AInvest Option Flow Digest — 2026-08-25 · $240.5M

Yesterday the tape was hedging. Today it is collecting — and doing plumbing. Eleven names, twenty-seven legs, and the balance is lopsided in a way we rarely see: $174.4M of credits taken in against $66.1M of debits paid out.

One desk sold gold premium three separate ways in a single session. One sold a call barely 1% above an all-time high. One sold a strike a leveraged fund has never closed above. And in SpaceX, a desk moved $41M of deep-in-the-money puts at parity — which is not a bet on direction at all, but a way of getting short a stock nobody can borrow. Against that, two desks paid up for genuinely asymmetric bets — one of them the largest contract count we have printed all month.

$240.5M net structural premium across 11 names, 27 legs. (Gross was $363.0M, but gross double-counts both sides of every spread, so we publish the net — what was actually paid out or taken in.)

⚠️ And a warning that applies to five of today's eleven names. In AA, SBET, HIMS, SPCX and BIDU the raw feed misrepresented the trade — a missing leg, a wider spread, or in one case a print that was cancelled twelve minutes later. Every one of those is corrected below, and in four of the five the real capital at risk is smaller than the headline. Read the structure, not the premium.

And the macro backdrop explains most of it. The Fed's 2026 debate is hold-versus-hike, not hold-versus-cut — the target range has been 3.50–3.75% all year, and the July 29 meeting held with three dissents all preferring a 25 basis-point increase. Selling premium into that is a coherent trade, not a reach.

🔁 OI Review — Yesterday's Predictions, Scored

We published 29 falsifiable open-interest predictions on Monday. Here is how they did, including the one our own stated threshold refutes.

✅ The biggest call landed

PBR — we said the 148,801-contract, four-strike program was a dividend-capture trade ahead of Tuesday's ex-date, and that a collapse in open interest would confirm mass exercise (the inverse of the usual test). Three of four strikes were retired almost entirely: the $17 line fell 16,125 → 363, the $16 fell 3,686 → 126, the $15 fell 2,354 → 103.

The $13 line is the interesting exception — 42,000 contracts traded and open interest barely moved (−234). The coherent reading is that this buyer was filled by someone writing new calls rather than by existing holders: open interest rose ≈42,000 on the print and fell ≈42,000 on exercise, netting to zero. An inference, not a proof — but it fits.

JNJ — we warned in advance that next-day open interest would not be a clean test on the October $190 call, because deep-in-the-money calls get exercised around an ex-dividend. It went 6,093 → 5. Without that published caveat it would have looked like a catastrophic misread.

✅ Fifteen legs landed at or near forecast

Leg	Predicted	Actual	Capture
VLO   Dec $390C	≈4,542	4,544	99.96%
NVDA   Jun-27 $140P	≈145,700	145,595	99.9%
BEKE   Apr-27 $20C	≈10,800	10,810	99.9%
MCHP   Oct $72.50C	≈150,100	150,522	99.7%
CCL   Dec $25P	≈27,000	27,072	99.7%
XEL   Jan-27 $75P / $70P	≈10,150 / ≈10,140	10,082 / 10,046	99.3% / 99.1%
NVDA   Jan-27 $180P	≈148,100	146,592	98.5%
GLD   Sep $430C	≈135,900	131,589	96.3%

PSKY's three-branch test resolved cleanly to the new-open branch (85,476, ΔOI +42,723). The August-19 holder did not exit; a fresh buyer arrived.

❗ Where we were wrong — and the pattern behind it

MCHP's December $65 line was refuted by our own published threshold. We predicted a collapse to ≈300 and wrote that "a decline of anything less than ≈100,000 contracts would refute it." It fell 119,200 → 58,769 — a decline of 60,431, well inside the refutation band.

What survives: it was a closing trade, because open interest cannot fall on an opening sale. What fails is the magnitude — roughly 51% of the block retired open interest and ≈49% was a transfer to new holders. The delta-preserving-roll finding stands; the "they retired the whole position" narrative does not.

And four legs landed in a middle we said did not exist. IBIT's November $40 and $46 calls came in at −9,150 and −9,215 (partial closes, ≈31% capture) rather than either branch. IBIT's June-2027 straddle and VLO's December $320 call both printed essentially flat — pure transfers, where the position simply changed hands.

The lesson, which we have now written into our own process: stop publishing open-versus-close as a clean binary. Volume meets a mix of opening and closing counterparties. From today, every prediction carries three branches and a capture rate — full open, full close, and partial-or-transfer, which is common rather than exotic. You will see that change in every article below.

📊 At a Glance

Ticker	Net Premium	Expiration Type	Catalyst (and its date)	The Option Play	What It Means
GLD	$163.6M credit	Weekly (Aug-28) · Quarterly (Sep-18) · Monthly (Oct-16) · LEAP (Jan-2027)	Jackson Hole + payroll benchmark revision, both 10:00 AM Aug 28   — expiry morning	Three packages: a 5-leg hedged restructure, a 3-day call spread sold, a short straddle	Selling premium three ways.   Delta-hedged with 780,130 shares
EWZ	$12.5M debit	Monthly (Nov-20)	Brazil election — Oct 4 first round, Oct 25 runoff	500,000-contract 43/45 call spread	Defined-risk event bet.   ≈8:1, and the polls point the other way
AMGN	$12.27M debit	Monthly (Oct-16)	Neither earnings nor MariTide is inside the window	In-the-money put above every price target	A valuation bet , not an event bet
TQQQ	$6.17M credit	LEAP-ish (Jan-2027)	3 FOMC meetings; Jan mega-cap earnings fall   after   expiry	Call sold above the fund's own 52-week high	Premium collection , with decay working for the seller
SCHW	$4.68M credit	Quarterly (Dec-18)	Q3 earnings Oct 15   + three FOMC meetings	Call sold ≈1% above an all-time-high close	A timing bet.   Consensus sits above the strike
SBET	$2.39M debit	LEAP (Apr-2027)	Ethereum price; Glamsterdam upgrade Q4 2026	6/10 call spread, 15,000 wide	A bet on the NAV discount closing   — short strike ≈ NAV
HIMS	$1.33M debit	LEAP (Jan-2028)	FTC suit filed Jul 29; peptide rulemaking 2027	Long LEAP put + 45,000-share hedge	Delta-neutral.   Convexity, not a directional short
SPCX	$33.9M debit	Weekly (Aug-28) · Quarterly (Sep-18)	Starship Flight 14 (NET Aug 28)   — same day one leg expires	Four deep-ITM puts at parity	A borrow trade, not a bet.   Synthetic short ≈540,000 shares
BIDU	$1.90M debit	Quarterly (Mar-2027)	EGM tomorrow Aug 26 ; Q3 ≈Nov	2:1 put-heavy risk reversal	Buying the bear case at the bear-case price
VOYA	$1.13M debit	Monthly (Nov-20)	Activist proxy filed Aug 6 ; Q3 ≈Nov 3	5,000 calls, prior OI of 5	Event speculation   — but the campaign peaks after expiry
AA	$0.68M debit	Quarterly (Sep-18) → Monthly (Oct-16)	Q3 earnings date unresolved — Oct 15 or Oct 21	Calendar spread at the $55 strike	Time decay, not direction

🔍 The Eleven, In Detail

1. 🥇 GLD — one name, three packages, $163.6M of credit

→ Read all three, including a spread sold into expiry-morning Jackson Hole

Gold was the whole story today, and it was not one trade but three, almost certainly three different desks.

Package A ($84.6M credit) — a five-leg restructure printed as a stock-and-options cross: selling deep-in-the-money September $392 calls, buying October $450s, and funding it by selling October $400 puts and $505 calls. The equity leg is confirmed — 780,130 shares at $424.30, which is 92.5% of the package's −843,500-share delta. A near-full hedge. This is position management, not a directional view.

Package B ($22.7M credit) — selling the August-28 425/440 call spread for ≈$4.10 on a $15-wide spread, with three days to expiry. And here is what makes it remarkable: Fed Chair Kevin Warsh speaks at Jackson Hole at 10:00 AM on Friday August 28 — expiry morning — and the preliminary annual payroll benchmark revision lands the same minute. They are selling premium into a two-event detonation with under six hours of life left after it.

Package C ($56.3M credit) — a short straddle at the January-2027 $360 strike, collecting $80.40 combined. Breakevens ≈$279.60 and ≈$440.40 against spot of $425.92, so the upside breakeven is only ≈3.4% away. In practice: a bet gold does not rally much further.

2. 🇧🇷 EWZ — half a million contracts on a Brazilian election

→ The largest contract count on the board, and why it is not a moonshot

500,000 November $43 calls bought against 500,000 $45 calls sold — a $12.5M net debit controlling 50 million shares of notional on a $35.42 ETF. Max value $100M against $12.5M risked, roughly 8:1, which implies about a 12% probability.

But it is not a bet on uncharted territory. EWZ's 52-week high is $42.02, so the $43 strike is only 2.3% above it. The Ibovespa hit an all-time high in April and now sits 12.8% below a peak set four months ago. This is a bet that an election-driven de-rating reverses.

The election is the driver — first round October 4, runoff October 25, twenty-six days before expiry — and we will be straight with you: the current polling points the other way. Lula leads both rounds, with the runoff a statistical tie at 46–45. The buyer is positioned against the polls.

3. 💊 AMGN — an in-the-money put above every price target

→ Why neither of Amgen's two big binaries is inside this window

4,125 October $460 puts bought at $29.75 — $12.27M — on a contract whose prior open interest was 3. Essentially zero, so this is 100% proven new risk with no open/close ambiguity at all. Rare, and worth saying.

The striking part is what is not in the window: Q3 earnings falls ≈November 3, eighteen days after expiry, and MariTide's Phase 3 completes in January 2027. So the ≈$13.79 of time value is not paying for a scheduled event. That leaves valuation — and Amgen is at an all-time high, +36.5% YTD, RSI 75, sixteen percent above its 50-day average, with consensus price targets ≈13–14% BELOW the current price. The $460 strike sits above every published target.

4. ⚡ TQQQ — a strike the fund has never closed above

→ The clearest explanation of leveraged-ETF decay you will read this week

15,000 January-2027 $90 calls sold at $4.11 — $6.17M. The strike is ≈28% above spot, and above TQQQ's own 52-week high of $88.09.

Here is the part most people get wrong: a 28% move in TQQQ is not a 9.3% move in the Nasdaq-100. Because the fund rebalances daily, the required index move depends on how smooth the path is — ≈+11.4% at 25% realized volatility, ≈+12.6% at 30%, ≈+15.7% at 40%. Higher volatility raises the bar for the buyer.

The proof is in this year's own tape: the Nasdaq-100 is +13.71% YTD while TQQQ is +32.61% — 2.4×, not 3×. And one more thing the buyer may not have priced: the entire January 2027 mega-cap earnings season falls after the January 15 expiry.

5. 🏦 SCHW — a call sold 1% above an all-time high

→ A timing bet against the Street's own price target

8,000 December $115 calls sold at $5.85 — $4.68M, printed exactly at the bid. Schwab closed at $114.53 on Monday, an all-time-high close about 1% below the strike, after running +23.5% in ten weeks.

And consensus sits above the strike, right on the seller's breakeven of $120.85 — average targets of $121.17 to $124.95, with almost every recent target at $118 or higher. So this is not a bearish call. It is a bet that the Street's twelve-month target does not arrive within 115 days.

The structural argument helping the seller: the self-help margin lever is nearly spent. FHLB borrowings are down 94% to $0.5B, which is most of why net interest margin expanded — and there is very little left to pay down.

6. ⟠ SBET — a call spread whose short strike is the company's own book value

→ 888,938 Ethereum, an 0.82× discount, and a $10 strike

15,000 April-2027 $6 calls bought against 15,000 $10 calls sold — $2.39M net debit, on a contract with zero prior open interest.

Sharplink holds 888,938 ETH (≈$2.20B) against a $1.85B market cap with no debt — so it trades at ≈0.82× net asset value, a discount. NAV per share is ≈$10.35. The short strike is $10. The trade reaches maximum value if the market simply stops discounting the assets, without Ethereum moving at all.

⚠️ The honest counterweight: a flat ETH price with the discount widening to 0.65× would zero this trade — and there is an accounting reason the discount may persist. Nearly 30% of the ETH stack is carried at cost, and the filing states plainly that it will not be adjusted upward for any increase in fair value. An ETH rally will not fix reported book value.

7. 💊 HIMS — one print cancelled, and half the headline with it

→ Delta-neutral, 45,000 shares, and a strike that traded fourteen weeks ago

The feed showed two 2,500-lot prints. The first was cancelled 12 minutes later. Only one stands: 2,500 January-2028 $23 puts at $5.30 — $1.33M, not $2.7M.

It printed as a stock-and-options cross, and the equity leg is confirmed: 45,000 shares against a 47,050-share package delta — a 95.6% hedge. Delta-neutral at inception, so this is a position in volatility and convexity, not a directional short.

The strike is not a tail strike either: Hims priced its convertible off a $22.29 sale in May, the 52-week low is $13.74, and the sell side's lowest price target is exactly $23. Against that, ≈29% of the float is short with only ≈4 days to cover — and March showed what that does on good news: +36% in a session.

8. 🔩 AA — a calendar spread wearing a call-buy costume

→ Why the $1.22M headline overstates the risk by 1.8×

The feed showed a single "$1.22M call buy." The tape shows two legs at the same instant, same $55 strike, different expiries: buying October and selling September. Net debit $680,000.

That is a different trade entirely. A calendar spread wants Alcoa to sit near $55 through September so the short leg decays, then move in October. Time decay, not direction.

And the timing is genuinely unresolved: Q3 earnings is either October 15 or October 21 depending on the source, and Alcoa has not confirmed. If October 15, the back leg captures earnings with a day to spare. If October 21, it expires before the print. Worth confirming before acting.

9. 🚀 SPCX — $41M of deep-in-the-money puts, and not a bet on SpaceX

→ Why four puts at parity is a borrow trade, not a short

Four September and August puts, $41.4M gross / $33.9M net, every one struck deep in the money and every one priced at or below its own intrinsic value — the $215 leg traded $1.36 under parity.

That shape is not a directional view. Net, it is a synthetic short of ≈540,000 shares at a blended $137.95, and the pricing tells you why someone would do it that way: against zero-borrow parity the bought legs price rich enough to imply a ≈7.6% to 10.4% cost to borrow the stock. With 34.91% of the float already short, borrowing SpaceX shares is expensive — so the short gets expressed in options instead.

⚠️ One caution on the chart and any return figure you see elsewhere: this ticker was reused. SpaceX listed on June 12, 2026; anything before that belongs to a different security. A naive year-to-date calculation prints ≈+535%. The honest anchor is +2.6% versus the $135 IPO price.

And note the asymmetry: three of the four legs are already past the point where early exercise is rational — but the August 28 leg is not, carries the most time value, and is the largest. It expires the same day Starship Flight 14 is scheduled.

10. 🔍 BIDU — the strike is not a round number, it is a price target

→ Someone bought the bear case at exactly the bear-case price

The feed showed a put buy. The tape shows three legs: 5,000 March-2027 $80 puts bought against 2,500 $130 calls sold — a 2:1 put-heavy risk reversal for $1.90M net, not $2.82M.

Here is the detail that makes it click: Morgan Stanley cut Baidu to Sell on August 19 and moved its target from $130 to $80 — the exact strike. The consensus low across 32 analysts is $80.13; the consensus average is $149.74. The buyer is paying to be right where one desk out of thirty-two sits.

The case for them: Q2 revenue fell 4% year over year — a sixth consecutive quarterly decline — the ADR dropped 12.7% on the print, and the CFO said advertising will "remain under pressure in the second half." The case against: AI Cloud infrastructure is up 50%, GPU Cloud is up 283% and accelerating, and the stock trades at 0.78× book with more cash than a third of its market value. Breakeven is ≈$74.36 — 16% below the 52-week low.

11. 🤝 VOYA — 5,000 calls on a contract that barely existed

→ An activist, a proxy filing, and a strike above every price target

5,000 November $110 calls at $2.25 — $1.13M, on a strike whose prior open interest was 5, unchanged for a week. Effectively 100% new risk, and the cleanest open on the board.

The quoted market was $0.05 bid / $3.90 ask — a spread wider than the price paid. Size like that simply cannot be done on screen, which is why it went to the floor.

And unusually, the tape has a documented reason. On August 6 an activist holding ≈4.5% launched a "No Confidence" campaign, filed a preliminary proxy, and demanded a strategic-alternatives review — accusing the board of refusing to engage third parties despite informal takeover interest. Voya's rebuttal called the materials "manipulative and deceptive" and did not address the takeover claim at all. The stock missed Q2 earnings by $0.47 and rallied anyway, to within 3.5% of its 52-week high.

⚠️ But the timing works against the buyer. The activist's own stated escalation point is January, six to eight weeks after this option expires, and consensus targets ($108.08 average, $109.50 median) both sit below the $110 strike. Breakeven of ≈$112.26 is above the 52-week high.

⏳ Tomorrow Morning Is the Scorecard

Nine of today's twenty-seven legs printed at a size at or below existing open interest, and four more clear it only marginally. We are changing how we publish these predictions, based on yesterday's misses: every one now carries three branches and a capture rate, not a binary fork.

Test	Full capture (≈100%)	Partial (40–90%)	Flat = pure transfer
GLD   Aug-28 $425C/$440C	≈118,000 / ≈119,000 if opening	—	≈63,000 / ≈64,000 unchanged   — the unwind reading
GLD   Sep-18 $392C	≈55,000 if opening	≈14,000–27,000 decline if closing	≈27,800 unchanged
GLD   Jan-27 $360C / $360P	≈17,600 / ≈13,000	partial	≈10,600 / ≈5,950 unchanged
EWZ   Nov $43C / $45C	≈556,000 / ≈518,000	40–90% capture	—
AMGN   Oct $460P	≈4,128	—	(prior OI was 3 — this one is arithmetic)
SBET   Apr-27 $6C / $10C	≈15,000 / ≈15,039	40–90%	—
SCHW   Dec $115C	≈10,365	≥+5,635 is the arithmetic floor	—
TQQQ   Jan-27 $90C	≈24,293	≥+5,707 floor	—
SPCX   Sep $210C/$230P	≈3,064 / ≈3,283	⚠️   a DECLINE here may mean EXERCISE, not a close — check untraded neighbouring strikes as controls	≈1,514 / ≈1,633 unchanged
BIDU   Mar-27 $80P	≈5,123	40–90% capture	≈123 unchanged
VOYA   Nov $110C	≈5,005	—	(prior OI was 5 — this one is arithmetic)

Read the "partial" column seriously. Yesterday, one GLD leg captured 42% while its twin captured 96% — two legs of the same spread, behaving completely differently.

👥 How to Read Today, By Who You Are

🎲 The YOLO trader. Three names are built for you. VOYA is the purest: 5,000 calls on a strike that had five contracts of open interest, with a documented activist campaign behind it — though the activist's own timetable peaks after the option expires, and breakeven sits above the 52-week high. Then two opposites. EWZ is the cleanest defined-risk lottery on the board — ≈8:1 on a dated event, maximum loss known in advance — but be honest that the polls currently point against it. SBET is the more interesting idea: a call spread whose short strike is the company's own book value, so it can pay off without Ethereum rising at all. The catch is real, though — a flat coin price with a widening discount takes it to zero, and that scenario is not exotic.

📈 The swing trader. Your calendar is unusually crowded. Jackson Hole and the annual payroll revision both land at 10:00 AM Friday, on GLD's weekly expiry. Then Schwab's Q3 print on October 15, Brazil's first round on October 4 and runoff on October 25, and three FOMC meetings between now and December — the last one nine days before December expiry. Also note what is not in these windows: Amgen's earnings and MariTide data both fall past October 16, and January's mega-cap earnings fall past TQQQ's January 15 expiry. Expiry dates are not calendars; check them.

💵 The premium collector. Today is your tape — six of eight desks were selling, and credits outweighed debits roughly six to one. But look at where they sold: Schwab's call went off 1% above an all-time high, TQQQ's above the fund's 52-week high, and GLD's three-day spread into a two-event Friday morning. These are not lazy far-out-of-the-money sales; they are considered bets that specific things will not happen by specific dates. The lesson is in the selection, not the act. And remember the standing caveat — we cannot tell from the tape whether any of these short calls is covered, and that distinction is the entire difference between an income strategy and a blow-up.

🌱 The beginner. Start with AA, because it is today's clearest lesson in why a headline can mislead: a "$1.22M call buy" is really a $680,000 calendar spread, and the difference is not cosmetic — it is a different trade with a different goal. This happened five times today — in AA, SBET, HIMS, SPCX and BIDU the feed showed something other than the real trade. Then read HIMS, where one of two prints was cancelled and the real trade is half the reported size. Then TQQQ, for the single most useful concept in leveraged products: a 3× fund does not deliver 3× over time, and this year's own numbers prove it — the index is up 13.7% while the fund is up 32.6%, not 41%.

⚠️ Risk, Plainly

A large trade is not a recommendation, and it is not a prediction. Every desk on this list has information, hedges, mandates and time horizons you cannot see. The tape does not carry the broker, the customer, or any stock position held away from the print. Everything we call intent is an inference from structure, size versus open interest, strike geometry and pricing — never a disclosure.

Three things worth more than any flow alert:

Check what the expiry actually captures. Three separate names today have a major catalyst falling outside the option's life. That is the most common unforced error in this business.

The headline size is often not the risk. AA's real capital at risk was 56% of the reported number; HIMS's was 50%; SBET's was 46%. Read the structure, not the premium.

On a negotiated block we cannot prove who bought and who sold. Where you see a "reported side" in our tables, that is a reporting convention — not evidence.

Nothing here is investment advice. Options carry the risk of total loss of premium, and short option positions carry risk far beyond the premium collected.