Interview, Fireside Chat
The AI Bubble WILL Burst | Should we be fearful of Chinese Open-Source | Jerry Murdock
- A potential economic disruption or a prolonged Iran conflict could trigger a credit market collapse between October 26th and March 27th, causing at least half of neocloud companies to cease operations and forcing Private Equity firms with 4 to 6x leverage to face asset deflation.
- Demand for AI intelligence is projected to be endless like the 1990s web boom, though current hyperscaler free cash flow is at historic lows; however, hyperscalers are uniquely positioned to survive dislocations due to consistent business models while the open source model and ASIC chip sectors form a distinct "tsunami" for the future.
- Frontier model companies are expected to grow over the next decade by executing on continuous and lifelong learning, with continuous learning architectures anticipated to replace all current models within two to three years, potentially ten years, necessitating new training methods rather than incremental updates.
- Open source models will initially lag in revenue due to low costs but will eventually capture significant market share and dominate highly specialized tasks, while frontier models remain expensive and exclusive to wealthy entities; open source will not cannibalize frontier models in the short term due to the latter's innovation cycle.
- Security risks are expected to intensify with more frequent hacks of models like OpenAI and Hugging Face, as enterprises increasingly upload secret data and developers run models in unsafe "yolo mode" without sandboxes, though thousands of specific sandbox forms will be needed for probabilistic agents.
- The market will shift toward companies offering model customization via ASIC chips and away from expensive GPUs, with long-term chip ownership deemed unnecessary for model companies despite short-term optimization needs, while the number of chip designers is expected to increase.
- Large tech giants including Meta, Google, and Microsoft will likely remain trillion-dollar companies for at least a decade due to stable consumer bases and enterprise communication dominance, though Meta may transition into a lower-growth dividend generator.
- NVIDIA is forecasted to reach a $10 trillion market cap in five years, while Apple faces significant risks if it acts as a passive AI consumer rather than an innovative strategist; Venture firms with portfolios in companies like Anthropic and OpenAI are expected to outperform in the AI transition.
- Enterprises that fail to adopt a thoughtful AI strategy or establish a system of record will lose opportunity within two years as the "co-work era" of autonomous agents begins to threaten the SaaS sector.
- Data sales are projected to pivot from frontier models to serving enterprises and mid-market companies with specialized data, creating a potential $200 billion opportunity, while blockchain is expected to achieve utility in agent payments and tokenization within five years.
- A confluence of a global financial event and a failure in model growth could push the industry into a "valley of disillusionment," potentially accelerated by Japan selling $300 billion in US treasuries to support the yen or a burst AI bubble.
- Fireworks is predicted to succeed by leveraging expertise in Python and PyTorch to move up the stack into fine-tuning, while the market for sandbox compute will shift toward models where customers bring their own compute to optimize visibility, leaving current high-margin sandbox providers with flawed strategies.