Behind President Joe Biden’s high-profile announcement on microchips last week — $8.5 billion in federal money for the Santa Clara, California-based chipmaker Intel — lies a very thorny global tech question that will take much more than a few billion dollars to answer. As part of the 2022 Chips and Science Act, the U.S. is trying to reboot its own semiconductor industry, a business it dominated for decades, using Intel as the flagship vehicle to restore American prominence. But in the past few years, the entire chip industry has been radically reshaped away from the U.S. — and is currently being reshaped again by the forces of artificial intelligence. Intel’s founding is part of tech legend, and the company was an engine of Silicon Valley’s rise. It eventually lost its chipmaking dominance to overseas players like Taiwan’s TSMC and South Korea’s Samsung. And since around the mid-2010s, the most advanced chip manufacturing has happened abroad. Even powerhouse U.S. “chipmakers,” like California-based Nvidia, rely on Taiwanese plants to manufacture the fastest chips they design. The rise of AI is changing this landscape again — creating a booming market for the most powerful, profitable high-end chips, and immense pressure on Intel to recover its lead in time to take advantage. “AI chips” is a broad category that usually refers to specialized products like graphics processing units, which were originally designed for other tasks and increasingly now used to power complex algorithms. At the moment, the AI market is driving even more business to TSMC, because it’s the main producer of Nvidia’s GPUs. But fast-shifting markets tend to create unexpected opportunities, including ones that could open them up, like newer attempts to customize chips just for certain AI applications. The new U.S. policy — spending billions of taxpayer dollars to go from making none to 20 percent of the world’s most advanced chips by 2030 — depends on the idea that all that intricate knowledge and technical knowhow can be re-planted in American soil, or fostered within a homegrown company like Intel. Intel itself is three years into a capital-intensive strategic reboot, and could use the money. When veteran engineer Pat Gelsinger returned as the new CEO in 2021, he set his sights on reinventing Intel as a foundry, a firm other companies trust to make the chips they design. He wants to overtake Samsung as the world’s second-largest contract chipmaker behind TSMC by 2030 — the same deadline the administration gives its subsidized fabs to begin production. Intel’s slump has been blamed on many missteps (some self-admitted): missing deadlines that delayed its transition to smaller process technologies, dividing its attention between microchip design and manufacturing, sticking with older techniques over the latest chip printing machines and losing opportunities in the mobile and AI markets. Another issue has been culture: Intel was known for an insular approach that limited external collaborations. As the industry grew to embrace more information-sharing and collaboration, Intel remained resolutely glued to former CEO Andy Grove’s catchphrase: “Only the paranoid survive.” Chris Miller, a Tufts University historian and author of the book “Chip War,” about the global semiconductor industry, says that cultural change could be a big sticking point. “They were always focused on protecting their in-house secrets,” he said. “Now, they have to share details about their manufacturing with their customers to convince customers to use their manufacturing services.” Viewing AI as its next make-or-break opportunity (and with Gelsinger relentlessly lobbying for the passage of CHIPS to secure federal support), Intel is now under immense pressure to win over customers and regain a spot in making the most advanced logic chips. Its third-place position on the global stage means that for most potential customers, it’s not the obvious choice to power tomorrow’s AI projects. But the company is trying to change that: So far, Intel has publicized a signed deal to make custom chips for Microsoft and says there are four large customers onboard for its next manufacturing process, which it promises will take the technological lead back from TSMC. Gelsinger is also openly courting Sam Altman’s OpenAI to source chips from its new federally-funded fabs. “It’s a vote of confidence by a very big tech company,” said Miller, about the Microsoft deal. “But I think Intel’s got a really long way to go before it attracts enough customers, enough revenue to make sure the foundry business is viable in the long run. I say the early signs are promising.” Intel’s latest corporate reports have been suffused with optimism: Last month, during a tech conference for its foundry business, the company projected beating TSMC to produce the world's fastest chips ahead of schedule. Its latest manufacturing technology — which Gelsinger said he’d “bet the whole company on” — is debuting later this year. Intel has plans to surpass that one with an all-new process in 2026. Gelsinger told reporters that the current AI surge is driving such “insatiable demand” for leading-edge chips that he sees “no fundamental risk” in securing enough customers for all the facilities he plans on bringing online, despite the industry’s notorious volatility Advanced packaging is another space where AI has created an opening — specializing in techniques to combine chips together within single packages that make it easier for companies to get faster processing speeds with existing technology. It’s quickly turning into a non-negotiable for cutting-edge AI work. Ben Bajarin, a semiconductor analyst and CEO at Creative Strategies, predicts investments like the New Mexico plants could be a "real differentiator" for Intel, as TSMC currently scrambles to meet intense customer demand for its packaging services. Still, success will hinge on many other factors, including those cultural adjustments. “You can't do it without the money,” Miller said of Intel. “But the money alone isn’t enough.”
| 
Comments
Post a Comment