How do you forecast demand for a new chip product with limited data?

When historical data is thin, I usually start with analog markets — find a similar component that went through adoption 5-10 years ago and use that adoption curve as a proxy. Combine that with bottoms-up sizing from design win pipelines (talk to a few FAEs or distributors, they often know more than they let on). Gartner and IDC have some niche semi reports worth the price if your company has a subscription.