Samsung Electronics will manufacture base dies for Nvidia’s next-generation custom high-bandwidth memory using Taiwan Semiconductor Manufacturing Company’s 3-nanometer process, according to a report published by Seoul Economic Daily on October 18, 2026, and picked up by The Economic Times, Business Times Singapore, and other outlets. The arrangement, still described as a report rather than a confirmed joint statement from either company, would mark the first time Samsung and TSMC have worked together at the HBM manufacturing stage, two companies that have spent two decades as the fiercest rivals in contract chipmaking.
The product at the center of the story carries the name NVHBM, Nvidia’s custom high-bandwidth memory line that the company unveiled in August 2026. According to the Seoul Economic Daily account, Nvidia has asked all three of its major memory partners, Samsung, SK hynix, and Micron Technology, to adopt TSMC’s process for the NVHBM base die while also standardizing on sixth-generation DRAM technology for the core dies that sit above it. That is a notable ask: Nvidia is not just picking a foundry, it is trying to force a common manufacturing baseline across three companies that have historically built memory with their own proprietary processes and tooling.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What NVHBM Is and Why the Base Die Matters
To understand why a seemingly narrow manufacturing decision is generating this much coverage, it helps to understand what a base die actually does inside a high-bandwidth memory stack. HBM chips are built as vertical stacks of DRAM dies, connected through silicon vias, sitting on top of a logic layer called the base die. That base die handles the interface between the memory stack and the GPU, routing signals, managing power delivery, and in Nvidia’s NVHBM design, hosting a custom memory controller that Nvidia itself specifies.
According to the Seoul Economic Daily report, NVHBM integrates that custom memory controller directly into the base die, with the explicit goal of standardizing implementation across multiple memory suppliers. In practical terms, Nvidia wants a base die built the same way, on the same process node, whether it ships from Samsung’s fab, SK hynix’s line, or Micron‘s facility. The DRAM stacks on top, the part that actually stores data, can still come from each supplier’s own sixth-generation process, but the logic layer underneath becomes common ground. That is the piece Nvidia reportedly wants built on TSMC’s 3-nanometer process, a node TSMC already uses for high-volume logic production and one with a long track record on yield.
For readers tracking the broader GPU memory squeeze, this story connects directly to the pricing pressure already visible in the market. Tech Insider covered how DDR5 RAM prices climbed to $450 for a 32GB kit as AI demand pulled DRAM production capacity away from consumer memory, and how RTX 5090 shortages pushed street prices to as much as $8,500, five times the card’s MSRP. A standardized NVHBM base die built on a single, high-yield process is, in effect, Nvidia’s attempt to keep its own supply chain from suffering the same kind of bottleneck that has been squeezing GPU buyers all year.
Why Samsung Would Turn to Its Biggest Rival
Samsung Electronics and TSMC compete head-on in contract chip manufacturing, a rivalry that has defined the foundry business since Samsung entered advanced-node production in earnest more than a decade ago. Samsung runs its own foundry division and has spent heavily trying to close the yield and customer-trust gap with TSMC, particularly at advanced nodes where TSMC has maintained a commanding lead in both volume and defect rates. A Samsung decision to hand base-die production for a flagship Nvidia memory product to TSMC, rather than running it through Samsung Foundry, is a visible concession on a node where Samsung has wanted to prove it can compete.
According to the Seoul Economic Daily report, Samsung Electronics accepted Nvidia’s requirements after an internal review, suggesting the decision followed deliberation inside the company rather than an unconditional handoff. The report also indicates Samsung plans to keep its in-house production model intact for standard HBM products and for custom memory built for other customers, meaning this is not a wholesale retreat from Samsung’s own foundry business. It is, as described in the reporting, a specific accommodation tied to the NVHBM program and to Nvidia’s demand for a common base-die process across all three suppliers.
One detail attributed to an unnamed source in the Seoul Economic Daily account is worth flagging as unconfirmed: the report suggests Samsung’s own foundry could still produce NVHBM base dies if demand for the product exceeds what TSMC’s capacity can handle. That would effectively make Samsung Foundry a backup supplier on a product that uses Samsung’s memory business as the primary customer, an unusual structure that has not been verified by Reuters or other outlets covering the story as of this writing.
Nvidia’s Memory Supply Chain Playbook
Nvidia’s push toward a standardized base die fits a pattern the company has followed across its hardware roadmap: wherever possible, Nvidia tries to reduce the number of distinct manufacturing variables it has to manage across suppliers, because every variable is a potential bottleneck when GPU demand is running at the pace it has for the past three years. The company has already shown this instinct with GPU packaging and with its approach to HBM qualification generally, where a single quality-verification standard across suppliers cuts the testing burden Nvidia’s own engineering teams have to carry.
That pattern also shows up in how Nvidia has handled pricing and allocation with its largest cloud customers. Tech Insider’s earlier reporting on how Nvidia B200 prices doubled around a $7 billion Tencent-Oracle deal shows a company willing to let component scarcity drive price, even for its biggest buyers, rather than sacrifice margin or allocation discipline. A common NVHBM base die reduces one more point of friction in that supply chain: if Samsung, SK hynix, and Micron are all building base dies on the same TSMC node, Nvidia can, in theory, shift allocation between suppliers more easily if one falls behind, rather than being locked into whichever supplier happens to have spare capacity on its own proprietary process.
It also lands against the backdrop of scrutiny Nvidia’s financing and supply arrangements have already drawn this year. Tech Insider examined how Nvidia’s roughly $500 billion web of AI investment commitments has started to diverge from its underlying technology case, a reminder that Nvidia’s supply chain decisions are increasingly read by markets as signals about how confident the company is in sustained AI infrastructure demand, not just as engineering choices.
The Three-Way Supplier Dynamic: Samsung, SK Hynix, Micron
SK hynix has been Nvidia’s dominant HBM supplier for several generations, a position built on being first to market with each new HBM standard and maintaining tight integration with Nvidia’s GPU roadmap. Samsung and Micron have spent recent years trying to close that gap, with Samsung in particular working to requalify its HBM products with Nvidia after earlier struggles getting its chips approved for Nvidia’s data center GPUs. A standardized NVHBM base die, if it plays out as described in the Seoul Economic Daily report, changes the competitive calculus for all three.
If the base die is common across suppliers, the differentiation between Samsung, SK hynix, and Micron shifts further up the stack, toward DRAM yield, stacking density, and thermal performance on their respective sixth-generation processes, rather than toward whoever has the best base-die logic process. That could, in principle, make it easier for Samsung and Micron to compete with SK hynix on a more level playing field, since the base-die bottleneck that has historically separated winners from laggards in HBM qualification gets outsourced to a single, already-proven TSMC node rather than built in-house by each company.
| Supplier | Reported Role in NVHBM | Core Die Process | Base Die Process |
|---|---|---|---|
| Samsung Electronics | Base die producer (per report), retains in-house production for standard HBM | Samsung sixth-generation DRAM | TSMC 3-nanometer (reported) |
| SK hynix | Asked by Nvidia to adopt common base-die process | SK hynix sixth-generation DRAM | TSMC 3-nanometer (reported) |
| Micron Technology | Asked by Nvidia to adopt common base-die process | Micron sixth-generation DRAM | TSMC 3-nanometer (reported) |
| TSMC | Reported base-die foundry partner for all three suppliers | Not applicable | 3-nanometer logic process |
Note that the table above reflects what has been reported, not confirmed specifications from any of the four companies named. None of Samsung, SK hynix, Micron, Nvidia, or TSMC has issued an on-record statement confirming the arrangement as of October 18, 2026, and the underlying source for the DRAM and base-die process details is the Seoul Economic Daily report relayed through Reuters and other outlets.
Historical Context: A Rivalry That Rarely Cooperates
Samsung and TSMC’s foundry rivalry dates back to the early 2010s, when Samsung began positioning its foundry business as a credible alternative to TSMC for advanced logic manufacturing, eventually winning major customers including, at various points, portions of Apple’s chip production before Apple consolidated most of its business with TSMC. Since then, the two companies have competed almost exclusively on separate tracks: TSMC dominating leading-edge logic for Apple, AMD, and Nvidia’s GPU dies, Samsung building out its own foundry customer base while also running a separate, large memory business that has nothing to do with its foundry operations.
What makes the reported NVHBM arrangement notable is that it blurs a line that has stayed mostly intact for years: Samsung’s memory division, not its foundry division, would be the one going to TSMC for manufacturing capacity. That is a different kind of cooperation than a foundry customer simply choosing TSMC over Samsung Foundry, because it puts Samsung in the position of being both a memory supplier to Nvidia and a customer of its own chief foundry rival on the very same product. Ground.news, among the outlets covering the story, framed the arrangement as a case where the Samsung-TSMC foundry rivalry effectively gets set aside so both companies can meet Nvidia’s base-die specification together.
This is not the first time Samsung’s memory ambitions have run into complications with a major chip customer. Tech Insider has reported separately on how Tesla has pushed back on Samsung over pricing for its AI5 and AI6 chips, a dispute that shows Samsung navigating demanding specifications and pricing pressure from multiple large customers simultaneously, not just Nvidia. The NVHBM base-die arrangement adds another high-stakes customer relationship to that list, one where Samsung appears to have accepted Nvidia’s terms rather than push back, according to the reporting.
Market Impact: Reading the Stock and Supply Signals
The report lands at a moment when memory stocks and AI infrastructure names have been trading on every signal about HBM supply and demand. TSMC’s own quarterly results have already shown the scale of the AI buildout flowing through its business; Tech Insider covered how TSMC’s profit jumped 70% as the AI boom ran longer than many analysts expected, a figure that underscores why winning even a slice of HBM base-die production, a market TSMC has not historically served directly, would matter to the company’s growth trajectory even if base dies themselves are a smaller, lower-margin product than TSMC’s leading logic business.
For Samsung, the market reaction will likely hinge on whether investors read this as Samsung ceding ground to TSMC or as Samsung securing its place in Nvidia’s supply chain by any means necessary. Memory companies have spent much of 2026 fighting for a larger share of Nvidia’s HBM business, and a report suggesting Samsung accepted Nvidia’s TSMC requirement after internal review, rather than resisting it, could be read either as pragmatic positioning or as a concession that favors TSMC’s leverage over Samsung’s own foundry ambitions.
SK hynix and Micron face a related question: if all three suppliers end up using the same TSMC base die, competition among them compresses into DRAM yield and stacking efficiency, areas where SK hynix has generally held an edge in recent HBM generations. Whether that edge persists once the base die is no longer a point of differentiation is one of the more interesting open questions this report raises for the memory market heading into 2027.
Competitive Comparison: How the Big Three Stack Up on HBM
| Factor | SK hynix | Samsung Electronics | Micron Technology |
|---|---|---|---|
| Historical Nvidia HBM position | Lead supplier across recent HBM generations | Working to close qualification gap with Nvidia | Smaller share, growing presence |
| Foundry relationship with TSMC | Not a TSMC foundry rival | Direct foundry competitor to TSMC | Not a TSMC foundry rival |
| Reported NVHBM base-die process | TSMC 3nm (per report) | TSMC 3nm (per report) | TSMC 3nm (per report) |
| In-house base-die alternative | No comparable foundry arm | Samsung Foundry cited as possible backup if TSMC capacity is exceeded (unconfirmed) | No comparable foundry arm |
The comparison above highlights why Samsung’s position is structurally different from SK hynix and Micron’s. Neither SK hynix nor Micron operates a leading-edge logic foundry business, so turning to TSMC for base-die production carries no symbolic weight for them the way it does for Samsung. For Samsung, the decision touches both its memory division’s relationship with Nvidia and its foundry division’s standing as a TSMC alternative, which is likely why the Seoul Economic Daily report specifically notes that Samsung’s acceptance followed an internal review rather than a routine supplier decision.
What Remains Unconfirmed
Several details tied to this story deserve caution before being treated as settled fact. A separate Seoul Economic Daily report has suggested NVHBM could offer substantially more memory bandwidth and lower power consumption than standard HBM4E, with specific percentage figures circulating in some coverage. Those figures were not confirmed in the Reuters account of the story and should not be treated as verified specifications until Nvidia, Samsung, SK hynix, or Micron confirm them directly. Similarly, the claim that Samsung Foundry could step in as a backup base-die producer if TSMC capacity runs short is attributed only to an unnamed source and has not been independently verified.
No pricing, shipment timeline, or production-start date for NVHBM has been confirmed in the reporting reviewed for this article, including the account carried by Capital.com. Readers should also note that none of the five companies central to this story, Nvidia, Samsung, TSMC, SK hynix, or Micron, has issued a public statement confirming the arrangement as described by Seoul Economic Daily, which remains the primary and original source for the report as of October 18, 2026.
Predictions: Where This Goes From Here
- Expect Nvidia, Samsung, SK hynix, Micron, or TSMC to eventually issue at least a partial confirmation or denial of the Seoul Economic Daily report within the next earnings cycle, given how much market attention HBM supply chain details have drawn throughout 2026.
- If the standardized base-die approach holds, look for SK hynix’s competitive advantage in HBM to narrow relative to Samsung and Micron over the next one to two product generations, since the differentiation shifts away from base-die logic toward DRAM stacking.
- TSMC’s role as a base-die foundry for memory companies, a business it has not historically served in this way, could become a new growth line item that analysts start tracking separately from TSMC’s core logic and GPU-die business.
- Samsung Foundry’s reported fallback role, if confirmed, would suggest Samsung is hedging against TSMC capacity constraints rather than fully outsourcing NVHBM base-die production, which would be consistent with Samsung’s broader strategy of maintaining foundry independence even while cooperating with TSMC on specific products.
- Expect continued price pressure on HBM and DRAM generally through at least mid-2027, as the NVHBM standardization effort is unlikely to ease the underlying capacity crunch driving prices for products like DDR5 and consumer GPUs higher in the near term.
Why This Matters Beyond the Memory Industry
For anyone tracking the broader AI hardware buildout, this report is a reminder that the bottlenecks shaping the next generation of AI infrastructure are not just about GPU compute dies, they are increasingly about the memory that feeds those GPUs. Nvidia’s willingness to standardize a critical manufacturing step across three competing suppliers, and to lean on TSMC, a company with no prior track record in HBM base-die production, to do it, signals how tightly Nvidia believes it needs to control its supply chain to keep pace with AI demand through 2027 and beyond.
It also raises a longer-term structural question for the foundry industry: if TSMC can successfully produce HBM base dies at scale, does that open the door to TSMC playing a larger role in memory manufacturing generally, a business it has largely stayed out of for decades in favor of letting Samsung, SK hynix, and Micron own memory while TSMC owns logic? A confirmed, successful NVHBM rollout could be the first real test of whether that division of labor still holds in an AI-driven chip market where the lines between logic and memory manufacturing are blurring faster than at any point in the industry’s history.
Frequently Asked Questions
What is NVHBM?
NVHBM is Nvidia’s custom high-bandwidth memory line, unveiled in August 2026, that integrates Nvidia’s own memory controller directly into the base die beneath the DRAM stack, with the goal of standardizing implementation across Nvidia’s memory suppliers.
Will Samsung stop making its own HBM chips?
No. According to the Seoul Economic Daily report, Samsung plans to retain its in-house production model for standard HBM products and for custom memory it builds for other customers. The TSMC arrangement is reported to be specific to NVHBM base-die production.
Why would Nvidia want three different memory suppliers using the same base-die process?
A common base-die process reduces the number of separate manufacturing and quality-verification standards Nvidia has to manage across suppliers, which can simplify qualification and make it easier to shift allocation between Samsung, SK hynix, and Micron if one supplier falls behind on capacity.
Has Nvidia or Samsung officially confirmed this report?
No. As of October 18, 2026, the arrangement is sourced to a Seoul Economic Daily report, relayed by Reuters and other outlets. Neither Nvidia, Samsung, TSMC, SK hynix, nor Micron has issued a public statement confirming the details.
Is this the first time Samsung and TSMC have worked together?
According to the report, this would be the first time Samsung and TSMC have partnered specifically at the HBM manufacturing stage. The two companies remain direct competitors in the broader foundry business.
Does this affect current HBM or GPU prices?
No immediate pricing impact has been confirmed. The report concerns next-generation NVHBM production, not current-generation HBM4E supply, which continues to face the capacity constraints that have pushed DRAM and GPU prices higher throughout 2026.
What is sixth-generation DRAM technology, and how does it relate to NVHBM?
Sixth-generation DRAM refers to the memory technology each supplier, Samsung, SK hynix, and Micron, would reportedly use for the core dies stacked above the common TSMC-built base die in NVHBM modules. Each company’s own sixth-generation DRAM process remains distinct, even if the base die underneath is standardized.
Could Samsung Foundry still produce NVHBM base dies?
An unnamed source cited in the Seoul Economic Daily report suggested Samsung Foundry could produce NVHBM base dies if demand exceeds TSMC’s available capacity. This detail has not been independently confirmed and should be treated as unverified.
