Overview of 20VC: OpenAI & SpaceX S1 Drops | NVIDIA's $81BN Revenue Quarter | Cloudflare and ClickUp Do Controversial Layoffs | Exa, OpenRouter and Polsia Raise Mega Rounds | Uber and Microsoft Declare AI ROI for Developers is Questionable
Harry Stebbings, Rory O’Driscoll, and Jason Lemkin dissect a packed week in tech: NVIDIA’s blockbuster quarter, the escalating debate over whether AI is actually delivering ROI, public-market timing for OpenAI and Anthropic, Elon Musk’s sprawling SpaceX/xAI narrative, and a wave of layoffs that may say more about AI efficiency than “COVID overhiring.” The episode’s core question is simple: if AI is so transformative, where is the proof in revenue, margins, productivity, and capital allocation?
The Biggest Macro Theme: AI Spend Is No Longer Just Hype
The conversation repeatedly returns to one question: what is the real ROI of AI?
- Corporate America appears more convinced than ever that AI spend will pay off.
- But the guests argue the next phase will be more demanding:
- It’s not enough to say AI is useful.
- Companies will increasingly need to show measurable gains against token spend, staffing changes, and margin impact.
- The key tension:
- Some companies are clearly getting leverage from AI.
- Others may be entering a more skeptical phase where “vibe-based” adoption gives way to hard budgeting.
Their shared view: the market is moving from experimentation to accountability.
NVIDIA: A Huge Quarter, But Expectations Are Already Wild
NVIDIA’s quarter was the day’s anchor topic.
The headline numbers
- $81.6B revenue
- ~$56B in profit
- $91B Q2 guidance
- $80B buyback
Why the stock barely moved
The panel argues the reaction makes sense because:
- NVIDIA is already priced for extraordinary AI capex growth.
- The market now expects NVIDIA to beat, raise, and accelerate every quarter.
- When a company is this embedded in portfolio and index exposure, “not falling” after earnings can be interpreted as a win.
Bigger takeaway
The guests emphasize that NVIDIA is less a “great quarter” story than a capital expenditure story:
- AI infrastructure spend could reach $3T–$4T by 2030 if Jensen Huang is right.
- The real question is not whether demand exists today, but whether the economics of the next $2T in spend justify the return.
OpenAI vs. Anthropic: The IPO Race Is On
A major segment focused on OpenAI’s confidential S-1 and how that changes the competitive landscape.
Why OpenAI should go public
The panel thinks OpenAI is now under pressure to list because:
- Anthropic is growing faster.
- Anthropic appears to be becoming profitable sooner.
- OpenAI risks looking like the “number two” if it waits too long.
The revenue race
They point out that:
- Anthropic’s Q1 run rate is now effectively catching up to or surpassing OpenAI on a trajectory basis.
- Anthropic’s revenue in Q1 was roughly on par with its full prior-year revenue pace, while OpenAI’s growth looked less explosive by comparison.
Anthropic’s advantage
They highlight:
- Rising gross margins
- Strong enterprise traction
- Premium pricing power
- Improving operating profile
Their broad view: Anthropic may already be the more attractive financial asset, even if OpenAI still owns the consumer mindshare.
SpaceX S1: Massive Ambition, Massive Skepticism
The SpaceX filing sparked equal parts admiration and disbelief.
What the panel likes
- Elon Musk has earned the right to attempt huge, world-changing bets.
- SpaceX, Starlink, and xAI-related infrastructure are all real businesses with real momentum.
- The filing showcases genuine operational excellence and extraordinary execution.
What the panel doesn’t buy
They are very skeptical of the valuation logic:
- The company appears to be bundling together disparate assets.
- Much of the story depends on a future in which:
- data centers move into space,
- AI infrastructure becomes massively larger,
- and SpaceX becomes central to that buildout.
Their conclusion
- Starlink is likely the biggest long-term value driver
- The “data centers in space” angle is viewed as optionality, not the base case
- The current valuation feels like it relies heavily on an Elon premium
In short: brilliant strategy theater, but hard to justify on DCF.
Layoffs: AI Efficiency vs. “COVID Overhiring”
The episode spent significant time on layoffs at companies like Cloudflare and ClickUp.
Their take
The panel strongly rejects the idea that these cuts are simply “COVID overhiring.”
Instead, they argue:
- Natural attrition since 2020 is already large enough that this explanation doesn’t add up.
- AI is likely playing a much bigger role.
- Companies are realizing they can do more with fewer people and are reallocating budget toward top performers.
On the PR messaging
They are skeptical of long public layoff explanations:
- It often sounds like blaming employees.
- It can come across as defensive or performative.
- The more honest message would be: the world changed, and the business model changed with it.
A new compensation model
A key insight:
- If AI makes top engineers and operators much more productive, companies may:
- cut headcount,
- and pay their best people far more.
- The result could be a world where revenue per employee keeps rising sharply.
AI Infrastructure Beneath the Foundation Models: The “Picks and Shovels” Trade
The episode closes with enthusiasm for the layer below the frontier models.
Companies discussed
- Exa: search for agents
- OpenRouter: model routing / model switching
- Manus: mentioned in the context of a complicated buyback
- Polsia/Pulsia: a polarizing AI startup with aggressive marketing and strong fundraising momentum
Why these matter
The guests argue that agents need a new stack:
- search
- routing
- observability
- structured access to current information
- databases and workflow tools
Exa’s thesis
Exa is interesting because:
- it is useful specifically for agents, not just humans
- it solves the “current information” problem for autonomous workflows
- it may not be winner-take-all, but it could become a durable developer tool with real demand
OpenRouter’s thesis
OpenRouter matters because:
- not every task needs the most expensive model
- enterprise users want the ability to optimize cost and quality
- model routing could become a standard part of the AI stack
Their broader conclusion: the picks-and-shovels layer of AI is still underappreciated.
Final Takeaways
- AI is entering a more quantitative phase. ROI will need to be proven, not assumed.
- NVIDIA’s business is still exceptional, but the market already expects perfection.
- Anthropic looks increasingly dangerous to OpenAI on revenue and margin trajectory.
- OpenAI should probably go public soon to avoid being framed as the lagging leader.
- SpaceX’s S1 is fascinating but highly valuation-dependent, with a huge Elon premium.
- Layoffs are increasingly an AI productivity story, not just a post-COVID correction.
- Agent infrastructure is a real investment theme, especially in search and model routing.
If you want the single sentence summary: this episode argues that AI is no longer about whether it works, but about who can prove the economics first.