Samsung to Invest US$73bn in AI Chip Development in 2026
AI / Machine Learning | 5 min read
Samsung Electronics has announced plans to invest more than 110 trillion won (approximately US$73.2 billion) in capital expenditure and research throughout 2026 — the largest single-year semiconductor investment by any company in history. The 22% increase over 2025 spending signals a decisive attempt to recapture leadership in the AI chip market, where Samsung has been losing ground to rivals SK Hynix in high-bandwidth memory (HBM) and TSMC in advanced foundry manufacturing. Samsung's CEO and Vice Chairman Jun Young-hyun identified agentic AI as the primary driver behind the increased spending, citing surging demand for AI servers, data storage systems, and next-generation memory.
The Investment in Context: Outpacing TSMC and Reversing the Tide
To put the scale in perspective: TSMC has projected 2026 capital spending of US$52–56 billion, focused primarily on advanced foundry manufacturing. Samsung's US$73.2 billion outlay spans memory production, foundry manufacturing, advanced packaging infrastructure, and R&D — making it a broader and larger commitment than any competitor this cycle. The investment also comes at a moment of competitive pressure: Samsung's foundry market share fell to just 7.2% in 2025 while TSMC commands approximately 69.9%, and SK Hynix has dominated HBM supply for leading AI chipmakers including Nvidia.
The 2025 performance provides further context. Samsung recorded annual revenue of approximately US$230 billion with an operating profit of around US$30 billion — one of its strongest financial performances in recent years, driven by the rebound in AI-related semiconductor demand.
HBM4 and the Memory Battle Against SK Hynix
The centrepiece of Samsung's AI memory strategy is HBM4 (High-Bandwidth Memory 4) — the next-generation memory standard that AI accelerators increasingly depend on. Samsung became the first to commercially ship HBM4 chips earlier in 2026, and debuted its HBM4E roadmap at Nvidia GTC 2026, receiving a public endorsement from Nvidia CEO Jensen Huang. Critically, Samsung has already sold out its entire 2026 HBM4 production capacity. Samsung's HBM4 solutions also target Nvidia's Vera Rubin platform — keeping it closely linked to future accelerator demand cycles. Mark Lipacis, semiconductor analyst at Jefferies, called HBM4 qualification a "genuine reset" for the company, noting that strong performance at scale could allow Samsung to recapture significant share from SK Hynix in the next-generation memory cycle.
Strategic Customer Wins: Nvidia, AMD, and Tesla
The investment is already generating strategic momentum. Samsung has secured a deal to produce Nvidia's Groq 3 processors using its advanced 4-nanometer process, reducing Nvidia's complete dependence on TSMC for certain chips. The company has also established a partnership to supply HBM4 chips to AMD for future AI memory requirements. In the automotive sector, Samsung has agreed to begin mass production of AI chips for Tesla — targeted for autonomous driving systems and robotics — by the second half of 2027. The company is simultaneously moving toward multi-year supply agreements (3–5 years) with large customers to stabilise revenue across the cycle.
Manufacturing Infrastructure: Pyeongtaek and Yongin
A significant portion of the US$73bn is being channelled into physical manufacturing scale-up. At the Pyeongtaek Campus, Samsung is accelerating efficiency improvements at its P4 plant and pushing ahead with the P5 production line. At Yongin, the company is building new manufacturing facilities as part of what is expected to become the world's largest semiconductor cluster. In the US, Samsung's Taylor, Texas campus is advancing its most sophisticated process nodes — including SF2 and SF3P — positioning it as a critical node in Samsung's global manufacturing network, with Fab 2 construction potentially beginning by late 2026.
Beyond Semiconductors: A Cross-Industry AI Ecosystem
Samsung's strategy extends beyond chip manufacturing. The company is pursuing acquisitions and partnerships across robotics, medical technology, automotive electronics, and AI data centre infrastructure — positioning its semiconductors as the connective tissue of a broader intelligent systems ecosystem. South Korea has reinforced this direction with a US$23 billion national chip industry support package, reflecting how semiconductor production has become a matter of economic stability and national security. Samsung plans to pay out 9.8 trillion won in regular dividends in 2026, signalling financial confidence despite the record capital outlay.
Key Takeaways
- • Samsung has committed US$73.2bn (₩110 trillion) to AI chip development in 2026 — a 22% increase over 2025 and the largest single-year semiconductor investment in history, surpassing TSMC's projected US$52–56bn capex.
- • Samsung's HBM4 production capacity for 2026 is already fully sold out. The company debuted HBM4E at Nvidia GTC 2026 with a public endorsement from Nvidia CEO Jensen Huang.
- • Key customer wins include Nvidia (Groq 3 processors on 4nm), AMD (HBM4 supply partnership), and Tesla (AI chip mass production for autonomous driving and robotics by H2 2027).
- • Physical expansion targets include Pyeongtaek P4/P5, Yongin (world's largest planned semiconductor cluster), and the Taylor, Texas campus (SF2/SF3P advanced nodes, Fab 2 from late 2026).
- • Samsung's CEO cited agentic AI as the primary demand driver, with the strategy spanning memory, foundry, advanced packaging, robotics, medical tech, automotive electronics, and AI data centre infrastructure.