Intel's Tower bid has shuffled off this mortal coil – so what about foundry plans?
Maybe Pat dodged a bullet – mature process nodes aren't the kind of thing shareholders get excited about
Analysis With its $5.4 billion bid to acquire Tower Semiconductor in ruins, Intel Foundry Services' (IFS) master plan has been turned on its head.
The failed buyout would not only have unlocked a bounty of mature process tech to compliment Intel's leading-edge nodes, but also much-needed expertise in how to run a successful foundry business.
While Intel has operated its own fabs for decades, the silicon slinger doesn't have any experience in the foundry arena, Gartner analyst Gaurav Gupta told The Register. The "integrated device manufacturer (IDM) business is very different from foundry. You have to have customer orientation, the service [mindset], the IP portfolio, all those customer relationships, and so on."
Buying Tower was a way for Intel to address many of these shortcomings, but unfortunately for Intel CEO Pat Gelsinger, Chinese regulators had other ideas. Faced with lengthy delays Intel this week admitted defeat, paying Tower $353 million to terminate the deal.
Despite the failed acquisition, Gupta doesn't expect it to have any major impact on IFS's buildout. "Yes, it would have brought that ecosystem experience and expertise, but leading-edge foundry business and competing against TSMC is a very different ballgame," he opined.
No going back
Since announcing its entrance into contract chip manufacturing in 2021, Intel has committed tens of billions to the project, brought on private equity partners, and lobbied for a sizable chunk of both US and EU subsidies.
To date, the chipmaker has fabs worth more than $70 billion under construction in Arizona, Ohio, and Germany. It has also announced billions more in upgrades to retrofit many of its existing facilities for contract manufacturing of custom silicon designs.
I think Intel's best bet is to get one of the hyperscalers to commit to them
So, while Tower would have played a key role in advancing Intel's foundry buildout, the chip giant was by no means dependent on it, Gupta emphasized.
And ultimately, Intel may be better off focusing on leading-edge process tech. Intel has been stuck on 10nm process tech for years, but is working to bring several more advanced leading-edge nodes to market. These include the long-delayed 7nm process – which it has taken to calling Intel 3 – and a 2nm node it is calling Intel 18A.
"From the foundry business perspective, obviously, the margins are more on the leading edge, which is where Intel plays," Gupta said – adding that, even if an acquisition is off the table, there's no reason that Intel can't leverage its advanced packaging tech to integrate other fabs' chiplets with its own.
It's worth noting that while the industry is moving toward a heterogeneous chiplet ecosystem, industry veterans tell us it's still a few years away from reality. So, while partnering with other foundries is certainly possible in the long term, it may not help Intel in the immediate future.
A rocky road ahead
That's not to say IFS isn't without challenges. While Intel has been talking up its advanced packaging and process tech for years now, Gupta says it's time "to move beyond just having the right technology, to proving that it's a viable option for customers."
That means winning customers – big ones. "They have announced a few customers like Ericsson [and] MediaTek for Intel 18A [and Intel 16], but still, they need to do more to get some of those customers up," Gupta explained.
- Biden administration restricts US investment in tech China's military might employ
- Beijing's silent treatment topples Tower Semiconductor merger with Intel
- TSMC and pals chip in for €10B German fab
- Gelsinger: Intel should get more CHIPS Act funding than rivals
Part of the problem, he argued, is that many of the manufacturers that consume vast quantities of leading-edge semiconductor capacity also compete with Intel. We suspect this could make contracting with Intel for manufacturing dubious for chipmakers like AMD and Nvidia, especially if they run into challenges in production.
Instead of courting its rivals, Gupta suggests Intel could find success building chips for the major hyperscalers, many of which are developing custom silicon with the help of TSMC. "I think Intel's best bet is to get one of the hyperscalers to commit to them," he stated.
An end to semiconductor M&A
More concerning for the broader semiconductor industry is the possibility that Intel's failed M&A bid may not be the last in the biz.
Over the past few years, several large chipmakers have used their financial position to expand their IP portfolios and addressable markets through mergers and acquisitions. AMD paid $49 billion for Xilinx; Nvidia spent $6.9 billion on Mellanox and attempted and failed to buy Arm for $66 billion.
But acquisitions of this scale often require regulatory approval in multiple regions – most notably China.
Worsening trade relations between the US and China – fueled in no small part by sanctions on sale and export of semiconductors, intellectual property, and AI accelerators – could leave M&A activity open to retaliation by regulators.
"You don't want to invest into something, do all the due diligence, wait for 18–24 months, and then have the deal not go through," Gupta said. Intel's $353 million payout to end the takeover must have been a bitter pill for Gelsinger to swallow.
Whether retaliation over the Biden administration's latest restrictions had anything to do with why Intel's Tower acquisition fell apart isn't clear, but the result may well have a chilling effect on the industry all the same. ®