Analysis There has already been much speculation about who will succeed Pat Gelsinger as Intel chief, but perhaps a more pertinent question is whether the company wants to continue to be a manufacturer or join the ranks of fabless semiconductor firms.
The chipmaking giant embarked on a turnaround plan envisioned by the outgoing Gelsinger to get back to its position as a technology leader and profit-making operation after reporting three successive quarters of losses this year.
As well as cost-saving measures that included laying off more than 16,000 staff or about 15 percent of the workforce, part of the plan was for Intel to spin off the actual chip manufacturing side – its foundry business – and delay expansion of its fabrication plants in the meantime.
Gelsinger discussed some of this in August, saying that Phase 2 of his IDM 2.0 strategy was about becoming more efficient and implementing cost actions. A month later, it was confirmed that Intel would spin out the Foundry division as an independent subsidiary, with its own board, in the hopes of bringing in new sources of capital.
But there was always the suggestion that the Santa Clara megacorp might cut its losses and spin off Foundry completely as this part of the business has been bleeding revenue, leading to a $7 billion operating loss in 2023. Is Intel prepared to do this?
"Anything at this point would be speculation," said Gaurav Gupta, Gartner VP Analyst for Emerging Technologies and Trends.
"With Pat being there, it seemed there was a strong inclination in keeping IDM and Foundry under the same company. They had created a subsidiary, but still it belonged to Intel," he added.
However, the problem with this arrangement is that any customers wanting to have silicon made by Foundry are likely to be competitors, and they could easily be discouraged by even the possibility of their IP being accessible to Intel's design engineers.
"Now, with Pat gone and statements from the board about Intel renewing focus on products and becoming a leaner company, this indicates a bigger separation between the IDM and Foundry (or products and manufacturing as Intel likes to call it), but how that might look and if/when that will happen is unclear," Gupta said.
The notion of a complete spinoff was dealt a blow by the US government, which finalized its deal with Intel for $7.86 billion in subsidy cash as part of the CHIPS and Science Act at the end of November, but with strings attached.
Those strings are that the CPU supremo must retain 50.1 percent ownership and/or voting rights in its Foundry business, and if that operation goes public, no single shareholder will be permitted to hold more than 35 percent of its stock – unless Intel remains the largest shareholder. The government doesn't want to sink public funds into assets that might then end up in foreign ownership, after all.
This means that Intel will be unable to realize the full value of its chip manufacturing arm unless it is prepared to also forgo that CHIPS Act subsidy. Whether this would go to the Foundry biz or just be quietly retained by the Commerce Department isn't clear, but Gupta believes the former.
"The CHIPS Act is for bringing resilience to domestic manufacturing and whether those fabs stay with Intel or get spun out, as long as they keep their commitment to building capacity and hit their milestones, they will get funding," he claimed.
But Foundry as a separate entity would likely find itself under financial constraints that would make it difficult to sustain the research and development effort needed to stay at the cutting edge of process node technology, Gupta warned.
There is also a more serious barrier to Foundry operating as a standalone contract manufacturer along the lines of Taiwan's TSMC, and that is its lack of design libraries and other tools that are required for third-party chip customers to make their IP work with Intel's production processes.
"TSMC has all this expertise in place, but it took them 20 years to get there," said Andrew Buss, IDC Senior Research Director for Europe. "Intel's fabs are still intertwined with its product group."
He told us that it was a target for Intel's upcoming 18A production node to have the design libraries and tools in place, but that it would still take time for the Foundry division to be able to operate as a full-blown contract manufacturer.
As for the company's products division, it already uses other manufacturers, namely TSMC, to produce silicon, which is shipped to Intel factories for packaging into the finished unit.
In fact, the Santa Clara biz has relied on its Taiwanese partner to produce things like GPUs and NICs for years, but more recently has started using it for some chip "tile" components for processors as well. A few months back, it emerged that Intel was effectively canning its 20A production process and using TSMC to deliver both its Lunar Lake and Arrow Lake CPU parts.
In theory, this division of Intel could survive as a so-called fabless semiconductor company, which focuses solely on the design of its chips and outsources the manufacturing to another firm. Processor rival AMD has operated like this for many years now since it spun off its own manufacturing arm to become GlobalFoundries.
But if Intel were to follow this approach, it may not wear well with the incoming Trump administration, which has vowed to bring manufacturing back to the US and punish companies that make use of foreign manufacturing with import tariffs that would push up the price of their products.
Instead, Gelsinger's plan was to build up Intel's Foundry division so that it could function as an independent contract manufacturer, allowing the company's products biz the choice of targeting whichever fabrication plant and process might be best suited for a particular chip, while the Foundry also served other customers.
"Intel is not big enough to sustain its own fabs anymore," Buss said, as the investment needed to develop cutting-edge production processes and equip the facilities to manufacture them has increased as node sizes get ever smaller.
The company was on track to succeed, according to Buss, but it was a long-term plan that could easily take a decade. It seems that Wall Street, and Intel's own board, didn't want to give Gelsinger that long.
"The challenge for the board now is whether to continue to drive that vision forward, or do they go back to Intel's fabs being intimately connected with its own products," he said.
Gupta echoed the view that Gelsinger had been doing the right things prior to his departure.
"Pat had a strategy with IDM 2.0 – building capacity and pushing through the CHIPS Act to get funding, but he inherited a lot of problems when he became the CEO and the rise of AI/GenAI didn't really help Intel, so the company continued to fall behind and bleed money," he said.
"I think going forward we will see a much stronger focus on products – unless Intel has figured something out on keeping manufacturing performance and cost effectiveness against TSMC – a very gigantic task."
The rise of AI has seen a huge amount of money poured into GPUs rather than CPUs, a trend that analyst firm Omdia has been closely following. It led to the paradoxical trend of overall server shipments declining, while server market revenue kept rising because of the cost of GPUs.
"End users are prioritizing investment in highly configured server clusters for AI to the detriment of other projects, including delaying the refresh of older server fleets," Omdia reported earlier this year.
Some, including our partner site The Next Platform, believe that this trend is already over, but Omdia doesn't see that just yet.
"I think that server refresh is still delayed. The fact we reduced our server shipments forecast for 2024 says it all really," said Vlad Galabov, head of Omdia's Cloud and Data Center Research Practice.
"There is a point where further delay is not possible and we expect next year to be a good year for server shipments," Galabov told The Register. This should be good news for Intel as it still accounts for 75 percent of all server CPUs shipped, but it now faces stronger rivals.
"Even as the refresh comes, I think Intel will face tough competition from AMD," Galabov said. "So far, most of the AMD share gain was driven by the top cloud providers opting to use their CPUs, so AMD has plenty of headroom for growth in the enterprise and I think this is where they are headed."
Meanwhile, hyperscalers such as Microsoft and Google are starting to ramp up their own Arm-based server chips, some of which will displace Intel CPUs. As our sister site The Next Platform has said, it is possible to envision a day when AMD, Intel, and the Arm collective all have large and relatively equal shares of the server CPU pie.
But in the short term, it has been suggested that Intel might seek to divest itself of some "non-core" assets, such as its telecoms biz. The company might also seek further Semiconductor Co-Investment Program (SCIP) opportunities, under which private equity firms take a stake in specific fabrication plants, to help offset some of the foundry costs.
Intel currently has two such arrangements; one announced earlier this year with Apollo Global Management for its Leixlip plant in Ireland, and another announced in 2022 with Brookfield Asset Management for its new Arizona fabs.
All of this adds up to a major challenge for whoever takes over Gelsinger's recently vacated hot seat.
"I think it will be a tough time for the next Intel leader. They will need to truly reinvent themselves to succeed in the long run," Galabov said. "I still think an Intel split into a product design group and a foundry business is likely, much like what happened with AMD and GlobalFoundries."
However, the latter has not been able to keep pace with rivals like TSMC and hasn't produced leading-edge silicon since abandoning development of its 7nm process node.
Galabov had this parting shot: "From Pat's sudden departure, I conclude that the news on 18A is not good. I expect that is the next big revelation to come from Intel." ®