Throughout the 59-year history of Moore’s Law, the only consistent truth in the semiconductor industry has been that performance wins. Competitors come and go, and new technologies like PCs, smartphones, and artificial intelligence rapidly change the landscape. But in the end, the company with the best technology and ability to manufacture in high volume wins the prize.
For decades, this was Intel. The company’s mantra was to drive technology leadership at all costs and push Moore’s Law to the limit. That’s how Intel became the most successful semiconductor company on the planet, with manufacturing technology consistently one generation (about two years) ahead of everyone else.
During the last decade, this all changed. Intel stumbled and lost its leadership position—and now it’s viewed by some as just another company struggling to survive.
Some pundits think there is a simple solution: split the company into two. That way, their argument goes, the Intel design company can compete with other chip designers (AMD, Qualcomm, NVIDIA, etc.) while the foundry part of the company can be free to serve all chip designers. This simplistic solution ignores the existence of Moore’s Law—and would be bad for Intel and the United States.
Every design company wants the best manufacturing technology available so as to maximize the performance of its chips. This holds true for companies that sell chips on the open market as well as those that design for the internal use of the likes of Apple and Amazon. Whichever foundry has the latest and greatest technology will win the lion’s share of all the chip designer’s business. Today, that position is held by TSMC. Much of the industry relies on the same manufacturer to build their products—their only manufacturing advantage is TSMC’s technology leadership.
Intel can attract these same chip customers by restoring its foundry technology leadership. Even if you compete with Intel designs, why put yourself at a disadvantage? As long as Intel has the best technology and is competitive on price, competitive issues can be resolved.
This is all unfolding at a time when the world needs a more globally diverse and resilient semiconductor supply chain. Geopolitical tensions around the world are escalating, making the urgent need for having a strong U.S. manufacturing capability on our own shores even more pressing.
Splitting Intel into two separate companies wouldn't do the U.S. any good if the Intel design business succeeds and the foundry business does not. In that case, the U.S. would remain dependent on a foreign supplier for leading-edge technology—and the $50 billion in the CHIPS ACT would have been wasted.
We’ve seen this movie before. Years ago, a struggling AMD split off its manufacturing capacity into Global Foundries. Pundits applauded the split at the time. A decade later, AMD is doing well using TSMC, while Global Foundries has little if any differentiated technology. Global Foundries just didn’t have enough research and development (R&D) budget, and with limited production and revenue, struggled to keep up with market leaders.
The economic reality is that it takes massive investment to drive Moore’s Law. In today’s semiconductor industry, only three companies (Intel, Samsung, and TSMC) have sufficient revenue to contend for technology leadership. If you split up Intel, the foundry portion will probably fail because of decreased R&D spending along with the complex realities of splitting up a huge multinational company in the midst of a multiyear turnaround effort.
Instead of wasting time and effort splitting Intel into two separate companies, why not focus on the real issue? The Intel of tomorrow needs to be like the Intel of 15 years ago—the driver and leader of Moore’s Law. That way, you get a win-win scenario with U.S.-based design and manufacturing technology.
This starts with rebuilding Intel’s technology leadership. The current CEO, Pat Gelsinger, has precisely the right strategy and attributes, and he is already driving the right changes.
Intel is on the verge of completing an unprecedented pace of node development to catch up to TSMC. It has taken the lead on next-gen technologies that will shape the semiconductor industry for years to come, such as high NA EUV lithography and backside power delivery. Yes, more work is needed—but this is a good start, and they must keep going.
The U.S. government must do its part as well. This includes increased funding of basic pre-competitive research in semiconductor technology in our research universities and national laboratories. The U.S. has made a small step in this direction with the creation of the National Semiconductor Technology Center, but there is still a ways to go, especially when you consider the NSTC budget for five years is less than what Intel spends annually on R&D.
There are immense challenges as leading manufacturers race to create technologies that combine 100 billion transistors into a piece of silicon the size of a fingernail. These are the most complicated things mankind has ever built, and semiconductor leaders will spend tens of billions of dollars to make it happen.
Let’s not kid ourselves: If the U.S. wants to be a leader in this game once again, simply slicing Intel in half is not the solution. We must be willing to invest and do the hard work of leading Moore’s Law into the future, not waste time rearranging the deck chairs.
I remember a similar time in Intel’s history—the dot-com bubble crash of the early 2000s. Customer demand evaporated. Wall Street said we should have layoffs, close factories, and cut R&D spending. The Intel board agonized over the situation but followed management’s plan to maintain investment in R&D and build new factories. The stock price crashed, but when demand returned, Intel was in a stronger position than before. It wasn’t pretty but it was the right thing to do. Pat Gelsinger is doing the right thing now.
Former Intel CEO Craig Barrett: Splitting up America’s leading chipmaker is a bad idea
'This simplistic solution ignores the existence of Moore’s Law—and would be bad for Intel and the United States.'
