Yesterday, Intel announced that it would delay opening its Fab 42 in Arizona, with no word on when it would finish upgrading the facility and ramp up its production. Intel spokesperson Chuck Malloy characterised the decision as a minor course correction, noting that Intel’s fab utilisation remains at 80 per cent.
“If we can maintain that 80 per cent capacity with the existing space, why spend the additional capital?” Mulloy said.
This story has got a lot of attention from people who see it as triumphant proof that Intel will go down in flames to ARM. On the other side, you’ve got Intel itself insisting that this is no big deal, with no real impact on Intel’s long-term roadmap. In this case, Intel has the better argument, but there are long-term implications that Malloy is glossing over.
The long road to ramp up
There’s a famous saying sometimes attributed to Robert Palmer, one-time CEO of Digital. “Building semiconductors is like playing Russian roulette. You put a gun to your head, pull the trigger, and find out four years later if you blew your brains out.”
Building fabs is extremely capital intensive. There are huge fixed costs associated with equipment installation, ramp, and maintenance. It’s far more economical to keep a smaller number of fabs running near full speed than to run a larger number at half or quarter speed. So when Intel decides to delay opening a facility, it’s doing so because it really doesn’t make good sense to bring the new facility up at this point. And unlike AMD, which often relied on a single fab for building its cutting-edge products, Intel has new processes ramping up at multiple locations. So does this impact Intel’s long-term roadmaps? Almost certainly not.
That leaves the second question – is this a sign of the ARM foundry model putting pressure on Intel? In a general sense, yes – ARM tablets and smartphones have certainly stolen business from the PC market, and Intel is making decisions based on that situation. Delaying the Arizona build-out, however, is more about efficient capital management than existential threats from the ARM ISA. Intel is simultaneously courting new foundry customers of its own, ramping 14nm in other places, and could still choose to make Fab 42 the centrepiece of a future 10nm production node.
So long as Intel continues to push ahead on executing future nodes and researching advanced semiconductor manufacturing techniques, the competitive implications for delaying the Arizona Fab 42 opening are limited. Changing those goalposts would be a sign that the company has run into more fundamental problems.