Nvidia, the memory crunch, and GPU prices

1.Jensen Huang did resign

Nvidia's board announced on 12 November 2026 that Jensen Huang had resigned as president, chief executive and a board member, effective immediately. The board had decided the night before, according to CNBC's reporting.

The trigger was a week of revelations about his personal life. The New York Times and others reported he had an affair with a businesswoman, Alice Huang, who had worked for a Chinese company and Nvidia customer called Megaspeed and whom US intelligence agencies believe was recruited as a spy. A second affair, with an Nvidia vice-president named Kimberly Powell, was reported the night before he went. Huang apologised publicly for being "unfaithful to my wife" and gave no detail.

Colette Kress — Nvidia's finance chief since 2013 — is interim CEO while the board searches for a permanent replacement, inside or outside the company. Harvey C. Jones became the board's first-ever chair.

Sources: Nvidia press release · AP · NPR · CNBC · CNN · Reuters · Euronews

2.The "circular financing" story — real, but not what the headlines said

The Wall Street Journal reported on 12 November that Nvidia had been investing in AI startups and smaller cloud providers that then used the money to buy Nvidia chips and build data centres. That is vendor financing, and it is not new or hidden: Nvidia's own quarterly filings disclose guarantees and credit support for customers.

The report describes an August 2026 meeting in which CFO Kress argued for scaling back those commitments — the OpenAI arrangement had been floated at roughly 8 gigawatts of compute, up to $250 billion, and was cut to about 4.25 gigawatts, up to $125 billion. Huang is said to have reacted badly to the pushback, and to have acknowledged privately that big cloud customers were drifting toward their own custom chips, leaving Nvidia with a committed volume of TSMC wafers and memory "coming at him like an avalanche".

So: the financing is real and worth watching. The conspiracy version — Nvidia quietly manufacturing the shortage — isn't supported by anything published. Nvidia hadn't responded publicly to the report as of 14 November.

Sources: claim-by-claim analysis · TechPowerUp · Wccftech · Forrester

3.The part that really sets prices: memory

Forget the boardroom. Graphics cards are expensive because the memory on them is scarce, and that is an AI-data-centre problem, not an Nvidia-management problem.

• Memory contract prices have risen every quarter of 2026 — roughly +93–98% in Q1, +58–63% in Q2, +13–18% in Q3 and a forecast +10–15% in Q4 (TrendForce). Slower growth on a base that has already trebled is not a discount.
• JP Morgan expects DRAM prices to be up about 400% by the end of 2026 against 2024.
• AI data centres are consuming roughly 70% of global DRAM this year. High-bandwidth memory for AI chips is set to be ~30% of all DRAM output in 2027, up from 20%.
• Most of next year is already sold. Micron has sold about 75% of its 2027 HBM output; Samsung nearly 80% of its 2027 memory, much of it on five-year contracts.
• Micron: conditions will be "much tighter in calendar 2027 and 2028" than 2026. SK Hynix's CEO: 2027 will be "the worst year in the industry's history from the supply perspective". Its chairman: four to five years to build enough wafer capacity, with production ~20% behind demand until at least 2030.
• Nvidia itself told investors in August that supply would be the main bottleneck until at least the end of its 2028 financial year, and flagged rising memory costs eating into margins.

Sources: Engadget · TrendForce/Micron round-up · Team Group CEO warning · the numbers

4.What cards cost right now

• The RTX 5090 launched in January 2025 at $1,999. Per Tom's Hardware's tracker it now sells for about $4,900 above that.
• In Germany, RTX 50-series prices are up 66% on average, and the 5090 by 132%.
• Nvidia has raised prices three times in 2026, by 20–30% on some cards; AMD lifted Radeon RX 9000 prices by up to 20%.

Sources: Engadget · eteknix tracker

5.Does the resignation change any of that?

Most of the price forecasts out there were published in September, October or early November — before Huang went — but that matters less than it sounds, because they are driven by memory supply and AI demand rather than by who runs Nvidia. The commentary that does account for the change is fairly consistent:

• Nothing changes for consumers in the near term. Nvidia's GPU allocation is largely fixed, and data centres remain the priority over gaming. — Avi Greengart, Techsponential
• The consumer cost problem is memory and storage, and Nvidia "doesn't play a direct role there at all". — Greengart again, on the same point
• Possible upside: if Nvidia is weakened, AMD gains share and competition improves for buyers. — Louise Wooldridge, Ampere Analysis
• Possible downside: a finance-led interim chief is more likely to focus on data-centre customers, meaning fewer new consumer initiatives. — industry reaction summarised by Yahoo Finance
• The next GeForce generation has reportedly slipped to 2028.
• Bloomberg Intelligence expects little change to Nvidia's 2027 and 2028 earnings estimates; Nvidia's board insists operations and the product roadmap continue uninterrupted.

Read it this way: the resignation is a stock-and-strategy story, not a graphics-card-price story. The one thing that would genuinely bring prices down is a slowdown in AI spending — which is also the thing the resignation has people worried about.

Sources: Yahoo Finance analyst round-up · analyst reactions · Euronews

6.Will you be able to buy a card in the next couple of years?

Yes — but not cheaply, and probably not in 2027.

• No forecast currently published shows prices falling in 2027. IDC: "no relief to the memory shortage before the end of 2027."
• New memory factories only ramp in the second half of 2027, with meaningful output in 2028.
• The most-cited model has retail RAM peaking around February 2027, staying near that level through 2027, then declining — back to normal around 2029 in the base case, and late 2028 even if AI spending is cut sharply.
• Waiting a few months does not help: early 2027 is projected to be higher than today.
• When memory does ease, GDDR7 costs could fall 15–35% from the peak by 2028–29 — but high-VRAM card prices may not follow all the way down.

Practical version: if you need a card now, buy sooner rather than later, because the next six months look worse, not better. If you can wait, 2028 is the earliest realistic point where things meaningfully improve, and that's the same answer whether or not Nvidia has a new boss.

Sources: price-outlook model · IDC/GDDR7 analysis · TrendForce/Micron