Tesla, Nvidia Funded Their Own Chip Supplier — Circular?

By Daniel Rao

Nvidia just committed $2 billion to the company that builds the trays its chips sit in. Tesla threw in another $830 million. And Wall Street's reaction was swift and allergic: by the time most of America had finished its coffee on Thursday, Nvidia, Intel and AMD were all sliding in premarket trade — because when your biggest customers start bankrolling your factory, investors hear exactly one word.

Circular.

The report, from Seoul's Maeil Business Newspaper and carried by Reuters on October 22, says Tesla and Nvidia will invest a combined ₩3.7 trillion — roughly $2.8 billion — in Samsung Electro-Mechanics to build production lines for FC-BGA chip substrates in the South Korean city of Sejong, with the output supplied back to the two companies under a long-term contract. Read that sentence again: the buyers are funding the supplier, whose new factories exist to feed the buyers. It is a closed loop with $2.8 billion drawn around it — which is why the market's first instinct was not celebration.

A Samsung chip-fab cleanroom in Pyeongtaek, South Korea — 2022 file photo. The reported $2.8 billion deal would build new lines at Samsung Electro-Mechanics' separate Sejong site. Photo: The White House, public domain, via Wikimedia Commons.

What was reported — and what was actually confirmed

Precision matters here, because the reporting and the confirmation are two different stories. What Maeil reported, citing industry sources, and what Reuters relayed: Tesla invests ₩1.1 trillion, Nvidia ₩2.6 trillion, new FC-BGA production lines in Sejong, substrates for AI semiconductors, long-term supply contract. (Reuters converts at ₩1,322.44 to the dollar — that is where the $2.8 billion figure comes from.)

What has actually been confirmed: a day earlier, on October 21, Samsung Electro-Mechanics and Nvidia announced a formal strategic partnership — joint investment in dedicated ABF package-substrate facilities at SEMCO's Sejong and Busan sites, with deliveries of substrates for Nvidia's AI compute chipsets and high-speed networking switches beginning in 2026 under a multi-year agreement.

One caution, from our own checking: we read the PRNewswire release behind the official announcement end to end. The ₩3.7 trillion figure appears nowhere in it — nor do Tesla or the per-company breakdowns. As of this writing, none of the three companies has confirmed the $2.8 billion. The numbers belong to Maeil's sources; the partnership belongs to the record books.

The least glamorous bottleneck in AI

FC-BGA stands for flip-chip ball grid array. Think of it as the foundation slab under a skyscraper: the AI chip — Nvidia's GPU, Tesla's AI silicon — is flipped face-down and bonded onto the substrate through a grid of tiny solder balls, which carry power and thousands of high-speed signals between the die and the rest of the board. No substrate, no working chip.

A flip-chip ball grid array (FC-BGA) package — the grid of solder balls carries the dense, high-speed connections AI chips need. This is the component the reported $2.8 billion would go toward making. Photo: Wikimedia Commons, CC BY-SA 3.0.

And packaging, quietly, has been one of the AI buildout's genuine choke points. ABF — Ajinomoto Build-up Film, the material these substrates are built on — was in brutal shortage during the 2021–22 chip crunch, and AI accelerators are far more substrate-hungry than the smartphone chips of that era. That shortage history is the single best argument the bulls have.

Why the market flinched

The numbers on Thursday morning were ugly across the board. Nvidia fell 2.2% in premarket, erasing a 2.5% rally from Wednesday that had followed Jensen Huang's keynote at GTC Berlin. Intel dropped nearly 5%, AMD 3.8%, Marvell 5.3%, Broadcom 2.4%, Qualcomm 2.9%, Micron 3.5%. Nasdaq 100 futures were down almost 2%. By midday the Nasdaq Composite had shed 2.4%, and Nvidia closed the session down 4.4% at $212.21.

Tesla beat — $0.48 adjusted vs $0.40 expected on $28.05 billion of revenue — but cut its 2026 capex forecast to ~$20 billion from "more than $25 billion." Meta posted record revenue of $64.6 billion but took a $9.89 billion legal charge and unveiled a new non-voting Class C share, effectively a 2-for-1 split preserving Zuckerberg's control. Both slid in premarket — Tesla 5.3%, Meta 1.9%.

Barron's nailed the specific poison in the Samsung report: a "classic case of bad timing," because the deal "further underscores the possible fragility of 'circular financing'" — investing in and lending to your own customers and suppliers, creating "a positive feedback loop of demand that may not reflect reality." The precedent investors keep citing: Nvidia's investment in OpenAI in late 2025, where reports suggested much of the funding would circle straight back into leasing Nvidia chips. Stacy Rasgon, an analyst at Bernstein, wrote in an investor note when that deal was announced that it would "clearly fuel 'circular' concerns" — and warned its sheer size would "likely fuel these worries much hotter than what we have seen previously." Samsung Electro-Mechanics now fits the template uncomfortably well: Nvidia money builds the lines, the lines make substrates, the substrates go into Nvidia chips.

"Investors grapple with AI growth concerns, higher oil prices, and ongoing fiscal stress," Henry Allen, an analyst at Deutsche Bank, told Barron's on Thursday. "The latest earnings after the US close failed to reassure investors… with tech stocks struggling to gain traction since."

The defence: this is how chip capacity actually gets built

The counterargument deserves a full hearing, though — customer-funded capacity is one of the oldest tricks in semiconductors.

Prepayments, take-or-pay contracts and co-invested lines are how the industry has financed expansions for decades. A substrate maker facing a multi-trillion-won fab decision faces brutal demand risk: build it and they don't come, and you're finished. A long-term contract backed by customer money converts that risk into a schedule. SEMCO gets revenue visibility; Nvidia and Tesla — which designs its own AI silicon — get supply assurance through 2028.

And the bottleneck is real. The official announcement says it plainly: the deal exists to "alleviate supply chain bottlenecks in the rapidly expanding global AI infrastructure market." According to Korean coverage, SEMCO's shipments to Nvidia are expected to grow at a double-digit annual rate from 2026, and by 2028 Nvidia could account for a significant share of the company's total package-substrate capacity. When your supplier might otherwise sell that capacity to someone else, writing the cheque yourself is not charity — it's procurement. It is the same bottleneck logic behind efforts like the Google–NVIDIA–Emerald AI alliance to keep the data-center buildout from stalling on infrastructure.

Debora Shoquist, Nvidia's EVP of operations, said in the release: "The pace of the AI infrastructure buildout requires us to expand manufacturing capacity on a global scale. Our work with Samsung Electro-Mechanics broadens our network of substrate partners and adds manufacturing capacity to help us scale production and deliver the accelerated computing platforms our customers need."

And Duck-Hyun Chang, Samsung Electro-Mechanics' president and CEO: "It is a great honor to join NVIDIA, a global leader driving AI innovation, as a key infrastructure partner." He added that combining the company's FC-BGA manufacturing with its MLCC expertise would ensure "NVIDIA's AI platforms are delivered to the market on time with the highest levels of performance and quality."

Our read: the market isn't wrong to notice — but it may be wrong about what it means

Here is the uncomfortable truth both camps keep dancing around: the circularity is real, and it might also be fine. Nvidia financing its own supply chain is only alarming if you believe AI demand is fragile. If you believe, as Huang does — he spent Thursday morning telling CNBC he wants lenders to treat Nvidia's chips as a "financeable asset," an asset class in its own right — then customer-funded capacity is just vertical integration by another name, and the timing was awful.

But — and this is the part the bulls keep skipping — there is a version of this story where the financing treadmill never stops. Nvidia invests in OpenAI, which leases Nvidia chips. Nvidia backs neoclouds, which buy Nvidia chips. Nvidia now funds the substrate maker, whose substrates go into Nvidia chips. (CoreWeave, the neocloud poster child, fell 9.6% on Thursday on its own debt worries. It had asked Wall Street for up to $6.4 billion in a single morning last month.) Each deal is defensible. The pattern, taken together, is an industry increasingly financing its own demand — and an investor is entitled to ask what happens to "record demand" if the financing ever pauses. The IMF made the same point at the macro level this month, warning that hundreds of billions in AI investment now move in loops where big tech effectively funds its own revenue.

Nobody knows, because the structure of the reported investment is undisclosed. Is it equity? A prepayment against future supply? A loan? "Investment," as Maeil used it, could mean any of those, and they imply very different things. A prepayment is procurement. Equity is a bet. The market, lacking the distinction, priced in the worst interpretation — which, on a day when oil spiked past $112 and the 10-year yield sat near 5%, was always going to happen.

The supplier side rallied: Seoul disagreed with Wall Street

Seoul disagreed with Wall Street — at least on the supplier side. Samsung Electro-Mechanics shares jumped in early trading on October 22: ChosunBiz reported the stock up 3.08% to ₩1,204,000 at 9:35 a.m., at one point touching a 5.65% gain, even as the KOSPI fell more than 1% (MK reported a 4.05% rise to ₩1,232,000). Rival substrate maker Korea Circuits rose roughly 7% on the same morning. Separately, Nvidia used its October 21 GTC keynote to name Samsung Electro-Mechanics alongside Japan's Ibiden and Taiwan's Unimicron as substrate partners, and said it plans to expand substrate capacity 2.5 times over three years (mk.co.kr). Meanwhile East Asia Brief reported the ₩3.7 trillion co-investment covers roughly 87% of a 4.27 trillion won Sejong expansion Samsung Electro-Mechanics had already disclosed on September 29 — with a companion 2.51 trillion won substrate facility in Vietnam and deliveries slated from 2026.

What we still don't know

• Whether anyone confirms the numbers. The ₩3.7 trillion figure comes from Maeil's industry sources. Until one of the three companies puts it in a filing or a release, it is a report — well-sourced, but a report.
• Whether this is new money or the same money with numbers attached. The official October 21 partnership involves "joint investment" of an undisclosed size. Is Maeil's figure that same commitment quantified, or an additional one? A day apart, conspicuously related — but unlinked on the record.
• What Tesla's AI chips have to do with it. Tesla co-funding a substrate plant deserves more attention: it designs its own AI silicon for vehicles and robotics, so its presence suggests those chips are in the long-term supply plan. That's itself a story.
• The structure. Equity, prepayment, loan or hybrid — the market is guessing, and guessing darkens the mood.

Frequently asked questions

What is FC-BGA, and why does it matter for AI chips?
Flip-chip ball grid array: the chip is bonded face-down onto a substrate through a grid of solder balls carrying power and thousands of high-speed signals. AI accelerators use enormous, connection-dense dies — without advanced substrates they cannot be packaged at all.

Has the $2.8 billion deal been confirmed?
No. It was reported by South Korea's Maeil Business Newspaper, citing industry sources, and carried by Reuters on October 22, 2026. Separately, Nvidia and Samsung Electro-Mechanics confirmed a strategic partnership on October 21 — but that announcement disclosed no investment figures.

Is the reported investment the same thing as the official Nvidia–Samsung Electro-Mechanics partnership?
Not as far as the public record shows. The official partnership (October 21, via PRNewswire) covers joint investment in dedicated ABF substrate facilities in Sejong and Busan, with supply starting in 2026. The $2.8 billion figure (October 22, via Maeil) names Tesla as a co-investor and gives specific won amounts. The timing is conspicuous, but no official source has linked them.

Why is Wall Street calling this "circular financing"?
Because the money flows in a loop: Nvidia funds the supplier, the supplier builds the lines, the lines produce substrates that go into Nvidia's products — the same concern raised over Nvidia's late-2025 OpenAI investment.

Sources: Reuters (Oct 22, 2026 — Maeil Business Newspaper report; Jack Kim reporting); Samsung Electro-Mechanics–NVIDIA partnership via PRNewswire (Oct 21, 2026); mk.co.kr (partnership details); Barron's (Oct 22 — Henry Allen/Deutsche Bank quote); TipRanks (Oct 22 — circular-financing explainer); Bernstein via i3investor (Sept 2025 — Stacy Rasgon note); Stocktwits (Oct 22 — Huang CNBC interview, GTC Berlin); WABX/Reuters (Oct 22 — futures, oil, yields); AIStockWire, Zacks, FinanceFeeds (Oct 21–22 — Tesla/Meta Q3); EL7/AP (Oct 22 — session close).

About the author: TechPulse USA's news desk covers the AI industry across the US, Europe, Canada and Australia — model launches, lab politics, and the numbers behind the hype.