Cathie Wood's New Stock Picks (September 2026) — and the Catch

Cathie Wood’s newest disclosed trades are dated Thursday 3 September 2026, and the whole file is four rows. One of them is a buy.

That is the answer, and unlike most people’s stock picks it is not an opinion — ARK publishes what it traded, every trading day, with share counts. What follows is the September file, then the holdings behind it, then August and July for the contrast, and then the part almost nobody writes: how these numbers get misread, including by outlets repeating them right now. It is not investment advice.

What are Cathie Wood’s newest stock picks?

One buy, and the same micro-cap swept out of three funds.

That is not an excerpt. That is the entire file — four rows, four funds, one name on the sell side. ARKQ and ARKX disclosed nothing at all.

Two things make it worth more than its size. The first is what the percentages mean: ARK gives each trade as a share of the fund it happened in, and the largest of these is 0.02% of ARKG. The Brera sales are smaller than a rounding error — on ARKK’s own numbers, 6,591 shares at 0.0004% of a $6.42 billion fund is a sale of roughly $26,000.

The second is that Brera is gone from ARKK’s and ARKW’s current holdings files entirely. These were not trims. They were the last of it being swept out, which is why the share counts are so odd — 6,591 and 2,610 are what is left at the bottom of a position, not what anyone sets out to sell.

The third fund is a good illustration of the trap in the last section. ARKF also sold Brera that day, and there is no way to confirm from ARK’s files that ARKF’s position is gone — because the holdings file ARKF is currently serving is dated 2 January and still shows 1,018,952 shares. So this page says the position closed in two funds, not three.

Set that against the month before. The 3 August file ran to 46 rows across five funds. The September file has four rows across four. On the disclosed record, ARK has gone quiet.

One caution the file itself insists on, and it cuts against reading too much into that: ARK’s disclaimer says these files are “not comprehensive lists of a day’s trades”. “ARKQ did not trade” can only ever mean “ARKQ disclosed no trades.”

What does ARK actually hold now?

The trades are the noise. This is the position — ARKK’s own holdings file, dated 4 September and priced at the previous close.

43 holdings, $6.42 billion, and the top ten are 50.2% of it. The bottom ten come to 5.8% between them. Whatever ARK’s daily trade file looks like on any given day, the fund is a concentrated bet on about a dozen names, and half of it sits in ten.

ARKK’s NAV was $87.14 on 3 September, a day it rose 4.44%.

What did ARK buy and sell in August 2026?

A month earlier the file looked nothing like this. On 3 August, ARK bought Amazon in every one of the five funds that traded — and sold half a million shares of Roblox.

That is the pick of it, not the whole file: ARK’s 3 August disclosure runs to 46 rows across 27 name-and-direction pairs in five funds. Also traded that day and not listed above: DoorDash bought (17,314), Strata Critical Medical sold (32,846), Komatsu sold (26,858), Kodiak AI bought (21,516), Iridium sold (19,528), Snowflake sold (18,729) and Natera sold (18,060), among others.

The Amazon buy is the tell. It is the only name touched by every fund that traded that day, and it comes days after ARK was selling it — 9,895 shares sold on 28 July and 4,902 on 29 July, then 73,835 bought back on 3 August. Amazon closed at $230.86 on 28 July and $284.02 on 3 August, so ARK sold into the low and bought roughly five times as much back after a 23% move. Whatever that is, it is not a long-held conviction being topped up.

The Roblox sale needs its own footnote, because the obvious reading is wrong. ARK did not do the bulk of its Roblox selling ahead of the collapse — about 85% of it came after. The disclosed sells are three dates: 69,135 shares on 28 July, 22,586 on 29 July, then 505,751 on 3 August. Roblox closed at $48.67 on 30 July and $35.60 on 31 July, a 27% drop in one session. About 85% of ARK’s selling came after that.

One fund is missing from the table entirely: ARKG disclosed no trades at all on 3 August. It is the one fund buying anything on 3 September — a reversal you would miss entirely reading either file on its own.

What did ARK buy and sell in July 2026?

ARK made 673 disclosed trades across 22 trading days in July. The biggest destination was SpaceX at about $183 million net — roughly a quarter of the month’s disclosed buying — then Tesla about $111 million, Circle about $60 million, Meta about $58 million and X-Energy about $45 million. On the sell side it disclosed a Strata Critical Medical sale on every one of the 22 days, and ARKK sold out of Figma entirely.

Those dollar figures value every disclosed trade at that day’s close. Gross disclosed buying for the month was around $760 million, so SpaceX took about 24% of it — the largest single destination by a distance, and more than Tesla and Circle combined.

Volume is a different question from conviction, and persistence says more than dollars. ARK bought 35 names in July without ever selling them and sold 41 without ever buying; these are the ones it touched on the most separate days:

Two structural moves sit inside that noise. ARKK exited Figma — it held 1,535,688 shares in mid-July and the name is gone from its August file, sold down through 23, 24, 29, 30 and 31 July, the last tranche a tail-end 96 shares. ARKW did not follow: it trimmed twice, on 23 and 24 July, and still holds 797,927 Figma shares, 1.20% of that fund. And it opened a new position in Scribe Therapeutics, 361,253 shares into ARKG on 24 July in a single buy, straight in at 0.34% of the fund.

The single largest trade of the month relative to a fund was ARKG buying 14,637 shares of Eli Lilly on 8 July — at 1.04% of the ETF, the only trade all month that exceeded 1% of a fund.

How did July’s picks actually do?

Dating a call is only worth something if somebody checks it later, so here are three from the past two months, marked from the day of the call to the 3 August close.

That AMD line is worth sitting with, because the obvious version of it is wrong. AMD is up 126% in 2026 — but essentially all of that was in place before ARK sold a share. Measured from its first trim on 1 June, AMD is down about 5%, and the 3 August close sits roughly 8% below ARK’s share-weighted average sale price of about $526. The sell decision looks better on the record than it does in a year-to-date table, which is exactly the sort of thing dated trades let you check.

What is fair to say is that “out” was the wrong word. We described the chip move in July as a rotation — Nvidia in, AMD out — and ARK still owns 1,054,886 AMD shares across five funds, including ARKW’s second-largest holding.

The single-day forensics behind two of those calls are in our earlier pieces on ARK buying Meta into a 7.95% drop and on the June AI dip-buying spree .

The conviction call: SpaceX

Up for a third consecutive month, and now 6.28% of ARKK.

The share count is the line to read, because it is the one price cannot move. It is up 269,069 shares, or 10.7%, over the month.

That rise is not proof of buying on its own — as the last section explains, holdings files also move with fund creations, and ARKK grew over the month. But the weight rules most of that out. SpaceX’s implied price in the file is $149.74, up 30.7% from the $114.53 it closed at on 3 August, against ARKK’s own NAV rising 18.5% over the same stretch. A position that simply sat there while both moved would have drifted to roughly 5.5% of the fund. It is at 6.28%. The gap is small, and it points one way.

If one position carries more weight than the rest, it is SpaceX, which began trading on an exchange on 12 June 2026 — we covered the $163 million ARK sold to fund that IPO purchase at the time. ARK bought it on 6, 7, 8, 10, 13, 15, 17, 20, 22, 27, 28 and 29 July — every single transaction a buy, not one sell — for about $183 million net.

It took second place in the fund during July, displacing Tempus AI, and has held it since.

Wood has not been subtle about why. On FOX Business on 22 July she said of the company: “We think this could become the most important company in history, and I mean in global history.”

And here is the honest scoreboard, which is the part a picks roundup usually leaves out. SpaceX closed at $114.53 on 3 August, about 46% below its record close of $211.39 on 16 June — and ARK was buying most of the way down. A month later the file implies $149.74, which is roughly 29% below that June record. So the position has recovered a good deal of ground without regaining all of it, and ARK’s average cost is not public, so this page is not going to translate that into a profit or a loss.

Where do these picks come from?

They come from ARK’s daily trade file — fund, direction, ticker, share count and the trade as a percentage of the ETF, published after every close. That is a different document from ARK’s daily holdings CSVs, its quarterly 13F and its 13G filings, and knowing which one you are reading matters more with ARK than with almost any other manager, because the four answer four different questions.

• The daily trade file is the closest thing to a live pick list: fund, direction, ticker, shares, and the trade as a percentage of the ETF, posted after each close. It is what this article’s dated trades come from.
• The daily holdings CSVs show positions, not activity — and they are priced at the previous trading day’s close, so a file dated 4 August carries 3 August prices.
• 13F filings are quarterly and lag by 45 days. ARK’s most recent covers 31 March 2026; the June-quarter filing is not due until mid-August, so any article claiming “ARK’s latest 13F” right now is describing a stale one.
• 13G filings start when a stake passes 5% of a company and are amended as it changes — including on the way back down. ARK filed three amendments on 8 July, all with a 30 June event date, and two of them report stakes already below the line: Strata Critical Medical at 3.16% and Canton Strategic Holdings at 4.74%.

ARK’s own disclaimer, printed inside the trade file, is the limit on all of it: the files are “not comprehensive lists of a day’s trades” and exclude initial and secondary public offering transactions as well as ETF creation/redemption activity.

And the trade file moved. The address this article used in August now returns a 404; the working one is whatever ARK’s own trade-notifications page currently links as “Latest Trade File”, and it is on a different host. That is worth knowing if you fetch it yourself, because of how the failure looks: nothing errors in a way you would notice, and anything that assumes a missing file means a quiet day will tell you ARK stopped trading. The fix is dull and it is the only one that works — follow the link from ARK’s page each time rather than trusting an address you saved.

The catch: how these numbers get misread

Three traps, all of which are live in coverage right now.

A disclosed trade is not a change in position. ARK’s holdings move with creations and redemptions as well as with trading, so subtracting two holdings files does not give you the trades. The cleanest proof is Recursion: between the mid-July and early-August files, ARKK’s position fell by 246,221 shares while ARKK’s last disclosed Recursion trade was a buy on 13 July. Nothing was “sold” in the sense a headline would mean.

Dates get shifted by a day, and it matters. A widely syndicated claim has ARK buying 109,129 Circle shares “on 1 August 2026”. Those exact shares, with that exact fund split, are the 31 July trade date. 1 August was a Saturday. The same one-day shift is behind a story run in late July about ARK buying “roughly 160,000 shares” on 24 July. That trade is real, but it is Tesla, not SpaceX, and it is dated 23 July — 160,151 shares, bought the day Tesla fell 14.5%.

Old URLs keep working after they go stale — and new ones stop working without warning. Both halves of that were live again this month, in opposite directions.

Two of ARK’s six holdings files, ARKF and ARKX, still answer with a perfectly valid CSV that is dated 2 January 2026. The other four came back dated 4 September in the same batch of requests, fetched within the same minute. Nothing in the response flags the difference; the date is a column inside the file, and if you do not read it you will publish eight-month-old positions alongside current ones and never see the join. ARKF’s file, for instance, still shows Shopify at 9.65% of that fund — a January number sitting in a September table.

The trade file is the same trap running the other way: its address changed and the old one now 404s. A 200 is not freshness, and a 404 is not silence.

The rule that covers both: the date lives inside the file, not in the URL. Read it every time, per file, and never once for ARK as a whole.

One last number, and a sentence this article is taking back. ARKK’s NAV was $87.14 on 3 September, against $73.54 on 3 August — up about 18.5% in a month, and up 4.44% on that single day.

The August version of this page ended by saying ARK’s trades had not, this year, been profitable relative to owning the index. That was true of the numbers in front of it and it is not a sentence that survives this month, so it has been withdrawn rather than quietly edited. It is also not being flipped: one strong month is not a verdict either, and a fair comparison against the index trackers on the same day could not be sourced for this update. What stands is the narrower claim the page was always really making — the trades are interesting, disclosed, and unusually well documented, which is a statement about the paperwork rather than about the returns.

This is a roundup and an explainer, not investment advice — Drawpie isn’t a financial adviser. ARK’s disclosed trades are historical facts as of the dates shown, not recommendations, and a fund’s decision carries no implication for any individual’s circumstances. Do your own research and consider a licensed professional before acting on anything here.