Family Offices Cut Nvidia as Compute Share Slips to 28%

A single line in a quarterly filing says the Soros family office sold about 206,000 Nvidia shares in the third quarter. It is not a large position by the standards of the fund’s overall book. It is one of twelve.

Filings made public this week show at least a dozen family offices reduced their Nvidia holdings in the same period, with Blue Crown Capital among those trimming exposure. Family offices are not usually read as a single signal, since their mandates, tax positions and time horizons differ. When a dozen move the same way in the same quarter, the pattern is worth a second look.

The more pointed number came from JPMorgan. Among the compute purchases that frontier AI laboratories have disclosed, Nvidia’s share has fallen to 28%, the bank said, with Broadcom, AMD, Amazon and Google absorbing the rest.

That figure describes a different business from the one investors usually discuss. Nvidia’s revenue still comes overwhelmingly from selling accelerators. The 28% measures the disclosed contracts under which laboratories buy compute, and it moves when a lab signs with a custom-chip vendor, builds on a hyperscaler’s own silicon, or splits an order across suppliers.

The direction of travel is what matters. Frontier laboratories have spent two years trying to reduce dependence on a single supplier, not because the chips are poor but because allocation, price and delivery schedules were controlled by one company. Custom accelerators designed with Broadcom, AMD’s competing parts, Amazon’s in-house chips and Google’s tensor processors all exist to give buyers a second option.

For Nvidia the risk is not the loss of a single contract but the loss of the pricing power that comes with being the only credible choice. A supplier holding 28% of disclosed demand still sells an enormous quantity of hardware. It no longer sets the terms alone.

The share price has already priced in some of that. Nvidia has fallen more than 25% since Huang resigned, a decline that reflects both the governance uncertainty and questions about demand that were building before it. Reports that he might return, and a board position that has not been made clear, have kept the stock trading on headlines rather than on earnings.

Family offices behave differently from hedge funds in ways that make their selling less informative about short-term views. Many hold concentrated positions for decades, sell to fund a specific obligation, or move shares into trusts and foundations. A reduction in one quarter can mean conviction, or it can mean a liquidity need, and the filings do not distinguish.

The disclosed-compute measure carries its own limits. It counts what laboratories have chosen to make public, which is a fraction of what they buy, and it weights contracts rather than volumes. A lab that signs a small custom-chip deal and a large accelerator order can move the percentage without changing its actual dependence.

Both data points still point the same way. Investors with long horizons are trimming, and buyers with technical teams are diversifying. Neither group is selling because the company is in trouble. Both are acting because the position became large enough, and the alternatives good enough, to justify spreading the risk.

There is a mechanical reason family offices sell a stock like this one. A holding that has multiplied several times over in a few years becomes a concentration problem for a portfolio that is meant to fund a family across generations. Trimming it is a risk decision rather than a view on the next quarter, and the trimming usually continues for years.

Analysts said the next set of filings will matter more than this one, because they will cover the quarter in which the resignation occurred. If family offices cut again while the share price was falling, the selling was conviction. If they held, the third-quarter reductions were housekeeping.

What the filings cannot show is what happens to the 28%. A supplier that keeps 28% of disclosed demand and most of the installed base still sits at the centre of the market. A supplier that falls to 28% and keeps falling is a different company, and the difference is decided by contracts that will not be public for a year.

The board’s position on the founder adds a second uncertainty that has nothing to do with chips. A company making a generational transition in its customer base is also making one in its leadership, and the two decisions interact. Buyers deciding whether to commit to a multi-year supply agreement want to know who will be running the supplier.

For the family offices, none of that changes the arithmetic. They held a stock that had become a large share of their portfolios, and they reduced it in a quarter when they were required to say so. The interesting part is not that they sold. It is that twelve of them did it at once, and that the buyers they are making room for have names the market did not use to say two years ago.

Related Posts

  • November 20, 2026
  • 12 views
Sandisk Completes Eight-for-One Split as NAND Outlook Darkens

Eight for one. Sandisk’s stock split took effect on Nov. 19, and the shares began trading on the adjusted basis on Nov. 20, a change that divides the price by…

  • November 20, 2026
  • 11 views
A Lawmaker’s Privacy Suit Puts Model Design on Trial

For three years British courts have sidestepped the question of whether the company that builds an AI model answers for what its users produce with it. A written defence filed…