“CPUs are still essential, but with Nvidia dominating the AI chip market, AMD and Intel are having trouble competing,” Chang said. AMD is targeting $5 billion in AI chip sales by 2025, and Intel’s AI efforts are minimal, centered around the Gaudi platform. “Both companies will continue to invest in GPUs and AI accelerators and will see some revenue growth, but their share of the data center market will likely continue to decline.”
What will happen after Trump comes to power?
Geopolitical and economic factors, such as export restrictions, supply chain disruptions, and government policies, have the potential to reshape the semiconductor industry. President-elect Donald Trump has hinted at plans to impose high tariffs on semiconductor imports after taking office on January 20.
The CHIPS and Science Act promises billions of dollars in support to companies with semiconductor development and manufacturing operations in the United States. The bill allocates $39 billion to companies including TSMC, Intel, Samsung, and Micron. These companies are planning or already building new manufacturing or research facilities.
However, to receive support, each company must meet specific milestones. Until then, funds will not be executed. The promise of billions of dollars in aid goes a long way toward revitalizing semiconductor production in the U.S., but Morales pointed out that the Chips Act’s 25% tax cut is an even bigger benefit.
“Even companies like Intel will get a tax break of about $50 billion,” Morales said. This is unprecedented. “This is where the real benefit comes from.”
Although President-elect Trump has called government funding to encourage reshoring the wrong approach, industry experts do not expect significant cuts in CHIPS Act funding after Trump takes office. “We expect there will be some modifications to the CHIPS Act, but there will be no extreme measures to cut funds that have not yet been distributed,” Morales said. “The CHIPS Act has bipartisan support, and any attempt to change it will face pushback from states that benefit, like Arizona and Ohio.”
To date, high-performance processors that run energy-consuming cloud data centers have dominated the market, but in the future, energy-efficient AI processors for edge devices are expected to continue to receive attention.
“This year we will see PCs and smartphones that integrate AI, or wearable devices that use smaller, more precisely tuned models to perform AI inference,” Morales said. “This is the direction we are going in, and it will be a very important market in the coming years.”
He also predicted, “AI inference will play a larger role in data centers than we have seen so far.”
Transition from LLM to SLM and Edge Devices
Companies and other organizations are shifting their focus away from single AI models to multimodal AI, or LLM, which processes and integrates multiple types of data (text, images, audio, video, sensory input, etc.). Input from a variety of resources allows the model to gain a more comprehensive understanding of its data and improve performance on a variety of tasks.
Report from S&P GlobalAccording to , more than 80% of enterprises expect AI workflows to increase in the next two years. About two-thirds expect to be pressured to upgrade their IT infrastructure.
However, Sudeep Keshe, chief innovation officer at S&P Global Ratings, said that while AI is evolving toward small, task-focused models, larger, more general-purpose models will also remain essential. “These two types of models coexist, and each has its own specific domain.” “It will create opportunities,” he said.
A key challenge is developing computationally and energy efficient models, which will have a major impact on chip design and implementation. Chip manufacturers must also address scalability, interoperability, and system integration issues. These elements will drive technological advancements across industries, improve autonomous systems and enable future developments such as edge AI, Kesey added.
In particular, industry interest in AI inference is expected to increase as companies move away from cloud-based LLM and adopt SLM that can be deployed on edge devices and endpoints.
“This is an environment where abundance and famine coexist,” said IDC’s Morales. What will happen in the year ahead? Data center growth has been tremendous and will continue into 2025. What is especially exciting is that companies are starting to prioritize IT budgets toward AI. “This will create a second wave of demand for processors,” he said.
dl-itworldkorea@foundryco.com
Source: www.itworld.co.kr
