I was at GloFlo when it happened. They shut down because of lack of customers and Mubadala did not want to fund another process tech which was not going to bring any profits.

But your point is still valid, even if they did somehow come up with a process recipe, with HVM worthy defect density, meeting their perf targets, they have to get customers. All this will easily take a decade.

Because they have been using FPGAs until now.

This is the greatest gift by TSMC. They have made it their life mission that designing chips should not be some black magic and any company should be able to do it. All the work they have put in over the last 2 decades, working with EDA vendors, universities, companies. Making pdks realistic, getting their silicon to hit the promised targets, enabling IP vendors to quickly iterate and get to GDS as quickly as possible.

This was enabled by entire mobile ecosystem ARM. Open source arch + TSMC’s design ecosystem made designing chips to be a relative cake walk.

Look at Tesla, they don’t design most of their IP. They focus on compute IP, system. All the others, they buy it off the shelf from IP vendors. And they don’t have to worry because IP vendors knows their IP will work on the process already.

This era enabled companies to focus on what’s important to them, like compute and not on IP building.

Intel robbed the industry of progress by pretending chip design needs a 20K+ people, IP teams, multiple BUs, architects.