AMD TSMC Dependency Explained: The Supply Risk to AMD Stock

Many of the companies that design the world's fastest chips never make them. They draw the circuits and pay another company to turn the designs into chips. That company's factories then decide how many chips exist and when they ship.

For someone who owns shares in a chip designer, that split moves part of the risk outside the company. Demand can be strong and the product can be ready, yet sales still depend on factory space that another company sells to many customers at once. When most of that space sits in one place, an earthquake, a price rise or a trade rule there can reach the share price quickly.

A company that designs chips and has them made elsewhere is called fabless. The company that makes them is called a foundry. At the leading edge, one foundry is far ahead of the rest, Taiwan Semiconductor Manufacturing Company, or TSMC, and much of its output comes from Taiwan.

AMD is one of those designers. Its newest server processors, PC chips and AI accelerators all start as wafers made by TSMC. So the question for an AMD shareholder is how deep that reliance runs and where it could break.

What is AMD's TSMC dependency?

AMD's TSMC dependency is its need for one company, Taiwan Semiconductor Manufacturing Company, to make all the wafers for its processors and graphics chips at 7 nanometers or smaller. Those include its newest EPYC, Ryzen and Instinct chips. AMD designs its chips and has outside foundries make all of them (source: AMD Form 10-K, fiscal 2025).

A wafer is a thin disc of silicon on which hundreds of chips are printed at once. The nanometer figure, written nm, names a generation of chipmaking, and the lower the number, the newer the process and the more tightly packed its chips. AMD's annual report says a TSMC shortfall at those sizes "could have a material adverse effect" on its business (source: AMD Form 10-K, fiscal 2025).

TSMC is the most important supplier, though it is one of several. GlobalFoundries makes AMD's older 12nm and 14nm wafers under a supply agreement that sets a minimum annual capacity and pricing through 2026 (source: AMD Form 10-K, fiscal 2025). TSMC, United Microelectronics (UMC) and Samsung make AMD's programmable logic chips (source: AMD Form 10-Q, Q2 2026).

The finished chip passes through several more hands before it reaches a customer. These are the main suppliers for AMD's newest server and AI chips, in the order the work happens.

Every one of those steps happens outside AMD. TSMC handles the two that matter most for the newest chips: the leading-edge wafer and the advanced package. That is why the reliance weighs most on AMD's AI data-center business, which runs on the newest chips.

Where AMD's supply gets tight: Wafers, packaging and memory

AMD says the squeeze reaches well past wafers. Speaking at Citi's technology conference on September 8, 2026, AMD's finance chief Jean Hu called overall supply "very tight" in several areas. She named wafers on the most advanced processes, high-bandwidth memory (HBM), packaging and substrates (source: AMD at Citi's 2026 Global TMT Conference).

Each item on that list is a separate bottleneck, and any one of them can cap how many chips AMD ships. Hu added that suppliers' capacity falls short of what the ramp of the Venice server chip needs. So AMD now buys some production equipment itself and places it at suppliers' sites, a practice called consignment (source: AMD at Citi's 2026 Global TMT Conference). On a visit to Taipei on October 6, 2026, AMD chief executive Lisa Su said AMD needs more advanced wafer capacity and is now planning three to five years ahead (source: Reuters, October 6, 2026).

AMD has one design choice that eases the wafer side. It builds many products from chiplets, several smaller chips joined in one package, so only some of the pieces need the newest process. When an analyst raised the shortage of 3nm wafers during AMD's August 4, 2026 earnings call, Lisa Su replied that chiplets let AMD "ramp in fewer wafers in the new node" (source: AMD Q2 2026 earnings call).

Say you design a processor from eight compute pieces and one input-output piece. If only the eight compute pieces use the newest process, the ninth can come from an older production line with more room, so each finished chip needs less of the scarcest wafer.

She also said AMD is watching back-end and packaging capacity and substrates, not only wafers (source: AMD Q2 2026 earnings call). Back-end work is everything after the wafer is printed: cutting it into chips, packaging, testing and assembly. For an AI chip, packaging is a step of its own at TSMC.

CoWoS packaging and why AI chips need it

CoWoS, short for chip-on-wafer-on-substrate, is TSMC's way of placing a processor and its memory stacks side by side on a thin silicon layer called an interposer. That layer then sits on a base board called a substrate. An AI accelerator needs it because the graphics processing unit (GPU) and its memory must swap data far faster than a normal circuit board allows.

AMD and TSMC say TSMC's CoWoS-L and SoIC-X packaging are used across AMD's broader AI and data-center lineup (source: AMD Venice release, May 21, 2026). That puts a second TSMC step behind most of AMD's AI chips, after the wafer itself. AMD's quarterly report adds that interposers, substrates and some other parts come from only a limited number of suppliers (source: AMD Form 10-Q, Q2 2026).

AMD has started to widen the packaging base inside Taiwan. On May 21, 2026, AMD said it would invest more than $10 billion across Taiwan's chip industry to expand partnerships and scale up advanced packaging. The plan includes work with ASE and Siliconware Precision Industries (SPIL) on a new 2.5D bridge packaging method for Venice, and a panel-based version qualified with Powertech Technology (PTI). It also names substrate makers Unimicron, Nan Ya PCB and Kinsus (source: AMD Taiwan investment release, May 21, 2026).

Those partners add packaging lines, and they belong to the same Taiwan chip industry. The program spreads the work across more companies on the island, which helps with capacity while leaving the location risk in place.

HBM memory and the memory shortage

High-bandwidth memory (HBM) stacks memory chips in towers right next to the GPU so data can reach it fast. AMD's MI455X uses HBM4, the newest kind (source: Samsung and AMD, March 18, 2026). So a memory shortage can hold back AI chip output just as a wafer shortage can.

In March 2026, Samsung signed a memorandum of understanding (MOU) with AMD to line up as the main HBM4 supplier for the MI455X (source: Samsung and AMD, March 18, 2026). That kind of agreement records an intent to cooperate, and Samsung named no volume or price.

AMD's filings flag the price side. The 10-Q describes an industry-wide memory shortage that has pushed memory prices up. It adds that the shortage may make data-center build-outs more expensive for AMD's customers (source: AMD Form 10-Q, Q2 2026). On the August 4 call, a Bank of America analyst asked if AMD's heavy use of memory leaves it open to rising memory costs. Su replied that AMD can see clearly how much HBM it will get for the products it plans to ship in 2027. She added that AMD can change how much memory each system carries if customers want (source: AMD Q2 2026 earnings call).

BloFin also lists perpetual contracts on three memory makers: the SAMSUNGUSDT Perpetual (source: BloFin, Samsung contract details), the SKHYNIXUSDT Perpetual (source: BloFin, SK Hynix contract details) and the MUUSDT Perpetual (source: BloFin, Micron contract details). You can follow the memory side of AMD's supply chain on the same exchange.

What AMD pays: Wafer prices and purchase commitments

On June 27, 2026, AMD had committed to pay about $30.3 billion to suppliers and partners, with $17.4 billion due by the end of fiscal 2026 (source: AMD Form 10-Q, Q2 2026). The money mainly covers wafers, substrates and parts, plus cloud computing and software licenses.

A purchase commitment is a deal to buy a set amount at a set time. AMD calls these unconditional, so the money is owed even if demand changes. At the end of 2025, six months earlier, the total was about $12.2 billion (source: AMD Form 10-K, fiscal 2025). Worked out, $30.3 billion is about two and a half times that. The $17.4 billion due by the end of fiscal 2026 is also about 1.5 times AMD's record revenue for the second quarter, $11.5 billion (source: AMD second-quarter 2026 results).

AMD is also paying some suppliers early. Its prepaid expenses rose by $1.0 billion in the first half of 2026, mostly because of prepayments under supply agreements (source: AMD Form 10-Q, Q2 2026). Prepaying and committing in advance is how a fabless company reserves capacity, and it shows how tight that capacity has become.

The price of each wafer is rising as well, according to reports. In September 2025, TechSpot reported DigiTimes' account that TSMC planned to raise prices on its most advanced processes by 5% to 10% in 2026, with AI and high-performance chips at the top of that range. It also reported that 2nm wafers cost about $30,000 each (source: TechSpot, September 2, 2025).

In July 2026, Tom's Hardware reported Nikkei's account of a second round. TSMC would raise base prices on 7nm-class and newer processes by 5% to 10% from the start of 2027. Customers that want extra high-performance capacity would pay a further 10% to 15%, and the report named AMD among the customers given time to adjust (source: Tom's Hardware, July 21, 2026).

Both are press reports of private terms that TSMC has not published. For AMD, the effect would show up in gross margin, which is what AMD keeps from each dollar of sales after paying to make what it sold. That makes the commitments the other side of the risk: if demand fell, AMD's annual report says it could be left with excess inventory and higher costs (source: AMD Form 10-K, fiscal 2025).

Is there a second source for AMD's chips?

AMD says TSMC will remain its main wafer supplier. At the Citi conference, Hu was asked whether AMD would use Intel as a foundry. She said AMD co-develops much of its technology with TSMC and will keep it as its "primary supplier on the wafer side" (source: AMD at Citi's 2026 Global TMT Conference).

Matt Ramsay, who runs financial strategy and investor relations, added that AMD checks the advanced wafer and packaging technology of every vendor in the market. He also said the TSMC partnership stays significant for the foreseeable future (source: AMD at Citi's 2026 Global TMT Conference). Say you read that AMD is testing another foundry. On AMD's own account that is routine checking, and TSMC remains the plan for its leading-edge chips.

Two possible second sources have come up in press reports. In October 2025, Semafor reported that Intel was in early talks to add AMD as a customer of its factories. It said it was unclear how much of AMD's production might move, and that Intel lacked the technology to make AMD's most advanced chips (source: Semafor, October 1, 2025). Intel is also AMD's main rival in PC and server processors, the contest behind AMD and Intel as stocks (source: AMD Form 10-K, fiscal 2025).

In December 2025, TrendForce passed on a report from Korea's Sedaily that Samsung was in talks to make AMD chips on an improved version of its 2nm process. A trial run would come first, and a decision was expected around January 2026 (source: TrendForce, December 15, 2025). In March 2026, Samsung's memory agreement with AMD added that the two would discuss a foundry partnership in which Samsung would make future AMD products (source: Samsung and AMD, March 18, 2026). AMD has put out no press release announcing a foundry deal with either company (source: AMD press releases).

GlobalFoundries is a real second supplier, yet it makes only AMD's older 12nm and 14nm wafers (source: AMD Form 10-K, fiscal 2025). For the chips that drive AMD's growth, TSMC is the only foundry AMD uses today.

That makes AMD's reliance on TSMC close to total for its newest chips. The offsets, from chiplets and prepaid capacity to Arizona and a second source under evaluation, ease the risk while leaving the reliance in place.

AMD chips made in Arizona

TSMC's Arizona factories let AMD make some leading-edge chips in the US, at a higher cost. AMD plans to make its Venice server chip there after starting in Taiwan (source: AMD Venice release, May 21, 2026). Lisa Su has said US-made TSMC chips cost between 5% and 20% more than the same chips from Taiwan (source: Bloomberg, July 23, 2025).

The Arizona work started with an existing product. In April 2025, AMD said it had brought up and validated its 5th Gen EPYC processors at TSMC's new Arizona factory (source: AMD, April 14, 2025). In July 2025, Su said at an AI event in Washington that Arizona chips cost "more than 5% but less than 20%" more. She also said AMD expected its first chips from Arizona by the end of that year (source: Bloomberg, July 23, 2025).

Say a chip costs $1,000 to make in Taiwan. On Su's range, the same chip from Arizona would cost somewhere between about $1,050 and $1,200, worked out. AMD would either absorb that difference in its margin or pass part of it to customers.

The newest process still starts in Taiwan. On May 21, 2026, AMD said Venice, its next EPYC server chip, was ramping on TSMC's 2nm process in Taiwan and would ramp in Arizona later (source: AMD Venice release, May 21, 2026). TSMC's annual report says its first Arizona plant began high-volume output at the end of 2024. A second is being built, and work on a third began in 2025 (source: TSMC Form 20-F, 2025).

Packaging in the US comes later. Amkor is building what it calls the first US plant for high-volume advanced packaging, in Arizona, and expects the first factory to start production in early 2028 (source: Amkor, October 6, 2025).

Chip tariffs and the US-Taiwan trade deal

Tariffs are the other reason US production matters. A tariff is an import tax, and Section 232 of US trade law lets the president set one on national-security grounds. Proclamation 11002 used it in January 2026 and set out a two-phase plan (source: Proclamation 11002).

The first phase is a 25% tax on a narrow group of advanced computing chips, in force since January 15, 2026 (source: Proclamation 11002). Chips brought in for US data centers are exempt, and the proclamation lists other exempt US uses too. The second phase is wider chip tariffs once trade talks end, with an offset program that would give better terms to companies that invest in US chip plants (source: Proclamation 11002). AMD's quarterly report adds that its MI325 chips bound for China pay a 25% tariff when they enter the US for the inspection their export licenses require (source: AMD Form 10-Q, Q2 2026).

Taiwan signed an MOU with the US on January 15, 2026. It set a 15% reciprocal tariff on Taiwan's goods and promised better Section 232 terms for chips, with tariff-free quotas for Taiwanese firms that invest in the US. As of May 28, 2026, the US had not yet published the list of chip products those Section 232 tariffs would cover (source: Taipei Times, May 29, 2026). That 15% rate was one of the reciprocal tariffs the US Supreme Court struck down in February 2026 (source: CNA via OCAC News, February 24, 2026). Since July 24, 2026, Taiwan has instead faced a 10% US tariff under Section 301 of US trade law, part of an action covering 60 economies over forced-labor import rules rather than a chip measure (source: Focus Taiwan, July 24, 2026).

The broader tariff is still being shaped. On September 2, 2026, Commerce Secretary Howard Lutnick described the plan as "if you build here, you don't pay" (source: Korea Herald, September 3, 2026). AMD's 10-Q warns that tariffs on data-center hardware could raise its customers' costs and lead them to delay or cancel AI projects (source: AMD Form 10-Q, Q2 2026).

What a Taiwan shock could do to AMD stock

A Taiwan shock would hit AMD stock through supply first: fewer wafers, packages or racks, and so fewer chips to sell. AMD's annual report names the risk. It says tension between China and Taiwan could disrupt its foundries and suppliers there, and it notes the April 2024 earthquake near those foundries (source: AMD Form 10-K, fiscal 2025).

That earthquake is the clearest recent test. It measured magnitude 7.4 and struck near Hualien at 11.58PM UTC on April 2, 2024, which was the morning of April 3 in Taiwan (source: USGS, M 7.4 Hualien earthquake). The same day, TSMC said more than 70% of its factory tools had recovered within 10 hours, more than 80% at its newer Fab 18, and none of its extreme ultraviolet (EUV) lithography machines were damaged (source: TSMC statement, April 3, 2024).

AMD's shares closed about 1.2% higher on April 3, at $180.77, against $178.70 the session before (source: Yahoo Finance, AMD price history). The shares rose even though the quake had hit the island where its foundries are. A longer outage would test something else: how long AMD could keep selling from stock.

AMD held $8.5 billion of inventory on June 27, 2026, and $5.3 billion of it was still work in process, meaning chips not yet finished (source: AMD Form 10-Q, Q2 2026). AMD's cost of sales in the second quarter, its revenue minus its gross profit, came to about $5.3 billion over 13 weeks (source: AMD second-quarter 2026 results). At that pace, the inventory equals about 145 days of cost of sales, worked out. Most of it was still unfinished, though, and those chips need more factory work before they can ship. So the stock would cover a short disruption and leave a long one exposed.

Racks add a further link. In its Taiwan investment announcement, AMD named Wiwynn, Wistron and Inventec, alongside Sanmina, as partners building its Helios racks (source: AMD Taiwan investment release, May 21, 2026). A shock to that Taiwan supply chain could therefore hold up finished systems even when the chips themselves were safe.

BloFin offers AMD exposure through the AMDUSDT Perpetual, a contract with AMD stock as its underlying asset (source: BloFin listing notice, May 6, 2026). Leverage on it goes up to 20x (source: BloFin, AMDUSDT contract details). Taiwan's working day falls overnight in New York. BloFin may restrict AMDUSDT to reduce-only orders outside regular weekday trading hours, and on weekends and market holidays (source: BloFin stock futures notice, April 21, 2026).

With a reduce-only order you can make a position smaller or close it, but you cannot add to it or open a new one (source: BloFin Help Center, Reduce Only). If a big move comes overnight, a leveraged position whose margin falls to its maintenance margin plus the closing fee can be partly or fully closed by liquidation before the next regular session (source: BloFin Help Center, Forced Liquidation Rules).

That build-up between two sessions is overnight and weekend gap risk, and leverage magnifies it. BloFin's regular weekday hours for the AMD perpetual start at 1.30PM UTC during US daylight saving time (source: BloFin stock futures notice, April 21, 2026).

You can watch the contract's price and its current funding rate on the AMDUSDT Perpetual page.

AMD vs TSMC stock: The customer and the supplier

AMD and TSMC both profit from AI chip demand, from different points in the chain. From the end of 2025 to October 6, 2026, AMD stock gained about 203%, worked out from closing prices (source: Yahoo Finance, AMD price history). TSMC's US-listed shares gained about 59% over the same period (source: Yahoo Finance, TSM price history).

The two companies differ most in how much each keeps from a sale and how much each relies on its biggest customer.

TSMC keeps about 68 cents of gross profit from each sales dollar and AMD about 54 cents, so the supplier earns the wider margin. TSMC also depends on a few buyers: its ten largest customers made up 78% of its 2025 revenue (source: TSMC Form 20-F, 2025). AMD's own sales are spread more widely, with no customer at 10% of revenue in 2025 (source: AMD Form 10-K, fiscal 2025).

TSMC guided to third-quarter revenue of $44.6 billion to $45.8 billion (source: TSMC second-quarter 2026 results). Its August 2026 revenue was NT$514.8 billion, up 53.3% from a year earlier (source: TSMC August 2026 revenue report). Those monthly reports are an early read on chip demand, AMD's included.

BloFin lists TSMC as a perpetual too, the TSMUSDT Perpetual, with leverage of up to 20x (source: BloFin, TSMUSDT contract details). A trader who holds both contracts has exposure to the chip designer and to its main supplier at once. Using one contract to offset the risk in another is the idea behind hedging with the AMD perpetual.

Supply-chain dates to watch

Two fixed dates come next. TSMC holds its third-quarter earnings conference on October 15, 2026 (source: TSMC, third-quarter 2026 earnings conference). AMD reports its third quarter on November 3, 2026, after the market close (source: AMD, October 6, 2026). Its filing will then update three figures showing how much supply AMD has locked in: its purchase commitments, prepayments and inventory.

Say you see the commitments rise again from $30.3 billion. That would mean AMD has reserved still more wafers and parts ahead of demand, which supports future sales and raises the cost if orders slow. A fall would point the other way, toward a company that needs less capacity than it booked.

TSMC's monthly sales reports come sooner. Its August figure came out on September 10, 2026 (source: TSMC August 2026 revenue report). If TSMC's sales keep rising while its factories stay full, demand for advanced chips is still running ahead of supply.

Three slower stories matter as well. The first is TSMC's reported price rise for 2027, which would reach AMD's gross margin next year if it happens (source: Tom's Hardware, July 21, 2026). The second is the broader phase of US chip tariffs, which the Commerce Secretary described in September as still being prepared (source: Korea Herald, September 3, 2026). The third is any announced second-source deal with Intel or Samsung, which would change how much AMD relies on one company.

Supply news can move AMD's price sharply, part of the wider pattern in how much AMD's price swings. Using less leverage gives a supply headline more room to pass without forcing you out of a position.

Looking to trade the AMDUSDT Perpetual as AMD's supply story develops? To get started, you'll need to first create a BloFin account, fund your account with cryptocurrency, and go to the AMDUSDT Perpetual trading page, TSMUSDT Perpetual page, SAMSUNGUSDT Perpetual page, SKHYNIXUSDT Perpetual page or MUUSDT Perpetual page.

Frequently asked questions

Does AMD use TSMC?

AMD uses TSMC to make all the wafers for its processors and graphics chips at 7 nanometers or smaller (source: AMD Form 10-K, fiscal 2025). TSMC also packages AMD's AI chips (source: AMD Venice release, May 21, 2026). In April 2025, AMD said its Venice server chip was the first high-performance computing product to be taped out and brought up on TSMC's 2nm process (source: AMD, April 14, 2025). Taping out is the point where a finished design goes to the factory.

Is Nvidia made by TSMC?

Nvidia, like AMD, has TSMC make its leading AI chips, and TSMC's Arizona plant celebrated its first Nvidia Blackwell wafer in October 2025 (source: NVIDIA, October 17, 2025). The two rivals compete for the same factory space, a story told from Nvidia's side in Nvidia's dependence on TSMC. TSMC does not name its biggest buyers, but its annual report says its largest customer made up 19% of its 2025 revenue and the second largest 17% (source: TSMC Form 20-F, 2025). A big order from one designer can leave less capacity for the others.

How much has AMD committed to pay its suppliers?

AMD's unconditional commitments totaled about $30.3 billion on June 27, 2026, and most of the money falls due soon. The schedule in its quarterly report shows $17.4 billion for the rest of fiscal 2026, $5.5 billion for 2027 and $2.9 billion for 2028. It then shows $2.5 billion for 2029, $1.5 billion for 2030 and $0.6 billion after that (source: AMD Form 10-Q, Q2 2026). The filing gives no split by supplier, so TSMC's share is not public.

Does TSMC make chips in the US?

TSMC makes chips in Phoenix, Arizona, through its unit TSMC Arizona, whose first factory began high-volume output at the end of 2024. The US government is helping to pay for it. Under the CHIPS Act, the Commerce Department agreed to give TSMC Arizona up to $6.6 billion in direct funding, plus up to $5 billion in loans (source: TSMC Form 20-F, 2025). AMD's newest server chip, Venice, started 2nm production in Taiwan, and Arizona comes later (source: AMD Venice release, May 21, 2026).

Who is TSMC's biggest rival?

Samsung is TSMC's closest rival among contract chipmakers, though far smaller. TrendForce's ranking for the second quarter of 2026 put TSMC's market share at 72.5% and Samsung's, in second place, at 5.9% (source: TrendForce, September 9, 2026). Intel also sells factory space to outside chip designers, and Semafor reported in 2025 that it was courting AMD (source: Semafor, October 1, 2025). For AMD, that gap helps explain why a second source takes time: few other companies can make its newest chips at the volume it needs.

Researched and written by the BloFin Academy editorial team with AI-assisted drafting. All facts independently verified. Primary sources include AMD's Form 10-K for fiscal 2025, its Form 10-Q for the quarter ended June 27, 2026 and TSMC's Form 20-F for 2025, plus Nvidia's blog of October 17, 2025 and press reports from Reuters, Bloomberg, TechSpot, Semafor, TrendForce, the Taipei Times, CNA, Tom's Hardware and the Korea Herald. They also include AMD's releases of April 14, 2025, May 21, 2026, August 4, 2026 and October 6, 2026, its list of press releases, the Samsung and AMD news of March 18, 2026, and AMD's call and conference transcripts of August 4 and September 8, 2026. TSMC's sources are its statement of April 3, 2024, its second-quarter 2026 results, its August 2026 revenue report and its third-quarter 2026 earnings conference notice. The rest are the USGS record of the 2024 Hualien earthquake, Proclamation 11002, the Amkor release of October 6, 2025, and price history from Yahoo Finance. BloFin's sources are its AMDUSDT, TSMUSDT, Samsung, SK Hynix and Micron contract details, its AMDUSDT listing notice, its stock futures notice and its help center articles on reduce-only orders and forced liquidation, current as of October 7, 2026.

Nothing in this article constitutes financial advice, and nothing in it is a recommendation to buy, sell or hold AMD in any form. AMD's newest chips depend on TSMC's factories in Taiwan. A disruption there, a TSMC price rise or a new chip tariff can cut AMD's sales or margins and move its share price quickly. Supply news from Taiwan often arrives outside regular weekday trading hours, when BloFin may accept only reduce-only orders on the AMDUSDT Perpetual, and a leveraged position that falls to its maintenance margin can be liquidated. Past performance does not indicate future results. Do your own research and consider your risk tolerance before you trade on BloFin.