Descriptions:
As markets absorb the surprise release of Kimi K3 — Moonshot AI’s 2.8 trillion parameter open-weight model — Bloomberg Technology speaks with a Crosslink Capital partner focused on mid-stage venture and AI infrastructure investments. Drawing explicit parallels to the DeepSeek moment of the prior year, the guest argues that China’s AI frontier capabilities are now materially closer to U.S. leaders than consensus assumes, with open-weight models increasingly viable alternatives for enterprise workloads.
Rather than viewing cheaper, more capable models as a threat to AI infrastructure spending, the investor makes the Jevons paradox argument: inference costs falling 95% over recent years have historically expanded total AI expenditure, as previously uneconomic workloads become accessible. Kimi K3’s pricing — $3 per million input tokens, $15 per million output tokens — is cited as roughly in line with established frontier models, complicating the narrative that Chinese models are purely a low-cost disruptor.
The conversation sharpens around valuation implications: if a $20 billion Chinese startup can train a 2.8 trillion parameter competitive model, what justifies trillion-dollar valuations for OpenAI and Anthropic? The guest declines a definitive answer but argues both companies continue producing differentiated products. He identifies agent infrastructure — particularly identity layers, agent harnesses, and memory systems — as his highest-conviction investment area, citing portfolio company Teleport as a category leader. The interview offers a grounded framework for thinking about where AI value accrues as model costs commoditize.
📺 Source: Bloomberg Tech · Published July 20, 2026
🏷️ Format: Interview
