Samsung Is Betting $648 Billion on AI Chips Over the Next Decade

Samsung just put a number on how serious it is about the AI hardware race. The South Korean company is set to commit 1,000 trillion won, roughly $648 billion, to AI chip factories, data centres, batteries, and displays across South Korea over the next decade. That's the largest single capital commitment any tech company has made to domestic semiconductor infrastructure.

The announcement lands at a moment when South Korea is competing hard with Taiwan and the US for position in the AI supply chain. The South Korean government wants the country to be a top-three AI nation by 2030, and Samsung's investment is the industrial backbone of that ambition.

The actual problem Samsung is trying to solve

The headline number is striking, but the context matters more than the size. Samsung is fighting a two-front war in semiconductors right now. On logic chips, TSMC has a substantial lead at the leading edge: Apple, Nvidia, and AMD all manufacture their most advanced chips there rather than at Samsung. On memory, SK Hynix has pulled ahead in HBM, the high-bandwidth memory that Nvidia packages with its GPUs for AI training. Samsung has been catching up on HBM, but it spent much of 2025 fixing yield problems that held back its HBM3E supply.

Throwing $648 billion at both problems is a bet that capital can close those gaps before the AI infrastructure window narrows. It's not a guaranteed outcome. Semiconductor manufacturing doesn't scale linearly with investment, and catching TSMC on logic has been a stated Samsung goal for years. But the alternative, accepting a permanent second-place position in the market that will define computing for the next decade, is worse.

Why the AI chip race matters beyond the headline

The AI chip supply chain has a choke-point problem. Nvidia designs the GPUs that power AI training. TSMC manufactures them. SK Hynix and Samsung supply the HBM memory that gets stacked on top. A bottleneck anywhere in that chain limits how fast AI infrastructure can expand globally.

Samsung supplying competitive HBM and winning back leading-edge logic contracts would add meaningful redundancy to a supply chain that's currently very concentrated. It also gives Nvidia more leverage in negotiations with TSMC, and gives Samsung's own device division, Galaxy phones and laptops, access to the most advanced chips without depending on a competitor's fab.

For anyone building or buying AI hardware in the 2027 to 2030 window, whether Samsung's bet pays off will directly affect what's available and at what price. This isn't an abstract corporate story.

The competition context

China's government published a $295 billion AI infrastructure plan earlier this year, and the US has the CHIPS Act directing subsidies at domestic production. Samsung's announcement fits a pattern where every major tech economy is treating semiconductor sovereignty as a strategic necessity rather than a commercial preference.

The difference with Samsung is that this is private capital, not government subsidy. The company is betting its own balance sheet that AI hardware demand justifies this scale of investment over the next decade. Given where Nvidia's data centre revenue is heading, that's not an irrational bet.

As we covered when OpenAI revealed its custom Jalapeno chip plans, the AI industry's biggest players are all working to reduce their dependence on the current supply chain. Samsung's investment is a direct response to that same pressure from the manufacturing side.

The ASML export dispute with China is a useful reminder of how geopolitically sensitive semiconductor infrastructure has become. Samsung's South Korea focus is partly a bet on stable, export-compliant manufacturing capacity at a moment when supply chain politics are increasingly unpredictable.

If you're tracking the AI hardware space, also worth reading: our piece on Qualcomm's acquisition of Tenstorrent and what it signals about the chip architecture fight shaping up beneath Nvidia's dominance.