I think we’d heard about this before but for some reason it’s in my headlines today.
Basically the problem with bespoke accelerators is that if the market shifts you can’t necessarily repurpose what you were working on…
As an ASIC maybe it wasn’t a super-heavy investment. But NVidia may have the right idea on building something sufficiently general-purpose that it can with modest modifications be fit for crypto mining one year and AI the next.
Continually missing schedules indicates to me that they were spread too thin. Eliminating the “marginal” products should help with this, but it will be a couple of years before the effect is seen. The Intel R&D budget is still HUGE compared to AMD, NVDA, or TSMC.