MurmurCast
OpinionDiscussion

20VC: The AI Bubble Will Burst: Half the Neoclouds Will Die | China: Should We Ban Chip Exports & Be Fearful of Chinese Open-Source | Mag7: Who Dies and Who Thrives: Why Meta is Meh and Microsoft is Mega

The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch1h 6m

Jerry Murdoch, founder of Insight Partners managing $90B, discusses the AI bubble, predicting at least half of neoclouds will collapse within 36 months if a financial dislocation occurs triggered by global conflicts. He emphasizes hyperscalers are best positioned to survive downturns, advocates for open-source and specialized models over frontier models for most use cases, and warns that continuous learning models will eventually replace all current model architectures.

Summary

Jerry Murdoch provides a seasoned investor's perspective on the current AI landscape and broader market conditions. He argues that if geopolitical tensions (specifically Iran conflict) escalate and cause credit market disruptions, the AI bubble could burst between October 2026 and March 2027. His concern centers on the unprecedented debt levels hyperscalers have taken on—if credit markets seize up, asset values could decline sharply despite underlying business quality, triggering margin calls.

On neoclouds specifically, Murdoch predicts at least 50% will disappear within 36 months, with the primary differentiator being management quality and capital efficiency rather than the business model itself. He contrasts Fireworks (highly capital efficient, profitable) with Base 10 (raising at similar valuations but unprofitable), arguing investors should bet on profitability-focused founders. He believes hyperscalers will actually benefit from a dislocation since competitors get wiped out and assets become cheaper.

Regarding the frontier versus open-source model split, Murdoch envisions a future dominated by specialized, customized models. He argues tokens aren't fungible—customized models produce different token volumes and quality, making cost comparisons misleading. Open-source will capture volume at low cost while frontier models handle complex tasks at premium prices. However, he emphasizes this hierarchy assumes continuous innovation by frontier model companies; without it, a "valley of disillusionment" becomes likely.

Murdoch predicts that continuous learning and lifelong learning models will fundamentally replace all existing models within a decade. These architecturally different systems, unlike today's static-training models, will require new frameworks and cannot simply be bolted onto current frontier models. Chinese open-source models with potential backdoors won't matter long-term since they'll be obsolete in 10 years anyway.

On infrastructure opportunities, he favors companies solving complex integration problems (like E2B and Docker for sandboxing) over commoditized services. Model routing layers like OpenRouter (charging 5% markups) will be disrupted by decentralized exchanges (Akinaki, Venice IO) offering direct inference without intermediaries. He emphasizes sandboxing as critically underestimated for security, especially as probabilistic agents explore multiple solution paths simultaneously.

Murdoch advocates for founders driven by necessity rather than opportunity—those who "have to" build their business regardless of capital conditions. He uses founder examples like Vignan at E2B and Sadi Khan at an unnamed company (praised by Vinod Khosla) as embodying this commitment. He warns against low-margin strategies claiming they're temporary; instead, founders should build margin-focused cultures from day one or risk losing venture backing.

On the broader market, he expresses concern about complacency in credit markets—comparing today to 2008 when credit rating agencies and risk departments missed obvious problems. Multiple systemic risks exist: Japan's potential treasury unwind (they've already sold $300B to support the yen), private debt market spreads being too narrow, and the complexity of global AI spending creating multiple disruption opportunities.

Regarding the Magnificent Seven, Murdoch sees all three major players (Meta, Google, Microsoft) as multi-decade holds due to their massive consumer bases and communication control—they act as stabilizers giving them time to catch up on AI products even if they've underperformed on coding agents. Meta is positioned as the least exciting but most dividend-generating long-term hold. He's skeptical of most SaaS companies lacking genuine AI strategies (vs. superficial co-pilot additions), warning that the "co-work era" of autonomous agents poses existential threats to traditional SaaS.

Murdoch sees blockchain as finally finding genuine utility in agent payments and inference exchanges, currently in the "valley of disillusionment" but positioned for vindication within 5 years. He distinguishes between speculative greed-driven projects and utility-focused innovations like tokenized stocks and payment rails.

About this episode

<p>Jerry Murdock is the Co-Founder of Insight Partners, which manages over $90 billion in assets. Jerry personally backed companies including Twitter, Nest and Docker, while Insight's portfolio includes giants such as Shopify, Wiz and monday.com. Across three decades, Insight has helped produce 55+ IPOs and become one of the most powerful technology investment firms in the world. </p> <p>AGENDA:</p> <p class="isSelectedEnd">00:00 — Is the AI Bubble About to Burst?</p> <p class="isSelectedEnd">11:00 — Will Half of All Neoclouds Disappear Within 36 Months?</p> <p class="isSelectedEnd">13:00 — Can Open Source Models Actually Beat OpenAI and Anthropic?</p> <p class="isSelectedEnd">22:00 — Are We Entering the Golden Age of Cybersecurity Attacks?</p> <p class="isSelectedEnd">32:00 — Have AI Valuations Completely Lost Touch With Reality?</p> <p class="isSelectedEnd">42:00 — Is AI About to Wipe Out an Entire Generation of SaaS & Private Equity?</p> <p class="isSelectedEnd">47:00 — Should We Ban U.S. Chips From Being Exported to China?</p> <p class="isSelectedEnd">48:00 — Should We Be Fearful of Open-Source Chinese Models?</p> <p class="isSelectedEnd">49:00 — Will Continuous Learning Kill Every AI Model We Use Today?</p> <p class="isSelectedEnd">53:00 — Why Microsoft Remains a Mega Buy</p> <p>56:00 — Which Mag 7 Giant Would You Short — and Is Apple's AI Strategy a Disaster?</p> <p> </p>

Key Insights

  • Murdoch argues that if global credit markets experience dislocation triggered by geopolitical conflict (specifically Iran war escalation), the AI bubble will burst between October 2026 and March 2027 because hyperscalers have taken on unprecedented debt levels that become dangerous when asset values decline sharply.
  • He claims at least 50% of neoclouds will disappear within 36 months, with management quality and capital efficiency being the primary differentiators—not business model—because underprofitable companies burning cash are at extreme risk during any market disruption.
  • Murdoch states that tokens are not fungible; customized models produce different token volumes and quality because model specialization directly affects output verbosity and utility, making cost-per-token comparisons between frontier and open-source models misleading.
  • He predicts continuous learning models will fundamentally replace all existing models within a decade because they require architecturally different designs and new training paradigms that cannot be bolted onto current frontier models.
  • Murdoch argues that hyperscalers are paradoxically best positioned to benefit from a financial dislocation because their competitors get wiped out, assets become cheaper to acquire, and their ongoing business remains consistent enough to absorb the shock.
  • He claims model routing intermediaries like OpenRouter charging 5% markups will be disrupted within 3-5 months by decentralized exchanges (Akinaki, Venice IO) that eliminate the intermediary fee by connecting buyers and sellers directly.
  • Murdoch contends that sandboxing is the most critically underestimated security need because probabilistic agents explore multiple solution paths simultaneously (potentially 100 sandboxes with different libraries), requiring sophisticated knowledge of how models interact with tools.
  • He argues that SaaS companies without genuine AI strategies embedded in their core product—merely adding co-pilots as superficial features—face existential threats from the emerging autonomous agent era, with limited time to pivot before becoming obsolete.
  • Murdoch states that complacency in credit markets today mirrors 2008 because credit rating agencies and bank risk departments are failing to account for leverage and complex interdependencies in private debt, with narrow spreads between risky and safe assets.
  • He claims Japan's potential $300B treasury unwinding (already occurring to support the yen) could trigger immediate global problems if they need to sell more treasuries, as markets wouldn't absorb that volume without severe dislocation.
  • Murdoch argues that Microsoft and Meta will remain multi-decade holds despite underperformance on specific AI products because they control critical communication infrastructure (Exchange email, WhatsApp, Instagram) with 2+ billion users that cannot be easily replaced.
  • He predicts blockchain will transition from the valley of disillusionment to genuine utility within 5 years through applications in agent payments and decentralized inference exchanges, not through speculative cryptocurrency greed.

Topics

AI bubble and financial dislocation risksNeocloud survival and capital efficiencyFrontier models vs. open-source economicsSpecialized and customized AI modelsContinuous learning and lifelong learning modelsHyperscaler competitive advantagesCredit market complacency and systemic risksInfrastructure layer opportunities and pricingSaaS companies adapting to AI eraFounder quality and commitment as differentiatorBlockchain utility for agent paymentsSecurity and sandboxing importanceModel routing and decentralized inference exchangesMag 7 companies as decade-long holds

Transcript

If there is a dislocation, no one is better prepared to survive it than hyperscalers. Let's take neoclouds. Neoclouds right now, there's a whole bunch of them. I think at least half of them go away within 36 months. Fireworks is making a lot more money than Base 10. The more you customize the model, the more the token changes its value. This is 20VC with me, Harry Stebbings. Now, everyone is calling today peak froth. Prices are out of control. There's too much money. We don't know what's going to happen with this AI bubble. Well, you know what's really valuable in this time? Wisdom. Jerry Murdoch, joining me in the hot seat today, he's the founder of Insight.…

Full transcript available for MurmurCast members

Sign Up to Access
View original source →

More from The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch

Get AI summaries like this delivered to your inbox daily

Get AI summaries delivered to your inbox

MurmurCast summarizes your YouTube channels, podcasts, and newsletters into one daily email digest.