One Fortune, Explained: Jensen Huang

The Million Dollar Question: In 2012, Jensen Huang moved a block of Nvidia stock into an irrevocable trust for his heirs. What was that block worth on the day he moved it?
A) About $7 million B) About $70 million C) About $700 million D) About $7 billion

Read on for the answer.

The second post in this series looked at Warren Buffett, a fortune held in one asset for sixty years and disclosed in unusual detail because its owner chose to narrate it.

This one is nearly the opposite case. Jensen Huang narrates very little. There is no annual letter, no published giving pledge, no signature on the Giving Pledge, no document in which he sets out what he intends the money to do. What exists instead is a paper trail — Section 16 filings, a proxy statement, a foundation’s tax return — and it turns out to be more revealing than the quiet suggests, because the structures it describes are visibly still being built.

So this piece does not ask what he is worth. It asks what he owns, where each block sits, what the company pays him, what has already left his estate, and what moved this year. Documents only.

What he actually owns

The clearest single picture is Huang’s Form 4 for 17 June 2026, because Section 16 requires every pot to be listed separately. Ten lines:

Held by Shares
Directly 70,146,252
Jen-Hsun & Lori Huang Living Trust (1995) 468,131,547
The Huang Irrevocable Remainder Trust (2016) 109,040,602
The Huang 2012 Irrevocable Trust 31,421,011
TARG S2, M2, S3, M3 LLC (30,000,000 each) 120,000,000
TARG S1 and M1 LLC (6,632,667 each) 13,265,334
Total 812,004,746

At Nvidia’s close on Monday 24 August 2026 — $208.48 a share — that is $169.3 billion, and 3.34% of the company.

One qualification belongs here rather than in a footnote. About 140 million of those shares — the 2012 trust and the Remainder Trust, $29 billion between them — sit in irrevocable trusts he did not set up for himself. He is a trustee of both. Calling the whole 812 million “his” is the convention every wealth ranking uses, and it is doing some work.

Nvidia’s 2026 proxy statement counts it differently again. Measured on 23 March 2026, it puts him at 870,604,104 shares, or 3.58%. The gap is mostly the 58,683,830 shares held by the Jen-Hsun & Lori Huang Foundation, on whose board he and his wife sit. SEC rules make him report those; the same footnote adds that “Mr. Huang and his wife have no pecuniary interest in the Huang Foundation’s shares.” They are his to vote and not his to spend, which is why the Form 4 — which reports pecuniary interest — leaves them out.

Two other numbers from that page matter more than the total. Nvidia had 24,312,141,810 shares outstanding in March 2026. And the two largest holders on the list are not Huang: BlackRock at 7.43% and Vanguard at 7.31%, each roughly twice his stake. (BlackRock’s figure comes from a filing dated to end-2023, so treat it as approximate; the ordering is not in doubt.)

This is the structural fact that separates Huang from most founder fortunes of comparable size. There is no dual-class stock at Nvidia, no super-voting founder share, no pyramid of holding companies, no family voting agreement. He votes what he owns. His position rests on the board re-nominating him each year, which it does, listing him as the sole non-independent director among ten. A fortune this size that comes with no control mechanism attached is rarer than it sounds.

What Nvidia pays him

The Summary Compensation Table is short, and reading it straight after the ownership table is a small exercise in scale.

Fiscal year 2026 2025 2024
Salary $1,497,627 $1,486,199 $996,514
Stock awards $24,800,511 $38,811,306 $26,676,415
Non-equity incentive $6,000,000 $6,000,000 $4,000,000
All other compensation $4,045,691 $3,568,746 $2,494,973
Total $36,343,830 $49,866,251 $34,167,902

There is no bonus column, because Nvidia’s table does not have one. The $6 million is a variable cash plan payout against a $3 million target, doubled because fiscal 2026 revenue came in at $215.9 billion against a stretch goal of $160 billion. Any account describing a Huang “bonus” is describing this line.

All of it is small against the stake. Total 2026 compensation is 0.021% of the shares he holds; the salary alone is 0.0009%. On the Monday this piece prices — a day Nvidia fell 2.91% ahead of an earnings report — his holding moved by roughly $5.1 billion, about 3,400 years of salary, in six and a half hours. That is what “one asset” means in practice, and it is why compensation is close to irrelevant to fortunes shaped this way.

The $4,045,691 of other compensation is almost entirely one item. The proxy breaks it out: $3,975,533 for “residential security, security during personal travel, and consultation fees and driver services,” plus a $11,500 401(k) match and $21,904 in life insurance premiums, both available to every eligible Nvidia employee. The board’s stated reason is “the exceptionally high public profile and prominence of our CEO,” and the company adds that it does “not consider these security arrangements to be personal benefits.”

The aircraft line is worth quoting, because it is the one most companies fudge: “To the extent Mr. Huang has personal travel on NVIDIA-arranged private aircraft, he fully reimburses the Company for associated incremental costs. As a result, we incur no incremental costs from Mr. Huang’s personal air travel.” Zero, in a category where peer chief executives routinely disclose seven figures. “In Fiscal 2026, we did not provide any other perquisites or personal benefits to our NEOs.”

The pay ratio does not behave the way the genre expects. Nvidia’s median employee earned $282,050 in fiscal 2026 — an extraordinary figure, and a reminder that the ratio measures a workforce as much as a chief executive. Huang’s ratio is 129:1, low for a company this size, and low mostly because of the denominator.

One more line, easy to skip. Nvidia’s insider trading policy prohibits covered persons from “purchasing NVIDIA stock on margin or holding NVIDIA stock in a margin account, or pledging NVIDIA stock as collateral for a loan.” That is narrower than it first looks: the policy applies to “employees, board members, contractors and consultants,” says nothing explicit about shares held by trusts or LLCs, and expressly allows exchange funds “on a case-by-case basis if the fund is broadly diversified.” It is a company rule, not a law. But it does mean the most familiar route by which American founders turn stock into cash without selling — pledging the shares — is one Nvidia has closed off in its own securities, at a company where the chief executive sets the tone.

The 2012 trust, and the $7 million that matters

That is the answer to the question at the top: about $7 million.

In 2012, as Jesse Drucker reported in the New York Times in December 2024, Huang moved 584,000 Nvidia shares — worth roughly $7 million at the time — into an irrevocable trust for his heirs. Drucker’s sources could not confirm the exact structure from the filings; he reported that multiple experts said it was “almost certainly” the gift-loan-sale arrangement estate lawyers nickname an “I Dig It,” after the intentionally defective grantor trust at its centre. The filings call it The Huang 2012 Irrevocable Trust. It holds 31,421,011 shares today, worth $6.6 billion.

The split maths gets you most of the way. Nvidia split four-for-one in July 2021 and ten-for-one in June 2024, so 584,000 shares became 23,360,000 — about two-thirds of the present balance, with the rest reflecting further activity the filings do not narrate. The order of magnitude is the point. Roughly $7 million of 2012 stock is $6-plus billion of 2026 stock, and the growth happened outside his taxable estate.

The federal estate tax is a flat 40% above an exemption of $15 million per person in 2026. Applied to that trust’s current value, the tax the block will now never generate is on the order of $2.6 billion. That is arithmetic on published figures, not a number the Times or the filings state.

“Intentionally defective” is a term of art and a slightly absurd one. The trust is deliberately drafted so its assets sit outside the grantor’s estate for estate tax purposes while its income stays taxable to the grantor for income tax purposes. The mismatch is the feature: the grantor pays the trust’s income tax, which shrinks his own estate further, and the payment is not treated as an additional gift. It is a transfer the tax code declines to count.

None of this is unusual for fortunes at this level, which is exactly why it is worth writing down. The mechanism is ordinary. Only the multiple is not.

What a GRAT does, and why 2016 was the year

Four years later, Huang did something larger.

A grantor retained annuity trust is the simplest of the big estate-planning instruments. You put an asset into a trust for a fixed term and keep the right to a fixed annual payment out of it. The IRS values that retained right using a published discount rate — the Section 7520 rate, set monthly at 120% of the applicable federal midterm rate. Set the annuity so its present value matches what you put in, and the taxable gift is nominally zero. If the asset then grows faster than that rate, everything above the hurdle passes to the remainder beneficiaries free of gift and estate tax.

Two consequences follow, both in the standard practitioner literature. A failed GRAT costs almost nothing — it simply hands the asset back. And if the grantor dies during the term, the property returns to the estate under Section 2036(a). A capped downside and a mortality risk.

The hurdle rate is everything. The Huang Irrevocable Remainder Trust is dated 19 February 2016, a month when the Section 7520 rate was 2.18%. (Rates went lower still later — under 0.5% in late 2020 — so 2016 was cheap rather than uniquely cheap. In August 2026 the same rate is 5.2%.)

Nvidia stock cleared a 2.18% hurdle by a margin that makes the word “hurdle” comic. Drucker reported that four 2016 GRATs held just over three million pre-split Nvidia shares, worth about $100 million at the time, and put the family’s likely estate-tax saving from that vintage alone at roughly $6 billion. Three million pre-split shares, adjusted for the same two splits, is 120 million today. The Remainder Trust now holds 109,040,602$22.7 billion at Monday’s close. The filings do not say the two sets are the same shares, and the Remainder Trust plainly received stock at more than one point.

The machine is still running

The most interesting thing in the filings is not history. It is that the structure is under construction right now, and two Form 4s from this year show it happening.

March 2026. The Form 4 for 18 March reports two entities — the Lori Lynn Huang 2016 Annuity Trust II and the Jen-Hsun Huang 2016 Annuity Trust II — transferring 29,481,301 shares each, 58,962,602 in total, to The Huang Irrevocable Remainder Trust “upon termination of the Grantor Retained Annuity Trust.” Each also paid 30,884 shares back to the Living Trust to satisfy annuity payments, and two of the TARG LLCs transferred 3,367,333 shares each for the same purpose, which is why those two dropped from 10,000,000 shares to 6,632,667.

That is a GRAT completing, on paper, in public: the annuity paid, the remainder — $12.3 billion at today’s price — passing to the trust that holds it for the next generation. Two cautions. The filings never state when these trusts were created or how long their terms ran; the only date is inside the name, and the “II” suggests they were themselves successors to something. And the Remainder Trust already held 50,078,000 shares before March, so this was one instalment of the 2016 vintage, not all of it.

June 2026. Three months later, the Form 4 for 16 June shows the Living Trust falling from 528,531,547 shares to 468,131,547. Two things account for the 60,400,000 difference. A footnote records 30,000,000 shares contributed to each of TARG S3 and TARG M3 — two LLCs that had never appeared in a filing before. And a single reported transaction, coded G for gift, disposes of 400,000 shares: “266,670 shares to The Jen-Hsun & Lori Huang Foundation, a 501(c)(3) charitable organization, and 133,330 shares to a donor-advised fund.”

Note what is absent. In all of calendar 2026 Huang has filed four Form 4s, none reporting an open-market sale, every one with the Rule 10b5-1 affirmative-defense box unchecked. His last selling plan was adopted 20 March 2025 for up to 6,000,000 shares and expired on 31 December 2025; the Form 4s filed under it total exactly 6,000,000 shares for $1.05 billion. Nvidia’s quarterly filings disclose no replacement — though they only run through 26 April 2026, so a later plan could exist and not yet be public.

One naming observation, offered as an observation. TARG is GRAT spelled backwards. No filing and no published source explains the name, and none says what the S and M denote. But six LLCs sitting under a living trust, funded in matched pairs of identical size, appearing alongside the annuity trusts, is a pattern — and it is the same pattern the 2016 vintage has already run to completion once.

The foundation, and the fund inside it

The philanthropy is where the numbers stop being about tax and start being about intent, and where they are hardest to read from a single year.

The Jen-Hsun & Lori Huang Foundation was set up in 2007 with 370,000 Nvidia shares then worth $12.6 million. Its most recent public return, the Form 990-PF for the year ended 31 December 2024, reports $9.21 billion in total assets, up from $3.41 billion a year earlier — around the twentieth largest private foundation in the United States, and larger than the MacArthur Foundation. Lori Huang is president and Jensen Huang is secretary and chief financial officer. Neither takes a salary; the return records compensation costs of zero.

It disbursed $126.3 million in 2024. MacArthur, on a similar endowment, granted $356 million.

And of that $126.3 million, $82.5 million — 65% — went to DAFgiving360, the donor-advised fund sponsor formerly called Schwab Charitable. Named recipients took the rest, the largest being $30.5 million to Oregon State University and $10.8 million to Crisis Text Line.

This is the part worth understanding rather than merely disapproving of. A private foundation must pay out about 5% of assets each year. A donor-advised fund has no payout requirement at all. Money moved from the first to the second discharges the legal obligation on the day it moves and can then sit indefinitely. The Huangs’ account is called the GeForce Fund, after Nvidia’s graphics line, and Inside Philanthropy has listed the foundation among its “black box billionaires” for this pattern.

Three things cut the other way, and a piece that left them out would be misleading. The DAF share is falling — 83% in 2022, 77% in 2023, 65% in 2024 — while the absolute payout rises. The foundation has begun professionalising: Bloomberg reported in November 2025 that it hired its first senior staff, sold $620 million of stock to diversify, gave $22.5 million to California College of the Arts, and faces a required distribution of roughly $345 million for 2025 and $450 million for 2026 simply because the endowment keeps growing. And Nvidia’s own proxy discloses something the 990 does not: the foundation has been buying GPU compute from CoreWeave to give to universities and non-profit research institutes, “of which $108.3 million has been donated to date.” That is larger than any cash grant on the 2024 return, and it is a donation of the one thing the family is unusually well placed to supply.

In July 2026 the family pledged $75 million to Vanderbilt to establish a college of art, architecture and design in San Francisco. It is a matching gift from the Huangs rather than a disclosed foundation grant, and the university has until the end of 2028 to raise the match.

What people get wrong

That he sold the top. He did not sell at all in 2026. The 2025 plan ran its course, hit its 6,000,000-share ceiling and expired. His stake shrinks on some screens because shares are moving into trusts and LLCs, and different trackers count those differently — which is also why rankings disagree about him by tens of billions.

That the estate planning is exotic. The pieces are textbook: a grantor trust, a set of GRATs, a family foundation, a donor-advised fund. Any competent estate lawyer will describe all four before lunch. What is unusual is the asset. Put ordinary stock in a 2016 GRAT and you clear a 2.18% hurdle by a little. Put Nvidia in one and you clear it by a factor that turns a routine technique into a generational transfer.

That the low salary is modesty. It is arithmetic. When the position moves $5 billion on an unremarkable Monday, the compensation committee is noise. Buffett’s $100,000 and Huang’s $1.5 million are the same phenomenon: past a certain ratio of stake to salary, salary stops being pay and becomes a formality.

That he is running the buy-borrow-die play. Not with this stock, at least. Nvidia’s policy bars covered persons from pledging Nvidia shares as collateral or holding them in a margin account. Whatever tax efficiency this fortune has is not coming from borrowing against the position.

That the foundation is not giving. It gave $126 million in 2024, $108 million of GPU compute across an unspecified period, and is legally required to move roughly $450 million this year. The fair criticism is narrower and more specific: for years most of the payout went to a vehicle with no deadline of its own, so the reported giving overstated what reached operating charities. That gap is closing, under compulsion from the 5% rule rather than from any announcement.

That $8 billion is the saving. The Times figure covered the family’s savings across all the structures, estimated in December 2024 when Huang was worth about $127 billion. He is worth substantially more now, and the Remainder Trust alone holds $22.7 billion. Whether the true number is larger depends on assumptions nobody outside the family can check. Treat $8 billion as a floor from a specific date.

That any of this is hidden. It is in Form 4s a member of the public can read in ten minutes. Little has been written about it not because it is secret but because Section 16 filings are boring, and the interesting part — that one structure ran to completion in March and two more were funded in June — only appears if you read two of them side by side.

Bottom line

Buffett’s fortune has an end date, published in a press release, and a will he has promised will sit in a county courthouse.

Huang’s has a mechanism and no stated destination. The mechanism is legible and it is running: 812 million shares, 70 million of them held outright and the rest across a living trust, two irrevocable trusts and six LLCs; roughly $7 million of 2012 stock now $6.6 billion and outside the estate; a 2016 annuity trust that terminated this March and handed $12.3 billion to the trust below it; two new entities funded with 60 million shares in June; a foundation that was worth $9.2 billion at the end of 2024 and is now compelled to give away hundreds of millions a year whether or not anyone announces a plan.

What the documents do not contain is a statement of purpose. There is no letter and no pledge. That is not a failure of disclosure — nothing obliges him to explain himself, and he is under no duty to be Buffett. But it changes what the filings are evidence of. With Buffett, the compelled numbers and the voluntary narration point the same way, and the narration lets you test the plan against the arithmetic. Here there is only arithmetic.

And the arithmetic is unambiguous about direction. The money is moving down — into trusts, out of the estate, toward heirs — on a schedule that has now completed one full cycle in public and started another. The philanthropy is moving too, faster each year, but mostly because a 1969 tax rule says it has to.

Most writing about Jensen Huang is about the chips. The filings are about something slower and more consequential: a fortune being transferred, deliberately and in instalments, from one generation’s balance sheet to the next one’s, using instruments designed for exactly this and never before applied to an asset that behaved like this.


Methods and sources. Share counts by holding entity, the 400,000-share gift, the TARG S3 and M3 contributions and the March 2026 annuity-trust termination come from Jen-Hsun Huang’s Forms 4 for periods 18 March 2026, 16 June 2026 and 17 June 2026. Beneficial-ownership totals, the entity footnote, the Summary Compensation Table, the security and perquisite disclosure, the aircraft reimbursement language, the pay ratio, the median employee figure, shares outstanding, the BlackRock and Vanguard positions and the CoreWeave donation come from Nvidia’s 2026 proxy statement, filed 12 May 2026 and reporting ownership as of 23 March 2026; BlackRock’s stake is drawn there from a filing dated to end-2023. The pledging and exchange-fund provisions are quoted from the insider trading policy filed as an exhibit to the fiscal 2026 Form 10-K. The 2025 Rule 10b5-1 plan terms come from the fiscal 2026 first-quarter Form 10-Q; the $1.05 billion figure is the sum of the sale transactions reported on the Forms 4 filed under it, and the absence of a later plan was checked against the Item 408(a) tables in every subsequent 10-Q and the 10-K, which cover periods through 26 April 2026 only. The 2012 trust’s original size, the 2016 GRAT figures and the $8 billion estimate come from Jesse Drucker’s New York Times investigation of 5 December 2024; that piece is paywalled and the figures here are as reported. GRAT mechanics and the Section 2036(a) risk follow the McGuireWoods practitioner outline; Section 7520 rates are from the IRS tables. Foundation assets, disbursements and officer compensation are from the Form 990-PF for the year ended 31 December 2024, via ProPublica; the DAF share and the largest named grants are from Inside Philanthropy, January 2026; the staffing, diversification and required-distribution figures from Bloomberg, November 2025; the Vanderbilt terms from the university’s own announcement, July 2026. Prices are the Nasdaq close on 24 August 2026 and will be wrong by the time you read this; the share counts will not be.

This draft was fact-checked line by line against the filings before publication, and the check changed it in several substantive ways. It corrected the 2025 trading-plan proceeds from a widely repeated $713 million — which belongs to the 2024 plan — to $1.05 billion computed from the Forms 4; corrected the February 2016 Section 7520 rate to 2.18% and removed a claim that the August 2016 rate was near-record, since rates fell far lower in 2020 and 2021; corrected the foundation’s size ranking from fifteenth to roughly twentieth; corrected the Vanderbilt gift from a foundation grant to a family matching pledge payable through 2028; narrowed the description of the pledging ban, which applies to covered persons and Nvidia securities rather than to Huang personally and to borrowing generally, and which permits diversification through exchange funds; removed an assertion that the 2016 annuity trusts ran a ten-year term, which no filing states; separated his 3.34% pecuniary stake from the 3.58% he reports; and added the $108.3 million CoreWeave compute donation and the foundation’s rising required distribution, both of which cut against the original draft’s reading of the philanthropy and improved it. Two things here are explicitly observations rather than sourced claims, and are flagged as such in the text: that TARG is GRAT reversed, and the estate-tax figure applied to the 2012 trust. The individual 2024 grants below $10 million could not be independently verified beyond the Inside Philanthropy account, and reporting attributing the Times investigation to more than one author could not be corroborated.

Related reading: Trusts: How Wealth Is Held, Protected, and Passed On · Taxes: How Wealth Is Structured and Preserved · Generational Wealth: How Long Fortunes Actually Last · Philanthropy: Giving, Status, and Influence · One Fortune, Explained: Warren Buffett

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *