Trouble in the data center.
The computing universe is undergoing a fourth major shift, one which may leave microprocessor giant Intel behind, according to an analyst at Jefferies & Co.
As the cutting edge of technology—and market growth—moved from mainframes to mini-computers, then to personal computers and, most recently, to mobile phones, different companies rose and fell, Jefferies analyst Mark Lipacis noted on Monday.
Now the industry focus is shifting again, this time to favor artificial intelligence, neural networks, and the Internet of things. Intel is most at risk and Lipacis urged investors to sell the company’s stock on Monday, issuing a price target of $29. Intel shares, which have already lost 7% this year, dropped 1% to $33.59 in midday trading on Monday.
Get Data Sheet, Fortune’s technology newsletter.
“As the incumbent with dominant share, we think (Intel) has the most to lose,” Lipacis writes. Nvidia nvda , Advanced Micro Devices amd , Xilinx xlnx , and Cavium cavm are all likely to grow faster and gain market share, he says.
The move to smartphones hurt Intel’s PC chip business, but helped its server chip sales fed by the rise of cloud data centers. The server chip side had been growing strongly until recently, as expansion slowed to just 6% in the last quarter. Large companies and cloud data center owners are increasingly moving to chips from Nvidia, while AMD has just launched a competing line of chips for the first time in a few years, as well. But Intel is hardly standing still. It plans to unveil an entirely new line of upgraded chips for servers and data centers on Tuesday. The company is calling it the biggest launch for its server chips in a decade.
Intel is suffering because it has long relied on speeding up chips every few years by cramming in more and more transistors on each chip. But with the slowing of Moore’s law, that strategy has become less effective. Competitors that have designed chips to do more simultaneous work, known as parallel processing, have an advantage in improving performance as the strategy of printing more transistors at a smaller scale becomes less effective.
“Computing platforms historically relied on higher clock speeds in successive generations for improved platform performance,” Lipacis notes. “However Moore Stress is now at play, preventing higher clock speeds translating to higher- and-higher processor core devices. We think those companies that have architected their hardware and software platforms from the ground up for parallel processing are best positioned to benefit.”