Overview of 20VC with Jerry Murdock
This episode of 20VC features Harry Stebbings in conversation with Jerry Murdock, founder of Insight Partners, for a wide-ranging, opinionated discussion on the current AI boom, whether a bubble is forming, which companies will survive a market dislocation, the rise of open source and specialized models, China’s role in the AI race, and how the Mag 7 may fare over the next decade. Murdock argues that the AI buildout is real but highly vulnerable to credit shocks, that many neoclouds and app-layer companies will not survive, and that the most durable value in AI will accrue to hyperscalers, infrastructure, security, and highly specialized tooling.
Core Themes and Main Takeaways
1) The AI boom is real, but the market is fragile
- Murdock believes the current AI cycle is overheated and vulnerable to a correction.
- His core concern is not AI demand, but the financing structure behind the AI buildout:
- hyperscalers and infrastructure players are taking on large amounts of debt,
- private credit spreads look complacent,
- a macro or geopolitical shock could trigger a credit market disruption.
- He frames the biggest risk as a financial dislocation rather than a lack of product-market fit for AI.
2) Hyperscalers are best positioned to survive a crash
- If the market does crack, Murdock thinks hyperscalers are the most likely survivors:
- they have real cash flow,
- diversified businesses,
- and the balance sheet strength to outlast weaker players.
- In his view, a downturn would likely wipe out smaller competitors, while making assets cheaper for the largest platforms to acquire.
3) Neoclouds are at serious risk
- Murdock is blunt: at least half of today’s neoclouds may disappear within 36 months.
- He argues that survival will depend less on hype and more on:
- capital efficiency,
- profitability,
- management quality,
- and the ability to avoid excessive leverage.
- He repeatedly contrasts stronger operators with weaker, more expensive growth stories.
4) Open source and specialized models will matter a lot
- Murdock strongly supports the idea that AI will increasingly become specialized rather than one-size-fits-all.
- He believes open source models can win large parts of the market because:
- they’re cheaper,
- they’re easier to customize,
- and many enterprise use cases don’t require frontier-model-level capability.
- He sees the long-term market as a combination of:
- a few frontier model companies,
- many specialized models,
- and infrastructure layers that help customize, route, and deploy them.
5) “A token is not a token”
- He pushes back on the idea that all model tokens are economically identical.
- In his view, token value depends on:
- model customization,
- verbosity,
- task complexity,
- and whether the output is actually useful for the business.
- This is why he thinks companies like Fireworks can be more attractive than higher-valuation rivals like Baseten if they are more capital efficient and more profitable.
6) Security will be one of the biggest AI winners
- Murdock is highly bullish on AI security and sandboxing.
- He thinks many developers are underestimating the risks of deploying models and agents in unsafe environments.
- His key view:
- containers are not enough,
- sandboxes matter,
- and agent behavior needs to be controlled and observed carefully.
- He suggests the AI security market is still early and likely much larger than most people think.
7) Enterprise AI is constrained by trust and firewall issues
- Enterprise customers are wary of giving all of their data to frontier model providers.
- Murdock agrees with the concern that companies should be selective about what data goes into external AI systems.
- He sees this caution as part of the reason open source and behind-the-firewall systems will keep growing.
8) The next major wave: continuous learning and lifelong learning
- Murdock believes current models are static, but future models will be fundamentally different.
- He expects a transition toward:
- continuous learning models,
- then eventually lifelong learning models.
- These models, he argues, may replace today’s model generation entirely.
- He treats this as a major architectural shift, not just an incremental improvement.
9) China, chip exports, and open source risk
- On China and export controls, Murdock does not endorse regulation for its own sake.
- He wants a clear national strategy rather than just blanket rules.
- He also argues that many current Chinese open source models will not be the same dominant players in 10 years, because the model landscape will evolve rapidly.
10) The Mag 7 are still the safest large-cap bets
- If forced to choose, Murdock leans toward holding most of the Mag 7 for the long term.
- His reasoning:
- enormous consumer/user bases,
- durable products,
- and strong buffers against short-term AI disruption.
- He says Meta is the weakest of the bunch from a long-term moats perspective, but still not easy to short because of its scale and user base.
- Microsoft stands out as especially resilient because of enterprise communications and infrastructure.
Company and Sector Views
Likely winners
- Hyperscalers
- AI infrastructure
- Security / sandboxing providers
- Specialized model companies
- Open source enablers
- AI data providers
- Blockchain payment / inference rail infrastructure
Likely losers or weaker areas
- Overlevered neoclouds
- Low-margin AI app layers without real differentiation
- Generic SaaS companies that simply “sprinkle AI” on top
- Open router / routing layers with high take rates, if competitors emerge
- PE-backed businesses that are too leveraged in a market downturn
Investing Lessons from Jerry Murdock
What he says to look for in founders
- Extreme commitment
- Unusual conviction
- The sense that the founder has to build the company
- Real impact, not just a compelling pitch
- Specialized focus before expansion
What he says to avoid
- Excessive leverage
- Low-margin businesses that treat margin as optional forever
- Commodity AI plays with no moat
- Teams that are not thought through on security
- Companies assuming hype will last without building durable economics
His venture philosophy in AI
- Infrastructure is where the strongest opportunity is today.
- App-layer investments are more suspect unless they are truly differentiated.
- The best investors will focus on:
- model adjacency,
- customization,
- security,
- data,
- and workflow complexity.
Notable Quotes and Ideas
- “If there is a dislocation, no one is better prepared to survive it than hyperscalers.”
- “I think at least half of [the neoclouds] go away within 36 months.”
- “A token is not a token.”
- “The most important thing underestimated is the need for sandboxes.”
- “The market initially expands, then contracts, then starts again.”
- “Blockchain for agent payments” is one of the things he thinks will seem obvious in hindsight.
Final Takeaway
Jerry Murdock’s view is that AI is transformational, but not evenly so. The next few years will likely produce:
- major winners in infrastructure, security, and customization,
- major casualties among overfunded, overlevered, or undifferentiated companies,
- and a possible market reset if macro or geopolitical stress hits credit markets.
His core message: don’t confuse excitement with durability. The AI future is real—but only some companies will survive the cycle.