"We are seeing way too many business models (and valuation levels) with no realistic margin expansion story, extreme capex spend, lack of enterprise customer traction, or overdependence on "roundtrip" investments - in some cases all with the same company," Davis wrote. Yet, Davis argues the same exuberance driving landmark transactions at the chip level is also fueling excess elsewhere in the AI stack, particularly among third-party data-center developers betting on what he called the "build it and they will come" model."
"The warning comes just days after Nvidia agreed to license assets from Groq, a high-performance AI chipmaking startup Disruptive has backed since its founding (not to be confused with Grok, Elon Musk's AI chatbot). The transaction, which Davis has said is valued at roughly $20 billion in cash, represents the largest deal Nvidia has ever completed and underscores how aggressively the company is moving to lock up all the verticals in AI talent and intellectual property."
An Austin-based investment firm expects a significant financing crisis to hit the speculative data-center market by 2027 or 2028, driven by extreme capital expenditures and a growing mismatch between those building AI infrastructure and those who will use it. Many business models show no realistic path to margin expansion, carry extreme capex burdens, lack enterprise customer traction, and rely on roundtrip investments. Recent large-scale chip-level transactions, including Nvidia's licensing deal with Groq valued at roughly $20 billion in cash, demonstrate aggressive vertical consolidation in AI talent and intellectual property. The firm prefers backing owner-users such as hyperscalers over speculative landlords.
Read at Fortune
Unable to calculate read time
Collection
[
|
...
]