Tesla and Nvidia will put 3.7 trillion won, about $2.8 billion, into Samsung Electro-Mechanics to build production lines for the chip substrates used in AI semiconductors, South Korea’s Maeil Business Newspaper reported on 22 October 2026, citing industry sources. Reuters relayed the report: “Tesla will invest 1.1 trillion won and Nvidia 2.6 trillion won to build the production lines in the South Korean city of Sejong to supply the substrates under a long-term contract, the paper reported, citing industry sources.”
None of the three companies has confirmed the amounts. Nvidia named Samsung Electro-Mechanics as a substrate partner at GTC Berlin a day earlier without giving a figure. As we read it, if the report holds, Nvidia’s share would be about 1.8 times the $1.1 billion advance it disclosed for test supplier KYEC the same day (our arithmetic).
Update (24 Oct): reports differ on the size of the investment.Businesskorea reported on 22 October that Samsung Electro-Mechanics “will receive an investment of about 3 trillion won (about $2.27 billion) from Tesla and Nvidia” for its FC-BGA production facilities, and that “the investment scale of Tesla and Nvidia is known to be about 1 trillion won and 2 trillion won, respectively.” That is below the 3.7 trillion won (1.1 trillion from Tesla, 2.6 trillion from Nvidia) that Maeil Business Newspaper reported and Reuters relayed, the figures used in this piece.
Businesskorea quoted a Samsung Electro-Mechanics official: “It is difficult to confirm parts related to clients.” Neither total has been confirmed by the companies.
Update (26 Oct): Taiwan’s Economic Daily says Intel, Nvidia and AMD have reportedly prepaid more than NT$100 billion for substrate capacity. The paper reported on 24 October that industry sources say the three have paid deposits totalling more than NT$100 billion to reserve long-term high-end IC substrate capacity. Analysts estimate Intel’s share at about US$3 billion; the paper says AMD’s is put at about US$1 billion by industry interpretation and Nvidia’s at about US$1 billion by outside speculation, about US$5 billion in all.
The paper ties the reports to record capex at Taiwan’s substrate makers: Unimicron raised 2026 capex to NT$53.7 billion, Kinsus approved NT$23.5 billion over three years for ABF substrate equipment, and Nan Ya PCB approved a NT$46.8 billion factory. It put announced 2026 expansion at a record NT$161.2 billion. Dan Nystedt flagged the report on X as a rumour.
The per-company figures are estimates, the substrate makers do not comment on individual customers, and no comment from Intel, Nvidia or AMD was reported.
Business Pill · GPUS EXPLAINED
A short Business Engineer explainer on why AI runs on graphics chips. It teaches the general idea only and says nothing about any company or person in this piece.
The key insight: As we read it, the new element is a price tag on a supply deal Nvidia announced without one: Maeil reports 2.6 trillion won from Nvidia and 1.1 trillion won from Tesla for Samsung Electro-Mechanics’ substrate lines, while Nvidia’s own release says only that the collaboration “supports facility upgrades and new factories in Sejong and Busan”.
A substrate is the base layer that a chip is mounted on and that carries its connections to the rest of the board. The kind in question, FC-BGA (flip chip ball grid array), is used for AI accelerators, graphics processors and server chips. Reuters describes the deal as investment “to build facilities making advanced chip substrates, known as FC-BGA for AI semiconductors”.
Maeil’s own report, in Korean, adds detail (our translation). It says Samsung Electro-Mechanics signed a long-term FC-BGA supply contract with Tesla and is adding a dedicated line at its Sejong site. Tesla would provide about 1.1 trillion won, part of the money needed for the building and equipment, in exchange for guaranteed medium- to long-term supply. The substrates would go into the AI chips of Tesla’s self-driving cars and its Optimus humanoid robot.
On timing, Maeil reports that the plant is under construction, with completion targeted by the end of next year, equipment and trial runs by the first half of 2028, mass production from the second half of 2028 and supply to Tesla over five years, from 2029 to 2033. It says Samsung Electro-Mechanics disclosed in September a total investment of 4.27 trillion won for the Sejong expansion, of which Tesla would cover 1.1 trillion won. That is about a quarter of the expansion (our arithmetic).
For Nvidia, Maeil says the company did not disclose an amount at GTC, and that Maeil’s own reporting puts it at about 2.6 trillion won, for a dedicated line making AI computing chipset and data-centre network substrates.
Supplier support for AI chip capacity, in billions of dollars. Nvidia’s $1.1 billion advance to KYEC is from Nvidia’s 21 October 2026 release. The Samsung Electro-Mechanics figures, 2.6 trillion won from Nvidia and 1.1 trillion won from Tesla, are as reported by Maeil Business Newspaper and relayed by Reuters, and are not confirmed by the companies; dollar conversions at 1,322.44 won are our arithmetic.
What Was Posted on X
Pretty wild that GPUs are now the only balanced part of the AI supply chain while DRAM, SSDs, hard drives and chip substrates face lead times 2โ5x longer than normal.
That means even if $AMZN, $MSFT, $GOOGL and $META secure their $NVDA GPUs, they still need CPUs, memory and packaging materials like ABF that can take nearly a year to source versus just three months normally.
That’s a massive opportunity for $MU and $SKHY in DRAM and HBM, $SNDK in enterprise SSDs, $WDC and $STX in hard drives and $INTC and $AMD in CPUs as AI agents drive demand across entire infrastructure stack.
@stocksavvyshay wrote on 25 October that GPUs are now the only balanced part of the AI supply chain, while DRAM, SSDs, hard drives and chip substrates face lead times 2 to 5 times longer than normal, with packaging materials like ABF taking nearly a year to source versus three months normally. The stock picks in the post are the poster’s own view.
Google VP Bill Jia says the agent era needs more memory plus CPUs, flash and hard drives, so every component stays scarce
โI think memory will still be very scarce. Because this agent thing has happened. Before, because it was only, I mean, mainly still in inference, in this raw inference. In raw inference, what it lacks is accelerator, like GPU and TPU, this ASIC, what it lacks is accelerator and memory. That was before, at this raw inference stage.”
“After the agentic era, it needs a lot of CPU too, it needs even more memory, and it also needs a lot of flash and hard drive, storage. Like what I said just now, in the future agents will definitely bring better, I mean, enterprise oriented and consumer oriented new product, so the demand will be very large. So I think for all of these components, I think theyโll all still be fairly scarce.โ _______ Link and breakdown of the Google VP interview: https://firesidealpha.substack.com/p/googles-bill-jia-vp-of-engineering
Fireside Alpha (@firesidealpha) relays Google VP Bill Jia saying the agent era needs more memory plus CPUs, flash and hard drives: “I think memory will still be very scarce.”
What Nvidia and Samsung Electro-Mechanics Said on Record
Nvidia’s 21 October release says: “Samsung Electro-Mechanics is joining NVIDIA’s substrate supply chain, bringing immediate capacity from its existing operations in South Korea.” It adds that “The collaboration also supports facility upgrades and new factories in Sejong and Busan, South Korea” and that “By 2028, production allocated to NVIDIA is projected to account for a significant portion of Samsung Electro-Mechanics’ total annual package substrate production capacity.”
The release frames this as part of a wider plan: “NVIDIA plans to increase substrate capacity for its products by 2.5x over the next three years.” Jensen Huang is quoted as saying: “Demand is extraordinary. Capacity is the constraint”.
In the two companies’ announcement, as reported by Nocut News, Nvidia’s Debora Shoquist said: “Through our collaboration with Samsung Electro-Mechanics, we have expanded our substrate supply partner network and secured additional production capacity.” Neither the release nor the quoted statements gives a dollar figure for Samsung Electro-Mechanics, and we found no Tesla statement on the report.
What Nvidia Has Put Behind Its Suppliers
Nvidia’s release lists several supplier deals. The table sets them beside the reported Samsung Electro-Mechanics figures. Amounts in won are converted at Reuters’ rate of 1,322.44 won to the dollar (our arithmetic).
Supplier
What
Support
Source
Samsung Electro-Mechanics
Substrates, Sejong and Busan
About 2.6 trillion won, about $1.97 billion (reported, not confirmed)
Maeil via Reuters; Nvidia release names the partnership, no figure
Samsung Electro-Mechanics (Tesla)
FC-BGA substrates for Tesla AI chips, Sejong
About 1.1 trillion won, about $832 million (reported, not confirmed)
As we read it, the common thread is that Nvidia is paying ahead for supplier capacity rather than only placing orders, and the reported Tesla money follows the same pattern for a carmaker.
The Structural Read
As we read it, the buyers are funding the supplier’s factory: Maeil describes Tesla covering part of the building and equipment cost in exchange for guaranteed supply, and Nvidia’s release lists capacity reservation advances for IBIDEN and KYEC.
As we read it, the constraint Nvidia names has moved down the stack: its release says “NVIDIA plans to increase substrate capacity for its products by 2.5x over the next three years”, and Jensen Huang is quoted as saying “Capacity is the constraint”.
As we read it, Tesla is buying the same scarce part as Nvidia for its own chips: Maeil says the substrates are meant for the AI chips in Tesla’s self-driving cars and Optimus robot, with supply from 2029 to 2033.
Debora Shoquist, Nvidia, as reported by Nocut News
“Through our collaboration with Samsung Electro-Mechanics, we have expanded our substrate supply partner network and secured additional production capacity.”
Three Implications
FOR CHIP BUYERS As we read it, Nvidia’s release ties capacity to paying ahead: a “multibillion-dollar capacity reservation advance” for IBIDEN and a $1.1 billion advance for KYEC, with the reported Samsung Electro-Mechanics money on top.
FOR SUPPLIERS Nvidia says that “By 2028, production allocated to NVIDIA is projected to account for a significant portion of Samsung Electro-Mechanics’ total annual package substrate production capacity.”
FOR READERS OF SUPPLY-CHAIN NEWS As we read it, the amounts remain a newspaper report until a company or filing confirms them, which is why the table keeps the reported and the announced figures apart.
The framework argues that the capex race “is a sovereign-scale financing problem dressed in tech-company clothing”.
As we read it, buyers pre-funding a substrate maker’s factory is one form of that financing moving down the supply chain. The framework was published in May 2026 and does not discuss Samsung Electro-Mechanics.
What Is Not Established
The amounts come from Maeil, citing industry sources, and none of Tesla, Nvidia or Samsung Electro-Mechanics has confirmed them. Reuters summarises both lines as being in Sejong, while Nvidia’s release and Maeil mention Sejong and Busan for Nvidia. We could not find Samsung Electro-Mechanics’ September filing to check the 4.27 trillion won figure against it, and we did not find a Tesla statement. Whether the money is equity, a prepayment or a capital contribution is not spelled out beyond Maeil’s description.
Connecting the Coverage
On 21 October we reported Nvidia’s $1.1 billion advance to KYEC for new U.S. test sites, in our piece on Nvidia’s supplier advances, from the same GTC Berlin release that named Samsung Electro-Mechanics. Our GTC Berlin wrap the same day covered Nvidia’s European commitments.
Also on 21 October, our piece on Tesla’s AI compute reported its plan to reach 400 MW of AI training compute by the end of 2026, and our Optimus piece reported Musk saying the robot is a year or more from 20,000 units a week. What’s new here is a reported figure for what both Nvidia and Tesla would pay to lock in the substrates those chips sit on.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This piece is part of this week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Maeil reports, and Reuters relays, that Tesla and Nvidia will fund 3.7 trillion won, about $2.8 billion, of Samsung Electro-Mechanics’ AI substrate lines. Nvidia has confirmed the partnership but not the amount. As we read it, the figure would make substrate supply one more place where AI chip buyers pay in advance for capacity.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. The investment figures come from Maeil Business Newspaper’s Korean-language report of 22 October 2026, citing industry sources, and Reuters’ summary of it; the English renderings of Maeil’s details are our translation. None of Tesla, Nvidia or Samsung Electro-Mechanics has confirmed the amounts. Nvidia’s statements come from its 21 October 2026 release, and the Shoquist quote from Nocut News’ report of the companies’ announcement. Dollar conversions at Reuters’ rate and the ratios are our arithmetic. Nothing here is investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
