Market chatter on October 26, 2026 puts a large number on a small, unglamorous part of the AI supply chain. Reports carried by Moomoo, Finimize and marketscreener.com say Tesla and NVIDIA will together put a reported KRW3.7 trillion (about US$2.8 billion) into advanced chip-substrate facilities at Samsung Electro-Mechanics. The split, as reported, is KRW1.1 trillion from Tesla and KRW2.6 trillion from NVIDIA.
The product at the center is the flip-chip ball grid array substrate, known as FC-BGA, which sits under AI semiconductors and links the silicon to the board. It rarely makes headlines. Yet Korean chip stocks have been trading on every AI supply signal for weeks, as our Kospi market open coverage showed, and a deal of this size lands in a market already primed to react.
This analysis separates what is reported from what is confirmed, walks through the numbers and the timeline, and explains why a substrate order matters to anyone building or buying AI compute. One caution up front: the cited reports do not show a complete public investment filing that confirms each company’s full amount, so treat the headline figures as industry-sourced.
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 the KRW3.7 Trillion Report Actually Says
The core claim is simple. Samsung Electro-Mechanics is reported to receive a combined KRW3.7 trillion from two of the biggest names in AI hardware, earmarked for FC-BGA substrate capacity aimed at AI semiconductors. The reports break the money into KRW1.1 trillion from Tesla and KRW2.6 trillion from NVIDIA, which adds up exactly to the headline total.
Two details separate this from a plain capital-expenditure story. First, Tesla reportedly signed a long-term FC-BGA supply agreement with Samsung Electro-Mechanics, with a dedicated production line at the company’s Sejong Plant in South Korea. Second, Samsung Electro-Mechanics and NVIDIA officially announced a joint investment in expanded dedicated manufacturing facilities for ABF package substrates used in AI semiconductors. That second item is an official statement. The dollar split behind it is not something the announcement spells out in the material we reviewed.
Also, the reports put the total funding requirement for the Samsung Electro-Mechanics FC-BGA expansion at KRW4.27 trillion. That means the Tesla and NVIDIA money, if it materializes as reported, would cover most but not all of the build.
The Numbers at a Glance
The table below collects every figure from the reports and adds two simple calculations so the scale is easier to read. The derived rows are our arithmetic, not reported numbers.
| Item | Figure | Status |
|---|---|---|
| Combined Tesla and NVIDIA investment | KRW3.7 trillion (US$2.8 billion) | Reported, industry sources |
| Tesla allocation | KRW1.1 trillion | Reported |
| NVIDIA allocation | KRW2.6 trillion | Reported |
| Total FC-BGA expansion funding requirement | KRW4.27 trillion | Reported |
| Customer money as a share of requirement | About 87% | Derived (3.7 / 4.27) |
| Tesla share of customer money | About 30% | Derived (1.1 / 3.7) |
| NVIDIA share of customer money | About 70% | Derived (2.6 / 3.7) |
The takeaway from the arithmetic: NVIDIA carries roughly seven of every ten won in the reported pool, and the two customers together would fund close to nine-tenths of the stated requirement. That is an unusually high customer share for a component maker’s own capacity build.
Why FC-BGA Substrates Matter for AI Chips
A modern AI accelerator is not one slab of silicon. It is a package of compute dies, memory stacks and interconnect, all of which must be wired to the outside world. The substrate does that job. It routes power and signals between the chip and the circuit board, and it has to stay flat and stable under heat that would warp lesser materials.
As AI packages grow larger and denser, the substrates under them grow more demanding. More layers, finer wiring and tighter tolerances all push yields down and raise the skill needed to build them. That is why a substrate line is slow and expensive to add. You cannot flip a switch and double output.
For buyers such as NVIDIA, this creates a quiet bottleneck. A company can secure wafers and memory and still ship fewer accelerators than demand calls for if the packaging layer runs short. Financing dedicated substrate capacity is a way to turn that risk into a contract. We have tracked the same pressure from the memory side in our look at Micron’s $54.23B quarter and what it signals for Korean chips.
NVIDIA’s KRW2.6 Trillion: Sejong and Busan
NVIDIA’s reported expansion involves two Samsung Electro-Mechanics sites, the Sejong and Busan plants. The company’s tie-up with Samsung Electro-Mechanics is the better documented of the two deals, because the joint investment in dedicated ABF package substrate facilities was announced officially rather than only reported.
The scale matters for the market narrative. At KRW2.6 trillion, NVIDIA’s reported share is more than double Tesla’s. That fits the broader picture of a company shipping accelerators at a pace that makes any single missing part expensive. Our analysis of NVIDIA trading under 17x forward earnings covers how investors weigh that demand against the threat from custom chips.
One more point. Funding a supplier’s plant is not the same as buying its output. The reports describe money going into facilities. They do not publish NVIDIA’s purchase volumes, pricing or delivery commitments, and we will not guess at them.
Tesla’s KRW1.1 Trillion and the 2029-2033 Supply Window
Tesla’s piece is smaller but more unusual. The reports say Tesla signed a long-term FC-BGA supply agreement with Samsung Electro-Mechanics, with a dedicated line at the Sejong Plant, and that the agreement period runs from 2029 to 2033. That is a five-year window that opens after the Sejong expansion is scheduled to reach mass production.
A dedicated line is the key phrase. It suggests capacity reserved for one customer rather than shared across a general pool. That arrangement protects the buyer from allocation fights in a tight market, and it gives the supplier a predictable base load to justify the build.
What the reports do not establish is what Tesla will use the substrates for. Commentary has pointed toward autonomous-driving vehicles and the Optimus humanoid robot, but that link is unconfirmed unless an official company filing or statement says so. Readers following the robot side of the story can check our piece on Optimus production at 20,000 a week and Tesla’s 80% value claim for the claims that are on the record.
Production Timeline: End-2027 to Late 2028
Substrate plants take time. The reported schedule for the Sejong expansion puts facility construction completion at the end of 2027, then equipment installation and trial operations in the first half of 2028, with mass production following in the second half of 2028. Tesla’s supply period then starts in 2029.
| Milestone | Reported timing | What happens |
|---|---|---|
| Facility construction completion | End of 2027 | Sejong building shell and utilities finished |
| Equipment installation and trial operations | First half of 2028 | Tools installed, trial runs begin |
| Mass production | Second half of 2028 | Volume FC-BGA output starts |
| Tesla supply agreement window | 2029 to 2033 | Long-term supply from the dedicated line |
Read the dates plainly. No new capacity from this program reaches the market in 2026 or 2027. Anyone expecting the deal to ease today’s packaging tightness will wait at least two years. What it changes is the supply picture for the back half of the decade.
What Is Confirmed and What Is Not
Good reporting on a story like this needs a clear line between evidence and rumor. Here is where each piece stands.
- Officially announced: Samsung Electro-Mechanics and NVIDIA joint investment in expanded dedicated manufacturing facilities for ABF package substrates used in AI semiconductors.
- Reported by industry sources: the KRW3.7 trillion combined total, the KRW1.1 trillion and KRW2.6 trillion split, the KRW4.27 trillion funding requirement, and the Sejong and Busan expansion details.
- Reported, not independently documented: Tesla’s long-term FC-BGA supply agreement, the dedicated Sejong line and the 2029 to 2033 term.
- Unconfirmed: that both companies will invest the full reported amounts, since the cited reports do not show a complete public investment filing for each figure.
- Unconfirmed: specific end uses such as Tesla’s autonomous-driving vehicles and the Optimus robot.
- No quotes on record: the material contains no qualifying direct quote from Elon Musk, Jensen Huang or a named Samsung Electro-Mechanics executive.
That last point shapes how this article reads. Without on-record statements, we describe what outlets reported and what the companies announced, and we leave motive and intent to the reader’s judgment.
Market Impact for Samsung Electro-Mechanics and Korean Chips
For Samsung Electro-Mechanics, the reported money does three things. It funds capacity the company would otherwise have to carry on its own balance sheet. It locks in demand from two high-profile customers before the plant exists. And it signals to the market that substrates, long treated as a cyclical commodity corner of electronics, are becoming a strategic input for AI.
The wider Korean tech trade tends to move as a group on news like this, because memory makers, foundry names and component suppliers share the same AI demand story. A substrate deal does not directly raise memory output, but it removes one worry about whether finished accelerators can ship. We do not have verified price moves tied to this report, so we make no claim about share-price reaction.
Investors will watch for a formal disclosure. Korean-listed companies normally file material investment and supply agreements, and a filing would settle the open question about the full amounts. Until one appears, the market is pricing chatter.
Competitive Comparison: The Two Deals Side by Side
The Tesla and NVIDIA arrangements share a supplier and a product but differ in structure. The table lays out the differences using only reported or announced details.
| Factor | Tesla | NVIDIA |
|---|---|---|
| Reported amount | KRW1.1 trillion | KRW2.6 trillion |
| Share of KRW3.7 trillion | About 30% | About 70% |
| Reported sites | Sejong Plant (dedicated line) | Sejong and Busan plants |
| Deal form | Long-term FC-BGA supply agreement | Joint investment in dedicated ABF substrate facilities |
| Term | 2029 to 2033 | Not stated in the reviewed material |
| Official announcement | None identified | Yes, joint investment announced |
| End product | Unconfirmed | AI semiconductors |
The pattern is clear. NVIDIA’s deal is larger, officially acknowledged, and aimed squarely at AI semiconductors. Tesla’s is smaller, rests on reports, and carries a firm multi-year term. Both buyers want the same thing, which is guaranteed access to scarce packaging capacity.
Historical Context: From Shortage to Strategic Asset
Packaging used to sit at the back of the chip story. During the pandemic-era supply crunch of 2021 and 2022, substrates joined the list of parts that held up finished products, and buyers learned that a wafer allocation means little without the parts around it. Component makers then spent years adding capacity, mostly on their own money.
The AI boom changed the economics again. Accelerator packages grew bigger and more complex, which raised the value and difficulty of every layer underneath. Customers who once bought off a catalog started negotiating dedicated lines and co-funding plants. This reported deal fits that arc: the buyer pays to build, and the supplier commits to serve.
The same logic has shown up elsewhere in the stack. Memory vendors have taken prepayments, and cloud buyers have signed multi-year capacity deals. Our report on 32 NVIDIA B300 GPUs spotted at China Telecom shows how tightly accelerator demand and supply remain linked across regions.
What It Means for AI Model Builders and Buyers
Model developers rarely think about substrates, but they feel the effects. Accelerator availability sets how fast labs can train and serve new models, and anything that loosens a bottleneck helps. If the Sejong and Busan expansions deliver on schedule, the 2028 and 2029 hardware generations face less packaging risk than today’s.
The effect on prices is harder to call. More supply can ease pressure, but customers who fund capacity also secure the first claim on it. Smaller buyers may see little benefit if the new lines are spoken for by the companies that paid to build them. For alternatives on the silicon side, see how Intel’s D-Series and AMD’s MI355X compare on memory bandwidth, since rival accelerators draw on the same packaging ecosystem.
Risks and Open Questions
Several things could change the picture. The reported amounts could shift, since no complete filing confirms them. Construction could slip past the end-2027 target, and trial operations often take longer than plans suggest. Demand could also cool, leaving a customer-funded line with less to do than expected.
A fourth risk sits on the technology side. Substrate designs keep changing as packages evolve, and a line built for today’s FC-BGA specs must stay relevant through 2033 to serve Tesla’s full term. Samsung Electro-Mechanics will need to keep upgrading tooling to hold that position.
Five Predictions for the Next 12 Months
These are our analysis, not reported facts, and each rests on the timeline and structure described above.
- A formal disclosure is likely. Material supply and investment agreements usually reach filings, so expect either confirmation or correction of the KRW3.7 trillion figure.
- No volume relief before 2028. Given the reported schedule, the new capacity cannot change substrate supply during the next year.
- Rivals will copy the structure. Other AI chip buyers will likely seek co-funded or dedicated substrate lines with their own suppliers.
- Construction news will drive sentiment. Milestones on the Sejong build will matter more to the stock story than the headline amount.
- End-use clarity will lag. Tesla is unlikely to name the exact products tied to the supply deal early, so the Optimus and driving-system link will stay speculative.
For more on how supply deals shape the field, browse our AI and machine learning coverage.
The Bottom Line
The reported Samsung Electro-Mechanics deal shows where AI spending is heading next: into the unglamorous layers that decide whether accelerators ship. The KRW3.7 trillion total, the 70/30 NVIDIA-to-Tesla split and the 2029 to 2033 Tesla term make a coherent story. The gaps, including the missing filing and the unconfirmed end uses, keep it a story about chatter rather than a closed deal. Check back as filings land.
Frequently Asked Questions
How much are Tesla and NVIDIA reportedly investing in Samsung Electro-Mechanics?
Reports put the combined figure at KRW3.7 trillion, about US$2.8 billion. Tesla’s share is reported at KRW1.1 trillion and NVIDIA’s at KRW2.6 trillion.
What is FC-BGA and why do AI chips need it?
FC-BGA stands for flip-chip ball grid array. It is the substrate that connects a chip to the circuit board, carrying power and signals. Large AI accelerators need advanced versions because of their size, heat and wiring density.
Has this investment been officially confirmed?
Partly. Samsung Electro-Mechanics and NVIDIA officially announced a joint investment in expanded dedicated facilities for ABF package substrates. The exact amounts and Tesla’s role rest on industry-source reports, and no complete public filing confirms each figure.
When will the new substrate capacity start producing?
Reports say construction at Sejong should finish by the end of 2027, equipment installation and trial operations should run in the first half of 2028, and mass production should start in the second half of 2028.
What will Tesla use the substrates for?
That is unconfirmed. Speculation points to autonomous-driving vehicles and the Optimus humanoid robot, but no official company filing or statement in the reviewed material confirms either use.
Why is the reported funding requirement KRW4.27 trillion?
Reports describe KRW4.27 trillion as the total funding Samsung Electro-Mechanics needs for its FC-BGA facility expansion. The Tesla and NVIDIA money would cover about 87% of it by our arithmetic, leaving the rest to the supplier.
Did Elon Musk or Jensen Huang comment on the deal?
No qualifying on-record quote from either executive, or from a named Samsung Electro-Mechanics executive, appears in the material behind these reports.
