Cash-strapped Intel looks for $3B in savings to pursue '5 nodes in 4 years' dream
Cheap as chips? Not in this case
Intel is looking to balance investments in technology with cost cutting to offset reduced revenue as it prepares to introduce a number of new production nodes over the next few years.
The Santa Clara chip giant reported a $700 million net loss for the fourth quarter of its 2022 financial year, and saw revenues drop 32 percent, partly due to declining PC and laptop sales following the boom in demand during the pandemic.
Yet Intel still has to invest in future chip development, which also means investing in new production processes and fabrication plants, such as those it is preparing to build in Ohio and Germany, to actually manufacture the chips.
Speaking at Morgan Stanley's Technology, Media and Telecom conference, Intel chief financial officer David Zinsner said the company had made some progress on cost reductions "even in the back half of 2022," but that the "lion's share" would come in 2023.
And that $3 billion is just a down payment on the $8 billion to $10 billion savings that Intel plans to make by the end of 2025.
"It's roughly about $1 billion in cost of sales, and about $2 billion in opex is the way I would look at it," Zinsner said.
For the cost of sales, this would be largely just "tightening the screws on various areas of the spend," while more meaningful savings are expected to come from reducing the workforce "at least on a temporary basis," he explained.
On the opex side, Intel looked at all overheads and set a target to reduce them by about 25 percent versus a baseline of the first half of 2022, the CFO claimed.
"I think as many people know, there's some news out last week around a more streamlined graphics roadmap. So, that was part of it," Zinsner said, referring to the recent announcement that the Rialto Bridge GPU has been canceled and the Falcon Shores CPU-GPU chip pushed back to 2025.
However, Intel has to make certain key investments, especially as it plans to introduce a number of new production process nodes over the next few years, to catch up with rivals that it has lost ground to.
There are investments "necessary to drive five nodes in four years" which Intel absolutely does not want to touch, Zinsner said. "That's important for a number of reasons and it really drives the entire strategy. So five nodes in four years has to happen, we will make the capital investments necessary to go make that happen."
Zinsner did not specify, but from Intel roadmaps, it appears the five nodes comprise the Intel 7 process already in production; the Intel 4 process to be used for the Meteor Lake chips; an enhanced version called Intel 3; and then Intel 20A and Intel 18A, which are slated as 2nm and 1.8nm nodes.
Intel also needs to invest in manufacturing these chips. "And we have a heavy bias towards wanting to invest there because it's always cheaper to build the clean room than it is to equip it," Zinsner said, explaining that Intel likes to have a few "empty shells" ready to be kitted out quickly when the time is right.
- AMD flips the bird at Intel as it glides past in CPU-GPU stakes
- Intel rattles the tin for another €5B in subsidies to build German fab
- While Intel XPUs are delayed, here's some more FPGAs to tide you over
- Intel buries news of GPU cuts and delays in low-key Friday post
"The rest of the capital in equipping those clean rooms is really where we flex. And we're essentially looking at wafer demand out over 5 to 10-year periods and looking at what percentage of the share we expect to get of those nodes from an internal perspective in our own products and from an external perspective in terms of foundry customers," he said.
Intel has constrained spending there because demand has changed, he added.
When asked about Intel defending its market share, Zinsner claimed the company was doing well on the client device side, but the picture was more mixed from a datacenter perspective.
"Alder Lake was a good product. It's performing really well, as is the follow-on product Raptor Lake. And we'll have Meteor Lake out this year, on Intel 4," he said.
On the datacenter side, "Sapphire Rapids is out and ramping, which is important," Zinsner said, but conceded that "it was not at the time we originally expected it."
"I think it's going to be one of our fastest ramping if not our fastest ramping product that we've ever had," Zinsner claimed, and turning to follow-on products, he said that Emerald Rapids is coming, but the products he thinks would make a meaningful difference to Intel's competitive position will be Sierra Forest and Granite Rapids, due in 2024.
This will be for a couple of reasons: "The performance of Granite relative to what customers need and what the competitors will have is going to be fantastic," he said. "But also we will have our first E-core line [Sierra Forest], a really efficient core that answers the needs of customers that really have, kind of, a power crunch issue that really needs a more efficient core," he explained.
The E-core or Efficient Core is Intel's power-optimized core design, and Sierra Forest will use these to target cloud workloads, while Granite Rapids uses the high performance P-cores.
During the discussion, Zinsner admitted the US CHIPS Act tech subsidy played a big part in Intel's decision making over the last year or so.
"CHIPS Act was a big part of our strategy," he said. "We're happy to see that get approved and now it looks like the mechanics of how to apply for that is now starting to become clearer.
"There's also a European version of the CHIPS Act. And we're already in discussions with our European partners on offsets there to help contribute to the capital investments."
The word pork barrel springs to mind. ®