Intel’s $20 bn vision for a manufacturing comeback will take years to execute

New CEO Pat Gelsinger assures us that “Intel is back,” but fixing the company’s manufacturing woes won’t happen overnight. In only his sixth week on the job, Gelsinger laid out a broad plan to build new chip factories, expand outsourcing, keep next-generation technology on schedule, and seek help with future generations. All are good ideas during an industry-wide chip shortage, but that shortage will abate before any of these initiatives kick in.

The problem for Intel is that its transistor technology, measured in nanometers (nm), has fallen behind that of Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC), which build chips for rival AMD. Intel first fell behind when its 10nm transistors were delayed for three years. Last year, the company disclosed delays in its next-generation 7nm technology, which is now due in 2023. These delays ultimately cost the former CEO his job.

Doubling down on manufacturing
Intel announced plans to spend $20 billion on two new factories, called fabs, in Arizona that will produce advanced chips. This shouldn’t rate as news; the company regularly builds new fabs to replace outdated ones, and $10 billion is the going rate for leading-edge fabs. The spending is in line with Intel’s recent capital expenditures of about $15 billion per year. In any case, new fabs have long development and bring-up times, so these factories won’t start volume production until 2024. By comparison, TSMC is building 11 new fabs between 2019 and 2024.

The new fabs, however, represent a serious commitment by Intel to continue manufacturing its own chips. Many investors have urged the company to sell its fabs and outsource all chip manufacturing to gain access to TSMC’s leading-edge transistors. But Intel clearly isn’t taking that route. As a sop to these investors, Gelsinger said the company will expand its outsourcing but gave no details. Given the ongoing fab investment, this outsourcing expansion will likely apply to ancillary products such as GPU chips and AI accelerators rather than to mainstream PC and server processors.

After last year’s delay, Intel’s 7nm technology is on track to reach an important milestone next quarter, according to the CEO. But we won’t really know if the problems are fixed until the company begins to fabricate samples of the initial 7nm product (code-named Meteor Lake) around the end of this year. Even if everything goes well, the 7nm design will be more than two years behind TSMC’s comparable technology. In fact, TSMC plans to debut a new transistor in the same timeframe, leapfrogging Intel again.

The danger for Intel is that AMD will maintain a technology lead by using TSMC to build its PC and server processors. AMD’s current “Zen 3” products (Ryzen 3 for PCs and EPYC 7003 for servers) already employ more advanced transistors than Intel’s desktop PC and server processors; these smaller transistors reduce manufacturing cost and save power (electricity) as well. AMD plans to leap forward by another transistor generation next year, and unless Intel can accelerate its roadmap, it will never catch up.

A partnership for the future
Thus, the final leg of the new strategy may be the most important: Intel is joining forces with IBM to develop new transistors beyond 7nm. IBM has already developed two new generations in its labs, and Intel can help bring this cutting-edge technology into production. The partnership could enable the company to close the gap with TSMC, although given its existing plans for 7nm, no sooner than 2025.

Thus, the plan for Intel to come “back” won’t kick in until 7nm ships in 2023, the new fabs come online in 2024, and the first Intel-IBM technology ships in 2025. In the meantime, Intel’s inferior transistors leave it fighting an uphill battle against AMD. Transistor technology is just one aspect of processor design, and Intel still excels in other areas, but it’s a good time for CIOs to take a look at alternative processors.

The author of the article is Linley Gwennap, President of The Linley Group and principal analyst, focusing on processors for servers, mobile, and deep learning. He is also publisher and editor-in-chief of Microprocessor Report. The article was fisrt published in Forbes.