The semiconductor industry sees artificial intelligence as a top revenue driver in 2024, according to 172 chip executives surveyed by KPMG and the Global Semiconductor Alliance in the fourth quarter. Most of those surveyed work at companies with more than $1 billion in annual revenue
The survey results may not surprise anybody, given the massive impact of generative AI on companies, including enterprises using AI training and inference software running on AI chips in data centers in corporate headquarters and at cloud providers.
The KPMG 2023 survey found AI jumped to the second most important revenue driver for the industry, after placing fourth in the prior two annual surveys. Automotive still finished as the top revenue driver in the survey, its second year in a row.
“Automotive is still the top revenue driver, but AI has surged up the list and is the headline in many ways,” said Mark Gibson, KPMG Global and US technology, media and telecommunications leader, in an interview.
“If you think of AI running on a mobile phone or a car, it’s going to be embedded into the next series of chips,” he added.
Indeed, market leader Intel, Nvidia, Qualcomm, AMD and others are all aboard the AI chip bandwagon. Nvidia holds the largest GPU share by far for use in data centers, but Intel just launched its 5th Gen Xeon for servers alongside a Core Ultra semiconductor for mobile, including the new AI PC concept.
Intel joins others in the AI PC chip arena, including AMD, Qualcomm and Apple.
RELATED: The AI PC race is on. Is Intel winning with Core Ultra?
RELATED: Intel dives deeper into AI with new mobile and server processors
Semiconductor companies also ranked Ai and Gen AI among the top three strategic priorities over the next three years, trailing talent development and retention and supply chain resiliency. Those three priorities ranked ahead of mergers and acquisition and even cybersecurity. Also down on the list were company plans to participate in government subsidies like the CHIPS Act and environmental, social and governance initiatives.
“The cost of attracting and retaining employees will continue to go up and drive costs,” Gibson said. “Almost certainly costs will go up, even after inflation stabilizes.”
The survey reflected some conflicting views on expectations for a chip inventory surplus in 2024, which has dogged much of the industry for more than a year. The survey found 30% believe there is already an excess of semiconductor inventory. Conversely, the survey found that emerging technologies like AI will provide continued growth with 19% of respondents saying they believe chip demand will keep increasing with no inventory excess in the next four years. Excess inventory developed in 2022 and 2023 when companies stopped buying as many chips for PCs, especially, but for other products as well.
Gibson predicted a “much more stable supply chain” in 2024 with more ability to manage inventory.
He also said in the US that the CHIPS Act grants will help a rebound in the tech sector overall. Also, some tech companies that were laying off workers last summer are now hiring workers.
“The state of the economy will drive this [semiconductor] sector,” he said. “All of us are living with more and more things that require chips and don’t even realize it.”
Energy consumption concerns
As AI demands grow and more companies buy up AI chips, there will undoubtedly be a surge in demand for electricity to run data centers. Gibson said it is a “great question” of how critical the demand will be in 2024 and beyond, but also how serious the overall electronics market is going to be about going green and placing reliance on renewable energy sources.
All the major AI chipmakers advertise energy savings with their latest products, but AI chip purchases can certainly scale up well beyond the amount of energy saved. Data center growth is continuing in the US, even as some major cities like Amsterdam have checked such growth, denying permits to add extra electricity.
OpenAI’s ChatGPT 3.5 exploded on the scene a year ago, but some economists and analysts foresee a 30-year impact. “Remember, we are just in the early innings of a long game,” said Karl Freund, founder and principal analyst at Cambrian-AI, in an interview.
“Companies are just getting on board with deploying AI to improve productivity, deliver better server and lower costs. The attitude now is “get it working, then we can worry about lowering energy and other costs,” he added. “Consequently, while AI is causing a significant increase in data center energy consumption, this legitimate cause for concern is often overshadowed by the business imperatives of getting AI up and running as soon as possible with enough performance to meet business needs. Energy will become more critical to decision=makers in the long term, especially as data centers confront power scarcity.”
The big GPUs made by the three major producers consume more energy per rack, Freund noted, but they also deliver far more performance. “Efficiency can actually go up when you use a new Nvidia H200 over an older GPU for a given workload. And software advances for AI will drive down both capital and operating costs.”