pressure

The discussion contrasts Figma’s strong results and AI-enabled product evolution with the weakness of Wix and Squarespace. Figma is portrayed as benefiting from the broader software and AI buildout, even if it risks being bypassed by faster design-to-production tools. Wix and Squarespace are described as being squeezed by AI website generation and by Shopify’s dominance in commerce, with buybacks criticized as a poor use of capital when growth is deteriorating.

Figma showed accelerating net dollar retention.
Internal AI tools can improve design workflows and deepen product value.
Design-to-build tools like Lovable and Replit are a competitive threat.
Wix and Squarespace face AI website-builder pressure.
Shopify is cited as having reshaped the low-end commerce market.
Buybacks are criticized when businesses may need flexibility instead.
04Compute Starvation and the Cerebras Day-One Pop

The panel asks whether the hardware and semiconductor boom is sustainable or merely a bubble. The core variable is compute scarcity: if capacity stays tight, infrastructure names can keep performing well, but if supply catches up, pricing power could vanish quickly. Cerebras’ IPO is then framed as a strong signal of appetite for AI infrastructure, though the hosts caution that buying IPO day-one pops is usually a poor base-rate strategy.

Compute scarcity is the key determinant of infrastructure pricing power.
Permitting and data-center buildout may slow any oversupply.
The AI capex cycle is lifting semis, networking, and power-related names.
Cerebras’ IPO day-one surge signals strong risk appetite.
The market still rewards differentiated AI hardware stories.
Buying an IPO on the first-day pop is usually a bad base-rate trade.
05SpaceX IPO Date, Retail Mania, YC Tokens, and the OpenAI Lawsuit

This segment opens with rumors that SpaceX has set a June 12 IPO date, then questions how much the filing will reveal given recent deals and limited forward-looking detail. The hosts debate whether retail enthusiasm could produce a major first-day pop, before shifting to Sam Altman’s offer of OpenAI tokens to Y Combinator startups and what that means for startup financing and compute economics. The chapter ends with reactions to the OpenAI/Musk lawsuit being dismissed on technical grounds.

SpaceX is rumored to be preparing a landmark IPO.
The S1 may be historically useful but still incomplete.
Retail enthusiasm could drive first-day price action.
The float may be too large for a true meme-stock squeeze.
OpenAI token grants to YC startups could change financing economics.
The Musk/OpenAI lawsuit is discussed as a technical and procedural loss.
06AI Backlash, Layoffs, and the Politics of OpenAI

The closing segment broadens from the OpenAI dispute into a critique of the broader AI backlash. The speakers argue that years of public warnings about AI’s existential risks have collided with visible layoffs and cost-cutting, making the industry’s messaging politically toxic. They debate whether companies are being candid about AI-driven efficiency gains versus using AI as a pretext for headcount cuts, and end with a report that OpenAI may file as soon as Friday.

The OpenAI/Musk conflict remains a major overhang.
AI messaging has created political vulnerability.
Layoffs tied to AI are landing badly with the public.
Companies may be using AI language as a pretext for capex shifts.
The discussion calls for more honesty about tradeoffs.
OpenAI’s filing timeline may be accelerating.