It looks like chipmaker Intel has hit rock bottom in its product and foundry business in the second quarter of this year, and that sales are slowly – we won’t go so far as to say for sure – improving. But now the restructuring expenses and cost cutting will take effect and the bottom line will look a bit ugly for a while.

Then Intel will run out of excuses, and hopefully in time for its 18A manufacturing process to catch a bit of fire, including developing chips for myriad purposes and applying 18A processes to Intel’s homegrown client and server products. In other words: 2025 should be a lot better than the misery that prevailed in 2023 and 2024.

“Operationally, third-quarter results exceeded our expectations as we achieved key milestones at Intel Foundry and Intel Products,” Intel CEO Pat Gelsinger said in a call with Wall Street analysts discussing the numbers . “Underlying business trends are improving at a moderate pace and our fourth quarter outlook is slightly above current consensus. Overall, our increased focus on efficiency and implementation is having a positive impact across the company. We still have a long way to go and are acting urgently to achieve our priorities. We need to fight for every inch and perform better than ever before, and our teams are adopting that mindset as we build a leaner, more profitable Intel.”

In the quarter, Intel posted revenue of $13.28 billion, down 6.2 percent year over year, and posted an operating loss of $4.6 billion across all of its groups and units, a stark contrast to the $8 million loss Intel had in the same period last year. There are also another $5.62 billion in restructuring and other costs – much of it related to the layoffs of more than 15 percent of Intel’s workforce ($2.2 billion), goodwill impairments for Mobileye unit ($2.6 billion) and also the write-off of investments in equipment used to make chips using the Intel 7 process ($3 billion), which uses geometries in the 10-nanometer range etched and cannot be further developed to support extreme ultraviolet lithography on smaller transistor geometries. Oddly enough, Intel also recorded a $7.9 billion tax provision, giving the company a staggering net loss of nearly $17 billion.

Ouch.

Incidentally, operating income was hit by a $300 million write-down of accelerator inventory – presumably for Gaudi 2 and Gaudi 3 devices, but possibly also for some Max-series “Ponte Vecchio” GPUs that Intel may have lying around had , “due to lower sales expectations.”

Gelsinger said in the conference call that “overall adoption of Gaudi has been slower than expected, with adoption rates impacted by the product transition from Gaudi 2 to Gaudi 3 and the ease of use of the software.” And for this reason, Intel’s forecasts will change in 2024 Bringing in $500 million for sales of Gaudi accelerators (mainly for Gaudi 3 devices, we assume) doesn’t materialize. With a depreciation of $300 million, this seems to mean that Intel is only selling $200 million of Gaudi 3 devices. That means that of the 32,000 Gaudi 3 accelerators that Intel planned to sell in 2024, only 12,800 found a home. If you want inexpensive Gaudi 3 accelerators, Intel has 19,200 of them available. That’s just over 40 exaflops at BF16 floating point precision, which wouldn’t be a not-too-shabby cluster by any standards. However, you will have a lot of work to do with the software stack, and the Intel people who could help you may no longer work there.

So far, Intel appears to be sticking to its plan to evolve the Gaudi series into the “Falcon Shores” hybrid of its Ponte Vecchio/Rialto Bridge GPU and the Gaudi 3 design, and has iterated forward a notch or two. It would be a very bad thing if Intel abandoned AI acceleration, even if it costs money in the short term. The AI ​​server market is not going away, and as we have shown in many predictions, it will eventually account for half or more of AI server spending. Intel cannot afford to abandon this entire market, even if it needs to become a world-class manufacturer again.

By the end of the decade, Intel aims to generate external revenue of $15 billion per year from this foundry, which is slightly less than what Intel Products group Intel Foundry pays for etching and packaging today. Hopefully, if Intel Foundry generates more than $30 billion in annual revenue – and possibly even more – it will actually be a profitable business. Which is certainly not the case today, as you can see in the table below:

The main thing to be happy about in this chart is that the Client Computing Group, which makes CPUs and GPUs for PCs, is holding its own and accounting for the bulk of the free money Intel is generating. This is the fuel that will help Intel get through this terrible time.

If Intel had started a real data center GPU business back in 2012, when it was clear that the company would need one, and had lived up to its potential, the Data Center & AI group wouldn’t be in such bad shape right now. DCAI achieved revenue of $3.35 billion, up 8.9 percent year over year and up 10 percent quarter over quarter, thanks to the increase in the Xeon 6 server CPU line. With an operating profit of just $347 million, presumably reduced by $300 million in depreciation presumably related to the Gaudí accelerators not selling like hot cakes, operating profit fell by 11, Down 3 percent, but up 25.7 percent quarter-on-quarter.

We are a long way from the almost 50 percent operating margins that the old Intel Data Center Group once achieved.

But as this chart shows from the new groups Intel created in 2022 to describe itself to Wall Street (and which we’ve expanded into 2021), things have more or less stopped getting any worse, and that’s something. This is the only reason we can imagine that Intel stock should have risen in aftermarket trading following the report to Wall Street.

As you know, we like to show the trend lines of Intel’s “real” data center business over time, which has become much simpler with the closure or sale of its various network switch businesses (Barefoot Networks and Omni-Path). and the spinoff of its flash storage business.

As far as we can tell from our model, Intel’s actual data center business is 5/9 the size it used to be and is only nominally profitable. Assuming no surprise write-downs and charges in the future, this data center business, which at this point includes all of DCAI as well as portions of Altera and NEX and is much less complex, should grow as Intel’s 18A processes for future “Diamond Rapids.” “P-Core and “Clearwater Forest” E-Core Xeon 7 processors are used. Gelsinger said Clearwater Forest is on and Diamond Rapids will go to the factory shortly for etching.