Cerebras priced its IPO at $185 on May 13, opened May 14 up roughly 68%, and gave back about 10% on Friday, May 16. [1] As of Wednesday's close, five trading days have passed without a clean second print. Nvidia's Q1 data-center revenue of $75.2B (+92% YoY) and the $91B Q2 guide arrived in the same window. [1] The substitute-versus-complement question for Cerebras now has two anchors: its own price and Nvidia's number.

The paper's Tuesday brief on Berkshire and Cerebras took the position that the IPO was a watch item, not a thesis. Friday's slide is the first datapoint testing the wafer-scale pitch against the analyst notes still being written. Customer-concentration risk is the recurring objection: a single hyperscaler relationship can carry a wafer-scale roadmap, and a single hyperscaler exit can break one. Cerebras has not published a current customer-concentration table since the S-1 update.

The watch items now have specific thresholds: a return to the debut highs reopens the substitute argument; a continued slide toward the IPO price closes it for the quarter. The IPO was the largest of 2026.

The first-day pop and the Friday slide are paired prints from the same week. They tell two different stories: institutional appetite for an Nvidia alternative is real, and that appetite priced the early float above where the next round of analyst notes are willing to model. The customer-concentration question — disclosed in the S-1 update and not since revisited — sits behind both prints. Day five is the silence between them.

-- THEO KAPLAN, San Francisco