The Collaboration No One Expected
Samsung Electronics will use TSMC's 3-nanometre process to build base dies for Nvidia's next-generation high-bandwidth memory chips, the Seoul Economic Daily reported on Sunday. It's the first time Samsung has worked with the Taiwanese foundry at the HBM manufacturing stage, a shift that rewrites how the industry's biggest rivals share production.
Nvidia asked Samsung, SK Hynix and Micron to adopt TSMC's process for NVHBM base dies, plus sixth-generation DRAM for core dies, according to industry sources cited by the paper. NVHBM, unveiled in August, embeds Nvidia's custom memory controller into the base die to standardise implementation across suppliers. The original contract meant three memory makers designing separately, shipping from different fabs, under different process controls. NVHBM changes that by tying them to one controller and one node.
Samsung accepted the requirement after an internal review, but it plans to keep its in-house model for standard HBM and custom products for other customers. The Seoul Economic Daily quoted an unnamed source saying Samsung is also assessing whether its own foundry could produce NVHBM base dies if demand exceeds TSMC's capacity. That assessment is still underway, not settled.
The move is not a merger of rivals. It's a supplier-customer arrangement forced by customer demand. Nvidia needs consistency, and TSMC delivers it at 3nm. Samsung doesn't want to lose Nvidia's orders, so it's buying the process it doesn't have at that node for this specific product. At the same time, Samsung is betting its own 2nm roadmap — which has been catching up for two years — may not close the gap fast enough to matter before Nvidia moves on.
Why This Matters Now
The deal lands at a strange moment for semiconductor stocks. Analysts at Bloomberg note the longest chip selloff since 2006 has minted short sellers $32 billion, and AI-linked sales have softened faster than expected. Selling is not over. TSMC's earnings must calm a jittery market that has stopped betting blindly on AI buildouts.
The GPU factory shortage story matters because it sets a ceiling. If TSMC has spare capacity for Nvidia even after Q4 orders slow, then Q1 results won't crater prices. If the slowdown continues, then every expansion story loses its signal and investors retrench to margin plays alone. Nobody knows yet where the bottom is.
Taiwan's government is making its own moves to match Washington. According to Bloomberg, Minister of Economic Affairs Kung Ming-hsin said Taipei will add AI servers and cutting-edge semiconductors to a controlled-items list, making it a criminal offence under the Foreign Trade Act to ship restricted AI hardware to China. The measure is part of a broader US trade alignment, not a standalone policy.
These two developments — Samsung turning to TSMC for HBM, and Taiwan criminalising chip exports to China — show the same underlying force: the supply chain is no longer free to optimise for cost alone. Politics and customer demands now dictate which wafers go where.
The Technical Shift Behind the Deal
Samsung isn't outsourcing everything. It retains its own foundry for standard HBM and for other customers. According to the Seoul Economic Daily, Samsung is also assessing whether its in-house foundry could produce NVHBM base dies if TSMC's capacity is exceeded. That's a hedge, not a plan.
The 3nm process is the critical detail. TSMC dominates that node. Samsung has been catching up with its 2nm gate-all-around design, but for this specific Nvidia contract, the Taiwanese process is already qualified. Time to qualification beats time to build a new line. Qualification cycles run months; building a fab runs years.
It's not clear this will hold. If Samsung's own 2nm yield improves quickly, it could reclaim production. For now, the company is accepting dependency on its closest competitor for a product that generates billions in revenue. Samsung declined to comment for this report, but the company publicly announced its 2nm roadmap last quarter, which makes the direction clear.
The Seoul Economic Daily reported the deal based on sources familiar with Nvidia's procurement planning, not statements from Nvidia or Samsung. That sourcing means the information was gathered inside these companies rather than leaked externally.
Export Controls Reshape the Maps
Taiwan's new rules target AI servers and advanced chips specifically, not all electronics. That distinction matters. A server with restricted GPUs is a different risk profile than a consumer laptop or a TV set. By making it a criminal offence, Taipei raises the cost of smuggling well beyond customs fines, pushing failures from a paperwork problem into a prison sentence.
The timing ties to the Trump administration's push to close loopholes in the original chip embargo. US officials have raised concerns that Chinese firms acquire restricted hardware through third-party routing, and Taiwan's rules make that harder by adding domestic criminal liability alongside the existing foreign trade measures.
Reuters confirmed the broader story through its desk coverage, noting that similar restrictions had already been proposed for photovoltaic manufacturing technologies earlier in October. China's proposal covers twelve technologies used to make solar equipment, citing national security concerns, as part of a revision that also touches AI and biotechnology.
Both policies work in the same direction, even if their wording differs. One comes from Washington, the other from Taipei. Neither treats chips as commerce; both treat them as national instruments.
What It Means for Indonesia
Indonesia's own semiconductor supply chain is thin. The country relies on imports for chips used in telecommunications, automotive and industrial automation. Joko Widodo's administration has courted foreign direct investment in the sector, but domestic capacity remains limited to a few niche players and simple packaging lines.
Any tightening of Taiwan-China routes could redirect supply, not reduce it. If TSMC and Samsung coordinate around Nvidia's specs, that coordination reaches every assembler that buys memory in Southeast Asia. A Sinar Group, an Infatek unit, or a PT Tirip steel mill all depend on shipping lanes that cross Taiwan's new control zone.
It's not a crisis for Jakarta, but it is a reminder that chip choices are made in Hsinchu and Seoul, not Jakarta. That won't change with this deal, but the deal makes the dependency more visible. Indonesia sits in the middle: closer to Shenzhen than to either Hsinchu or Seoul, but dependent on routes that neither ends of the island system controls directly.
The Forward Look
Samsung's internal review accepted Nvidia's requirements. That's rare for a company with its own foundry ambitions. The next few quarters will show whether Samsung treats TSMC's 3nm process as a bridge or a permanent lane. Samsung's 2nm yield progress will matter most here, not theoretical process advantages, because what counts is what arrives on the wafer.
Meanwhile, Taiwan's criminal-export rule takes effect after the measures are finalised, and its impact will depend on enforcement. A closed border or passport check that stops legitimate exporters isn't something any policy paper can guarantee. The actual friction depends on how aggressively prosecutors pursue the offence once it becomes a crime.
The chip market doesn't reward predictions; it rewards capacity that arrives on time. For this round, TSMC has the capacity, Nvidia has the order, and Samsung has a reason to say yes. Indonesia's take is structural, not operational — its supply chain simply wasn't built to be sovereign in anything except legacy niches. Read the desk coverage for more on export controls and fabrication. Further reading: Samsung's 2nm road map announcements and related semiconductor intelligence roundups.