$NVDA filed its quarterly 13F on August 14, and the list of what it owns has been circulating since. It is a striking list: eight positions, $63.44 billion, up from $18.4 billion three months earlier.
Here it is, straight from the filing — accession 0001045810-26-000065, period ended June 30, 2026:
| Issuer | Shares | Value as filed |
|---|---|---|
| Intel | 214,776,632 | $29,989,261,126 |
| SpaceX | 122,764,805 | $20,975,594,582 |
| CoreWeave | 47,213,353 | $4,699,617,158 |
| Coherent | 7,788,161 | $3,072,195,870 |
| Nokia | 166,389,351 | $2,209,650,581 |
| Synopsys | 4,821,717 | $2,150,823,302 |
| Nebius | 1,190,476 | $328,773,757 |
| Generate Biomedicines | 833,325 | $14,058,193 |
| Total | $63,439,974,569 |
(SpaceX appears in the filing under its legal name, Space Exploration Technologies Corp.)
Those figures are exactly right, and they are also the least interesting thing in the document. A list of holdings tells you what a portfolio is. It cannot tell you what the manager did. For that you need last quarter's filing next to this one, and the two together say something no single-quarter list can.
Nvidia bought one thing
| Issuer | Shares Q1 | Shares Q2 | Change |
|---|---|---|---|
| Intel | 214,776,632 | 214,776,632 | unchanged |
| CoreWeave | 47,213,353 | 47,213,353 | unchanged |
| Coherent | 7,788,161 | 7,788,161 | unchanged |
| Nokia | 166,389,351 | 166,389,351 | unchanged |
| Synopsys | 4,821,717 | 4,821,717 | unchanged |
| Nebius | 1,190,476 | 1,190,476 | unchanged |
| Generate Biomedicines | 833,325 | 833,325 | unchanged |
| SpaceX | 0 | 122,764,805 | new |
Seven of the eight share counts are identical to the March filing. Not similar — identical, to the share. Nvidia did not add to Intel, did not trim CoreWeave, did not touch Nokia or Synopsys or Nebius. In a quarter that reads like $45 billion of aggressive dealmaking, the company made exactly one move.
Here is where the $45.1 billion actually came from:
| Source | Amount |
|---|---|
| New position (SpaceX) | +$21.0B |
| Price appreciation on holdings it did not touch | +$24.1B |
| Actual buying of anything else | $0 |
More than half the increase is the market, not Nvidia.
Intel is most of the rest
One line does the heavy lifting. Nvidia held 214,776,632 Intel shares on March 31 and the same 214,776,632 on June 30. The implied price moved from $44.13 to $139.63 — up 216% in a single quarter.
That one price move added $20.5 billion to the disclosed portfolio, which is more than the entire portfolio was worth two quarters ago.
| Issuer | Implied $/share Mar 31 | Implied $/share Jun 30 | Move |
|---|---|---|---|
| Intel | 44.13 | 139.63 | +216% |
| Nebius | 103.76 | 276.17 | +166% |
| Coherent | 238.21 | 394.47 | +66% |
| Nokia | 8.04 | 13.28 | +65% |
| CoreWeave | 77.47 | 99.54 | +28% |
| Synopsys | 396.48 | 446.07 | +13% |
Every one of those is a price the market set, on a position Nvidia left alone.
The SpaceX line is not what "new position" usually means
The $21 billion looks like the quarter's big purchase. It probably is not a purchase at all.
SpaceX listed on June 12, 2026 — eighteen days before the quarter ended. A 13F reports 13(f) securities, which broadly means US-listed equities. A private holding is invisible; the moment it lists, it appears. A stake that shows up as "new" in the first quarter after its IPO is far more likely to be a position that became reportable than one that became owned.
We cannot settle that from this filing alone — the 13F has no cost basis and no acquisition date, and a first-time filer's position looks the same whether it was bought last week or held for years. But it is the difference between "Nvidia spent $21 billion" and "Nvidia's existing stake became visible," and the filing does not support the first reading.
The figure was already stale on the day it was published
A 13F is a snapshot of June 30, published 45 days later. By the time the list was circulating, it described a portfolio six weeks in the past. Prices moved in between.
Using our own last price refresh, August 3:
| Issuer | Implied Jun 30 | Aug 3 | Move |
|---|---|---|---|
| CoreWeave | $99.54 | $107.01 | +7.5% |
| Intel | $139.63 | $140.94 | +0.9% |
| Nebius | $276.17 | $274.38 | −0.6% |
| SpaceX | $170.86 | $116.00 | −32.1% |
Holding the four unpriced names flat, the $63.44B portfolio was worth about $57.3B five weeks later — down roughly $6.1B, or 9.6%. Essentially all of it is SpaceX, whose $21.0B mark had become about $14.2B.
The headline number was never wrong. It was just never current, and the single largest new line is the one that moved most against it.
What the list leaves out entirely
Two things.
A 13F is not a portfolio. It covers 13(f) securities only. Nvidia's private positions — the AI labs, the robotics and model companies it has backed — do not appear here at all, at any size. "$63.4B" is not what Nvidia owns; it is the listed slice of what Nvidia owns, and it is a floor rather than a total.
The exits do not appear either. Compare far enough back and the churn is real. In the September 2025 quarter Nvidia also held Arm Holdings, Applied Digital, Recursion Pharmaceuticals and WeRide. By December all four were gone:
| $M | 2025 Q2 | 2025 Q3 | 2025 Q4 | 2026 Q1 | 2026 Q2 |
|---|---|---|---|---|---|
| Total disclosed | 4,333 | 3,844 | 13,105 | 18,374 | 63,440 |
| Positions | 6 | 6 | 5 | 7 | 8 |
The portfolio went from $4.3B to $63.4B in four quarters — a genuinely enormous change, and a real story. It happened in two steps that a single-quarter snapshot hides: a clear-out of the four small positions alongside a $7.9B Intel entry in Q4 2025, then Coherent, Generate Biomedicines and a CoreWeave stake nearly doubled from 24.3M to 47.2M shares in Q1 2026.
Why it matters beyond the trivia
Three of the eight are Nvidia's own customers or suppliers, and that is the part worth arguing about.
CoreWeave and Nebius buy Nvidia GPUs. We have covered both closely — CoreWeave's Q2 and Nebius's Q2 — and between them they committed billions of dollars of capex to Nvidia hardware in the same quarter this filing covers. Nebius alone spent $5.66 billion on property and equipment in Q2, against a $328M Nvidia stake. Nvidia's holdings in its own customers are small relative to what those customers spend with it, which is the honest way to size the circularity question rather than assert it.
Synopsys and Coherent sit upstream — EDA software and optical components — and Intel is now simultaneously a competitor, a potential foundry partner, and Nvidia's largest disclosed equity position.
Against Nvidia itself, though, the whole book is small. Trailing twelve-month revenue is $253.5B and trailing free cash flow is $119.1B. The entire $63.4B disclosed portfolio is:
- 25.0% of one year of revenue
- 53.3% of one year of free cash flow
- 1.3% of a ~$4.9T market capitalisation
A company generating $119B of annual free cash flow holding $63B of listed equity is not a hedge fund with a chip business attached. It is a chipmaker with about six months of free cash flow parked in eight names, one of which tripled.
What would change this read
- A cost basis for SpaceX. If Nvidia did put new money to work at the IPO, the "it bought one thing" framing understates the quarter substantially. The 13F cannot tell us; an S-1 or a subsequent disclosure might.
- The Q3 filing, due mid-November. It is the first that will show whether Nvidia acts on any of these — the current filing shows a manager sitting still while its book repriced.
- Intel's next two quarters. A 216% move on an untouched position is the single largest contributor here. If it reverses, the portfolio shrinks by more than the entire SpaceX line without Nvidia doing anything at all.
The number to watch is not the total. It is the share counts — the only column in a 13F that records a decision.
