The "matched set" problem
Cutress is chief analyst at More Than Moore. He describes chipmaking as assembling a complete "matched set" of components. Every piece has to be ready before a product can ship.
PCWorld's example is this. Nvidia could have plenty of finished AI silicon sitting around waiting for high-bandwidth memory (HBM). That would still free up no useful capacity for GeForce cards.
The supply chain looks like this:
- Nvidia designs the chips. TSMC manufactures most of the silicon.
- SK Hynix, Samsung and Micron supply the memory.
- AI accelerators typically pair HBM with sophisticated packaging.
- GeForce cards use consumer-oriented GDDR memory and different packaging.
- Some underlying manufacturing resources overlap, but PCWorld stresses these are not interchangeable production lines.
Six months before you even have silicon
Cutress told PCWorld that vendors typically order wafers from foundries like TSMC and then wait about six months from the start of manufacturing to the end. Treat that as his estimate of a typical schedule, not a universal foundry timetable.
Even on that estimate, the silicon alone takes half a year. Memory, packaging and board assembly still have to line up afterward. So even if Nvidia chose to prioritize GeForce tomorrow, the extra GPUs would not appear overnight.
Memory vendors have other priorities
Cutress also says memory makers are fixated on two things: HBM and DDR (or LPDDR). In his words, "We never hear them talk about GDDR."
PCWorld is careful about what this does and doesn't mean:
- GDDR isn't necessarily disappearing or in short supply.
- Cutress is unsure that the next generation, GDDR8, will arrive this decade.
- The point is about where memory vendors' attention and investment are going. It is not a claim that current GeForce memory is unavailable.
What Nvidia's own numbers show
The financial disclosures and supplier announcements back up the general picture. They don't prove anything specific about GeForce.
The $279 billion figure. Nvidia's 10-Q for the quarter ended July 26, 2026 reports supply commitments of $279 billion. These are commitments, not money already spent. The company describes them as supporting its data-center infrastructure systems, primarily memory and manufacturing facilities. They rose from $119 billion in the prior quarter. The filing spreads them over several years:
| Period | Commitment |
|---|---|
| Rest of fiscal 2027 | $92 billion |
| Fiscal 2028 | $87 billion |
| Fiscal 2029 | $88 billion |
| Fiscal 2030 | $6 billion |
| Fiscal 2031 | $5 billion |
| Fiscal 2032 and later | $1 billion |
The filing also says some of these agreements can be canceled, rescheduled or adjusted before firm orders, possibly at extra cost. It acknowledges certain supply constraints and says the complexity of data-center systems can cause production delays.
The October 21 expansion. At GTC Berlin on October 21, 2026, Nvidia announced capacity plans with six partners: Amkor, IBIDEN, King Yuan Electronics (KYEC), Samsung Electro-Mechanics, SPIL and Unimicron. They cover substrates, advanced packaging, testing and optical engines. Industry coverage notes the plans were framed around accelerated computing and AI infrastructure.
Reported targets include:
- KYEC testing: a $1.1 billion capacity reservation advance to KYEC will support new U.S. semiconductor testing facilities, with high-volume production targeted for April 2027. Nvidia said its total semiconductor testing capacity is expected to double from 2026 to 2027, with additional buildouts in Taiwan and Singapore. Expanded Singapore testing is expected to begin high-volume production in February 2027.
- IBIDEN substrates: The Ono Plant in Japan targets capacity at 3x current levels, with additional capacity expected to begin ramping in the first quarter of 2028.
- Unimicron substrates: Its Taiwan expansion is expected to begin adding capacity in the second quarter of 2027 and reach full capacity by mid-2028.
- Amkor packaging: Nvidia and Amkor plan to expand advanced packaging at Amkor's South Korea plant, targeting a greater than 300% increase in interposer wafer capacity by the end of 2028.
- SPIL optical engines: The target is 15 times current capacity by mid-2027, with another doubling by the end of 2028.
- Aggregate substrates: The IBIDEN, Samsung Electro-Mechanics and Unimicron agreements will, in aggregate, drive a 2.5x increase in capacity over the next three years.
These are Nvidia's forward-looking targets, not proof the capacity is running. Most of the dates fall in 2027 and 2028. That fits Cutress's point that supply is built over long horizons and in many stages.
"Capacity is the constraint," Jensen Huang said in the statement. Notably, he said it about AI infrastructure.
What this does and doesn't prove
- It supports: semiconductor supply needs memory, wafers, substrates, packaging and testing all to be ready. Fixing one bottleneck doesn't fix the rest.
- It doesn't show: that any particular AI chip or HBM allocation could be converted into GeForce production. No source here says how many extra cards could be made, and none establishes that GDDR is currently constraining GeForce output.
- Counterpoint: the Cutress material is analyst opinion relayed by PCWorld, and the six-month figure is his estimate. Nvidia has not publicly said it is deliberately starving GeForce. The filing simply shows where the money and capacity commitments are heading.
Why it matters for PC buyers
PCWorld's conclusion is that Nvidia could make more GeForce cards. It would need additional orders, coordinated component supplies and months of planning. That is a hard investment to justify while AI hardware earns far more per chip. The Nvidia filing is consistent with that. It splits the business into Compute & Networking and Graphics, and attributes its recent revenue growth mainly to data-center products.
For anyone waiting on GeForce prices to ease, the practical reading is cautious:
- Don't expect a quick fix from a change of leadership or priorities. A reprioritization takes months to show up on shelves.
- The large supply-chain expansions announced so far target AI hardware, and most land in 2027 and 2028.
- Any relief for consumer cards would depend on supplier decisions that Nvidia has not announced.
In short, writing a bigger check doesn't produce finished graphics cards. Nvidia would have to convince an entire supply chain to follow it.