Summary
Mike Volpi, who helped oversee Cisco’s mergers and acquisitions during the late-1990s internet buildout, joins an a16z conversation to compare that era with today’s AI infrastructure boom. His starting point is that demand for AI capacity, measured in tokens, is near infinite while supply is tightly constrained, a world he jokingly calls Jensen’s world.
Volpi draws both parallels and differences. Both periods featured huge expectations, surging demand and a flood of equity and debt capital. But he notes that fiber bandwidth grew dramatically through technologies like WDM, while compute improvement is more moderate, bound by something resembling Moore’s law. He also argues AI’s practical applicability is higher today, pointing to the revenue ramps of Anthropic, OpenAI and open-model hosts such as Fireworks, Together and Modal. He believes the current buildout will be fully utilized, while questioning whether today’s owners will be the long-term owners.
The discussion also covers M&A lessons from Cisco: repeatable integration processes, small dedicated deal teams, tolerance for failure at high frequency to capture outliers, and the trouble with large acquisitions such as Scientific Atlanta. It is useful listening for investors, founders and operators weighing how durable the AI infrastructure cycle will be.
📺 Source: a16z · Published November 11, 2026
🏷️ Format: Podcast
