Tesla-Samsung Collaboration, China’s System-Level Chip Innovation Breaks Through

The global competition in AI computing power has evolved from the minute battles over nanometer process nodes to a broader contest encompassing system innovation and ecosystem collaboration.

The global semiconductor industry landscape is undergoing a new round of profound restructuring, where geopolitical rivalries and corporate strategies are intricately intertwined.

Samsung Wins Massive $16.4 Billion Tesla AI Chip Order

Recently, South Korean tech giant Samsung Electronics disclosed in regulatory filings that it has signed a long-term foundry agreement worth 22.8 trillion won (approximately $16.4 billion) with an unnamed large multinational company for AI chips.

This news triggered a strong market reaction, with Samsung Electronics’ stock surging as much as 3.5% in early KOSPI trading, marking its largest single-day gain in nearly four weeks.

Although Samsung strictly adhered to commercial confidentiality in its official filing, revealing only the contract period from July 26, 2024, to December 31, 2033, Tesla founder Elon Musk confirmed via multiple posts on social media platform X that Tesla is indeed the client.

He specifically emphasized: “Samsung’s new fab in Taylor, Texas, will be dedicated to Tesla’s next-gen AI6 chip. This manufacturing base, just 25 miles from the Austin Gigafactory, will be a key pillar of Tesla’s AI compute strategy.”

According to technical details disclosed by Musk, Samsung is currently mass-producing AI4 chips for Tesla, while TSMC produces AI5 chips at its Hsinchu facility in Taiwan (planned to shift to its Phoenix, Arizona fab in 2026).

Samsung securing this exclusive 10-year contract for the AI6 chip signifies Tesla’s adoption of a “dual-source supply” strategy in advanced manufacturing – leveraging TSMC’s 3nm process advantages while ensuring capacity flexibility through Samsung’s Texas fab.

Industry observers noted that the 4nm Gate-All-Around (GAA) technology employed at Samsung’s Taylor fab has potential for deep synergy with Tesla’s Dojo supercomputer architecture. The fab represents a total investment of $17 billion with a planned monthly capacity of 100,000 wafers, but its ramp-up was previously delayed due to lagging key customer commitments.

This order may prompt Samsung to accelerate the installation of core equipment like EUV lithography machines, with mass production expected in Q2 2026.

Tesla faces pressure from rapid technological iteration. Its Autopilot team is developing the FSD V13 system, demanding four times the computing power of its predecessor from the in-vehicle chip.

Musk revealed that the AI6 chip will utilize a custom IP core developed by Samsung, targeting Nvidia’s Thor chip in power efficiency metrics. This necessitates deep collaboration between the two companies in areas like advanced packaging and chiplet interconnectivity.

WAIC 2025: Beyond the “Chip,” System-Level Innovation Achieves Computing Power Leapfrog

While international giants deepen collaboration in chip foundry, optimizing chip performance through innovations like advanced packaging and chiplet interconnectivity, innovations showcased by Chinese companies at the recent WAIC 2025 in Shanghai demonstrate that “beyond the chip,” system-level innovation is surpassing single-point performance, becoming the core driver of China’s AI computing power.

Take Huawei’s Ascend 384 Super Node as an example. It doesn’t merely stack chips but integrates hundreds of computing units into a single “super server” via high-speed bus technology, enabling the cluster to “work like one computer,” achieving over 2.5 times the performance of traditional clusters when training large models.

Simultaneously：

• SOPHGO’s “Shanghai Cube” debuted as a fully domestic computing cabinet.
• MetaX made its debut showcasing a full-link technology solution, designed and manufactured based on the domestic supply chain, covering a complete product line from chips, boards, servers to super nodes.
• Xfusion launched the world’s first multi-intelligence plug-and-play super cluster system, achieving an energy efficiency leap described as “one cabinet equals 32 cabinets.” A single cabinet supports 128 AI accelerator cards, employs fifth-generation 100% native liquid cooling technology, and saves over 20% energy.
• Moore Threads proposed the “AI Factory” concept, emphasizing full-stack synergy from chips and servers to intelligent computing clusters.
• StepStar, spearheading the establishment of the “Model-Chip Ecosystem Innovation Alliance,” is accelerating the formation of a fully autonomous end-to-end ecosystem spanning hardware to software, chips to models.

Against the backdrop of global geopolitics, technological roadmaps and industry maps are being redrawn. As details of the Samsung-Tesla contract emerge, this $16.4 billion commercial engagement is reshaping the power dynamics of the global AI chip foundry market.

Concurrently, the systemic innovation and ecosystem synergy within the Chinese market are becoming a crucial force in the global computing power landscape.

The global AI computing power competition has indeed ascended from the minute battles over nanometer processes to the vast arena of system innovation and ecosystem collaboration.