“No Quick Fixes” As Intel Losses And Restructurings Continue

Intel’s new chief executive officer, Lip-Bu Tan, has his work cut out for him, just like his predecessor, Pat Gelsinger, did several years ago. And given the even worse state that Intel is in – generating less cash on lower sales and therefore making it that much harder to raise cash to invest in the foundry that can make better chips in the future – Tan is in a much tighter spot than Gelsinger ever was.

But it is important to not lose hope. Look at how IBM, Microsoft, Apple, and AMD have all rebounded from their near death experiences. It is very difficult to make something as big and creative as any of these companies go to zero.

That said, the situation with Intel right now is not pretty, even if it will improve financially with layoffs and a flattening of the hierarchy of the company that Tan is now imposing and that could result in as much as 20 percent of the company being laid off on top of the 15 percent that have already been let go.

“My focus will be ensuring that our team builds products that are highly competitive and meet the needs of our customers as we enter a new era of computing, defined by AI agents and reasoning models,” Tan explained in a call with Wall Street analysts going over the numbers for the first quarter of 2025. “To achieve this, we are taking a holistic approach to redefine our portfolio to optimize our products for new and emerging AI workloads. We are making necessary adjustments to our product roadmap so that we are positioned to make the best-in-class products while staying laser focused on execution and ensuring on time delivery. However, I want to emphasize that this is not a quick fix here. These changes will take time.”

By the time Intel fixes this, half of the world’s CPUs will be homegrown Arm chips being designed by the hyperscalers and cloud builders, and the remaining half could be split evenly between Intel and AMD. A decade ago, Intel had north of 97 percent shipment share for server CPUs and 85 percent of revenues. And Intel really does not have an AI accelerator that it can afford to make in volume and sell at a competitive price that customers want to buy, after two such attempts so far: the “Knights” family of many-core HPC processors from a decade ago and what we will call the X^e family of GPUs that have been installed in the “Aurora” supercomputer at Argonne National Laboratory and pretty much nowhere else. The latter was supposed to be converged with the “Gaudi” family of AI accelerators about now with “Falcon Shores” devices, and now the future is being pinned by an as-yet undefined “Jaguar Shores” accelerator further out in the future.

We may never see Jaguar Shores, depending on what Tan cuts and when. And with AI servers now comprising half of server revenues, and the AI accelerators comprising most of the revenue and profits inside these machines, Intel is missing out bigtime on a revenue stream that could at least help fill the server CPU void and pay for its foundries.

Like we said, it is not pretty. But, don’t think for a second that every datacenter in the world is not looking for a cheaper alternative than Nvidia GPUs. Intel has to do to Nvidia in this era what it did to IBM, Digital Equipment, Sun Microsystems, Hewlett Packard, SGI, Fujitsu, NEC, Siemens-Nixdorf, and a handful of other companies that few remember.

Whatever Intel does, it will focus on rackscale integration – something that it has experience with but that it has not successfully commercialized. We would not be surprised to see Intel go all the way and try to disaggregate compute, memory, storage, and networking to make a more modular and upgradable system – although we concede that this may be a bridge too far from a manufacturing standpoint, just as the “Ponte Vecchio” X^e GPU was with its many levels of packaging and 47 chiplets on that package.

We shall see.

Intel has to deal with the present if it is going to build the future. In the first quarter, revenues were down four-tenths of a percent to $12.67 billion, and were driven by strong than expected sales of Xeon server processors. But because of high foundry costs due to lower utilization rates and aggressive pricing in the market for both server and PC CPUs, Intel booked and operating loss of $301 million and a net loss of $887 million.

The good news is that Intel has done away with the Network and Edge Group and how has a datacenter and a PC group, with Altera spun out separately and Mobileye now in the “Others.” Intel has sold off its software and flash businesses, written down the acquisition of FPGA maker Altera, and now we have a clear view of its datacenter business, which we have not had for many years.

Here is a table of the new groups, with the financials rejiggered for 2024:

Here is this same revenue data represented visually for those of you who like pretty pictures:

Here is what the revenue and operating profit looks like for the datacenter business looks like over the years at a large grain level:

We added NEX and Altera to the DCAI business as a proxy for the old Data Center Group in recent years, but have also done a finer grained analysis across all of the variosu technoliges Intel has sold into the datacenter and at the edge to create what we called Intel’s “real” datacenter business.

With the latest financial reporting tweaks from Intel, revenue and operating profit numbers for the datacenter are exactly and only what the Data Center & AI group reports. We don’t have to cut off the datacenter portions of all of the various divisions and try to reassemble a single pair of revenue and operating profit numbers.

And so, we do have all of those years of data to compare the current datacenter business against, which is cool, and it looks like this:

In the quarter, the DCAI group posted sales of $4.13 billion, up 7.8 percent and only down sequentially 5.2 percent, which is not bad for a Q4 to Q1 transition. Operating profit for DCAI was $575 million, up 37.9 percent year on year. These numbers are heading in the right direction.

Intel Foundry is the boat anchor that the Intel Products group is dragging behind it, as you can see in the table. Intel has a tad more than $21 billion in the bank, and now plans to spend $18 billion on capital expenses, down $2 billion from previous targets.

Perhaps, if the US government is so keen on indigenous chip manufacturing, it could extend Intel loans instead of giving it cash with different kinds of strings attached. Uncle Sam did this for the US automakers during the coronavirus pandemic, and they all paid back their loans when they recovered. To be fair, Intel made its own pandemic through greed and negligence, but it also has capital expenses that are unique in a vital industry. If cash is going to limit Intel’s recovery, then cash can speed it up, and if that is indeed in the national interest then perhaps we should act accordingly.