The artificial intelligence boom has lifted chipmakers including Nvidia, Micron and South Korea’s SK Hynix.
Now Intel, which has begun rebounding from a lengthy slump, is benefiting from a new shift in A.I. spending. Tech firms are not only buying A.I. chips known as graphics processing units, made by companies like Nvidia, but also increasingly purchasing microprocessors known as central processing units — which are exactly what Intel makes.
The bonanza was evident on Thursday when Intel reported its financial results. The Silicon Valley company said its revenue had risen 25 percent to $16.1 billion in its latest quarter, driven by nearly 60 percent growth in its data center segment, which includes the chips that power A.I. applications and cloud services.
The results and a projection for further growth were well above what Wall Street had expected, pushing Intel’s stock up nearly 9 percent at one point in after-hours trading.
“A.I. is driving unprecedented demand for compute,” Lip-Bu Tan, Intel’s chief executive, said in a statement, adding that the revenue growth was the strongest in 15 years.
Intel is still losing money from layoff costs and related restructuring charges that Mr. Tan ordered after taking the top job last year. The company lost $11 billion in the latest quarter, more than the $2.9 billion loss from a year earlier. The wider loss stemmed from a revaluation of shares held in escrow in connection with an $8.9 billion investment that the U.S. government made in Intel last year.
But Intel’s upward trajectory has been unmistakable, and its stock quadrupled between Jan. 1 and the end of June.
Much of this has been powered by the change in A.I. spending. Earlier in the A.I. boom, buyers wanted Nvidia’s GPU chips, which can do many simpler chores simultaneously. Central processing units, or CPUs, served a role in A.I. servers but might be outnumbered at least four to one by GPUs.
More recently, A.I. products called “agents,” which can “reason” and act on information, have become more popular. These programs require a greater proportion of microprocessors to run. As a result, research firms such as Creative Strategies have predicted that A.I. data centers will increasingly buy roughly the same number of CPUs and GPUs.
“It is not just a GPU game anymore,” Santosh Janardhan, Meta’s head of infrastructure and co-head of engineering, said recently. “CPUs are becoming at least as important, if not more.”
This shift benefits not only Intel but its rival Advanced Micro Devices, which has steadily taken market share in data center servers and successfully developed GPUs for A.I. jobs. On Thursday, AMD unveiled more powerful CPUs and GPUs, as well as new “rack-scale” systems — complete computing boxes taller than a refrigerator — that integrate CPUs, GPUs, networking, data storage and other technology.
Anthropic, a leading A.I. start-up, has also committed to buying the new hardware from AMD — which estimated the value of the deal at tens of billions of dollars — and will receive an investment of up to $5 billion from the chipmaker. OpenAI, Anthropic’s rival, also said it would use AMD’s new rack-scale system.
Nvidia, meanwhile, has pushed its own microprocessors for A.I., including a new model called Vera. Huge data center operators like Amazon, Microsoft and Google have also developed custom microprocessors.
Intel still retains some unique assets, including its own network of factories. It competes with Taiwan Semiconductor Manufacturing Company, the biggest maker of advanced chips, in both building chips and the increasingly important task of packaging them to work together.
That so-called foundry service has recently landed customers such as Apple, which had been relying almost entirely on TSMC.
Intel said growth in sales of its server chips over the last year had been the strongest on record, with its latest version, Xeon 6, one of its fastest-selling products ever.
Though demand for such data center products is strong, Intel said this week that it had trimmed some employees in that business to improve efficiency. It did not disclose numbers.
The company added that the productivity of its latest manufacturing technology was improving and that its foundry business had grown 31 percent to $5.8 billion, though the unit posted an operating loss of $2.1 billion.
The post Intel Benefits From a New Shift in A.I. Spending appeared first on New York Times.
