Aug 22, 2026, 10:00 AM ETInterview, Fireside Chat

The AI Bubble WILL Burst | Should we be fearful of Chinese Open-Source | Jerry Murdock

Jerry Murdoch predicts an AI bubble burst occurring between October 26, 2024, and March 2027 if the conflict in Iran escalates, citing potential credit market disruptions.
The current credit market is characterized by "complacency," with spreads between real risk and low risk being too narrow, mirroring conditions prior to the 2008 financial crisis.
Japan's potential to unwind $300 billion in US treasuries to support the Yen could trigger an immediate global market problem.
Hyperscalers are uniquely positioned to survive a dislocation due to consistent ongoing business, whereas "neoclouds" face a high extinction risk.
At least 50% of neocloud providers are expected to cease operations within 36 months due to capital inefficiency and economic disruptions.

Fireworks vs. Base10: Fireworks is generating significantly more revenue than Base10 due to superior capital efficiency and a willingness to monetize profits rather than just revenue.
Token value is dynamic and increases with model customization; the more a model is tuned for specific tasks, the more valuable its tokens become.
Open-source models are expected to capture the majority of token volume (low-cost tasks), while frontier models will retain the majority of revenue (high-complexity tasks).
Frontier models (e.g., OpenAI, Anthropic) are predicted to continue innovating indefinitely due to capital resources, preventing open-source models from cannibalizing their core business in the short term.
Enterprise demand for AI is currently met at single-digit fulfillment levels, with massive potential for growth as more businesses adopt the technology.
Continuous learning models, expected within 2–3 years, will fundamentally replace current static model architectures.
Lifelong learning models, which mimic human continuous adaptation, will eventually supersede continuous learning models and replace all current open-source and frontier variants.
The "co-work" era, driven by autonomous agents, poses an existential threat to traditional SaaS companies that have merely added "AI copilot" features without systemic change.

Sandboxes: The most critical and underestimated component of AI security is the sandbox environment; companies like E2B and Docker are positioned to dominate this space.
Chip Strategy: Long-term, owning the chip layer (ASICs) is viewed as unnecessary for most companies, though short-term ownership aids optimization for hyperscalers.
Data Strategy: Data is not a commodity; unique enterprise data and its evolving context represent a $200 billion opportunity for specialized data providers like Macaw.
Investment Thesis: Investors should focus on founders with an "all-in" commitment and businesses where they feel a unique, non-negotiable necessity to build the company.
Exit Strategy: The "land grab" phase of low margins to acquire market share is a valid strategy (like Amazon's early days), but long-term survival requires building a culture of effective margin generation.
Routing Layer: Model routing intermediaries (e.g., OpenRouter) charging 5% markups are unlikely to survive; direct exchanges (e.g., Akinaki, Venice.io) will likely eliminate these fees within 3–5 months.

Frontier Model Ownership: While Sam Altman proposed US government equity stakes, Murdoch views this as a political necessity rather than a strategic one, given existing US capital market depth.
Export Controls: China's open-source models are unlikely to pose a lasting threat, as the entire generation of models will likely be obsolete within 10 years due to the shift to continuous learning architectures.
SaaS Valuation: Companies with $400–500M in revenue (e.g., Airtable) may face significant discounting if they lack a compelling AI strategy; the "lipstick on a pig" approach of adding AI features to legacy systems is insufficient.
IPO Market: IPOs remain viable for rare few with strong revenue trajectories (e.g., Cursor, potentially Anthropic/OpenAI), but many venture-backed firms will exit via M&A rather than public markets.
Private Equity Risk: PE firms with highly leveraged portfolios (4–6x leverage) face severe risk if a financial dislocation causes rapid EBIDA drops or churn increases.

Microsoft, Google, Meta: All three are considered long-term "marriage" investments due to their massive, stable consumer bases and control over global enterprise communication (e.g., Microsoft Exchange).
Meta Risk: While stable, Meta is the weakest of the three long-term holds, risking a transformation into a dividend-yielding, low-growth utility similar to AT&T.
Apple Strategy: Apple's AI approach is currently a "watchful observer" strategy; their fate depends on whether they are intelligent consumers of the ecosystem or merely passive adopters.
Nvidia Valuation: NVIDIA's stock is viewed as potentially mispriced due to market plateaus and circular transactions obscuring real growth, though a $10T valuation in five years is predicted.
Best PE Firms: Insight, Menlo Ventures, Benchmark, and Khosla Ventures are highlighted as having the best portfolios and navigation strategies for the AI transition.

Complacency: The industry suffers from security complacency, with many developers running models in "YOLO mode" inside unsafe containers rather than robust sandboxes.
Blockchain Utility: While Bitcoin is seen as a vehicle for greed and risk, blockchain will find true utility in agent-to-agent payments and inference exchanges within five years.
Innovation Cycle: The current period is a "Cambrian explosion" of innovation; investors must distinguish between fleeting hype and genuine structural shifts in the ecosystem.