Jensen Huang and Elon Musk Join Forces... Intel Signals 'Return of the King' in 2030 [Chip Talk]

Shift from AI Training to Inference
Intel CPUs Reemerge in the Spotlight as Big Tech Makes Its Move
EMIB as Intel's Technological Gambit to Overcome CoWoS Bottlenecks

Once dismissed as a "setting sun," Intel is now signaling a resplendent comeback as the artificial intelligence (AI) inference market blossoms. As the role of control systems that organically connect devices within AI infrastructure comes to the forefront, Intel's central processing unit (CPU) is regaining its relevance. With Intel's vertically integrated capabilities spanning design, manufacturing, and packaging aligning with the United States' national ambitions for semiconductor supremacy, the industry is speculating that Intel may redeem its tarnished reputation and reclaim its position as a global semiconductor leader.

The Control Tower of the 2030 AI Inference Era

Just two years ago, Intel was facing what was widely regarded as the greatest crisis in its history, with its market capitalization falling below 100 billion dollars and even humiliating reports emerging that it could be acquired by competitor Qualcomm. Despite leading the semiconductor industry with CPU design for nearly 50 years since the late 1970s, Intel had fallen a step behind its rivals in entering the AI market, which is seen as the future growth engine. A prime example is Nvidia, whose revenue in 2021 was only about one-third that of Intel, yet has since become the world's most valuable listed company, highlighting a gap that Intel could not close.

Jensen Huang, CEO of NVIDIA. Photo by Yonhap News Agency

Now, Intel is once again disrupting the landscape of the semiconductor industry. The primary driver of this turnaround is the rapidly expanding "AI inference" market. As predictions arise that the AI inference market will surpass the AI training market in size by around 2030, Intel's CPU is being re-evaluated as the "control center" overseeing entire AI service systems. According to a Grand View Research report, the global AI inference market is estimated to be worth 97.24 billion dollars (approximately 144 trillion won) in 2024, with an expected compound annual growth rate of 17.5% through 2030.

Lip Bu Tan, CEO of Intel. Photo by Reuters and Yonhap News Agency

Lee Seungwoo, a researcher at Eugene Investment & Securities, explained, "In the early days of AI, training was paramount, so graphics processing units (GPUs) dominated everything. But as the focus shifts to inference and service, CPUs are becoming important again. Intel's Xeon line is now being redefined—not as a supplementary processor, but as the control system for AI services." He further noted that at the stage of AI services, which require more than simple calculations—such as data refinement, network management, and logical judgment—the importance of CPUs, which can intelligently manage the entire AI infrastructure, outweighs that of GPUs, which specialize in parallel processing.

In particular, the "AMX" accelerator embedded in Intel's latest Xeon processors enables ultra-high-speed matrix calculations even without a GPU, simultaneously delivering low latency and high cost efficiency for AI services where real-time response is critical, such as chatbots and autonomous driving.

Big Tech Partnerships and Support from the U.S. Government

Intel's signs of revival are also clearly reflected in its bold collaborations with global Big Tech firms. Google has recently decided to greatly expand its adoption of Intel’s Xeon CPUs while also partnering with Intel to co-develop Infrastructure Processing Units (IPUs), which will be central to next-generation data center operations. This move goes beyond mere component supply, signaling an intent to deeply integrate Intel's design assets into Google's standard infrastructure.

The shift in Intel’s relationship with Nvidia, once seen as a rival, is also noteworthy. Nvidia has announced direct equity investments and joint development projects with Intel, pivoting from sharp competition to a "partial cooperation" relationship. Jensen Huang, CEO of NVIDIA, has acknowledged that Intel's manufacturing capabilities and CPU technology are essential to expanding Nvidia’s GPU ecosystem. Furthermore, companies led by Elon Musk, such as SpaceX and Tesla, have also chosen Intel as a partner for building AI production infrastructure called "Terafab," further strengthening Intel’s position.

The industry is also paying attention to the U.S. national strategy underlying Intel’s resurgence. The U.S. government is reportedly focusing on nurturing Intel as a "national strategic asset" in order to reduce excessive dependence on TSMC and reconfigure the AI supply chain around domestic production. Previously, when Intel was driven to the brink and considered selling off its foundry division during the Trump Administration, the government declared the companies' fates intertwined by suddenly acquiring a 10% stake in Intel for 8.9 billion dollars (approximately 13 trillion won) in August last year.

Researcher Lee added, "With the U.S. government no longer leaving technological supremacy solely to the market, Intel has become the core of American infrastructure, transcending its role as a mere company. Now, Intel’s value is derived from its very existence, not just its profits. The market has already begun to bet on whether Intel can become a kingmaker again, as evidenced by its stock price."

TSMC Packaging Bottleneck: Will Intel Benefit?

The growing concern over bottlenecks in TSMC's CoWoS (Chip-on-Wafer-on-Substrate) process also presents an opportunity for Intel. As TSMC, which produces more than 90% of the world’s AI chips, struggles with capacity shortages in its CoWoS process, Intel's proprietary 2.5D packaging technology—EMIB—is drawing attention as an alternative. Instead of placing the entire chip on a precision silicon interposer, EMIB embeds small bridges only where needed within the substrate, reducing costs by approximately 30% compared to CoWoS.

An industry insider commented, "TSMC currently dominates the high-performance AI GPU market, but when it comes to custom AI chips (ASICs), Intel has a role to play. However, considering TSMC’s overwhelming experience and yield in mass production, Intel, which has only just started taking on foundry customers, will need more time to prove its reliability for large-scale orders."

The expansion of advanced packaging technologies like Intel's EMIB and the growing demand for high-performance CPUs are expected to provide strong momentum for Korean semiconductor companies as well. The increased adoption of Intel’s AI CPUs and IPUs directly translates to surging demand for high-performance DDR5 and next-generation memory. In particular, Intel’s advanced packaging processes require high-value-added substrates such as FC-BGA (flip-chip ball grid array), highlighting the anticipated performance improvement for Korea-based substrate suppliers that are major partners of Intel.

This content was produced with the assistance of AI translation services.