MARKET BRIEFING (6)

We buy eggs at Costco. Kirkland brand, two dozen, somewhere between $7 and $8 depending on the week. Our family goes through about a dozen a week, so one carton lasts us two weeks, and I honestly stopped thinking about the price after the madness of 2025.

Then on August 14, 2026, the FDA elevated the egg recall to Class I — its highest risk level, the one reserved for products that can cause serious health consequences or death — and my wife asked me the only two questions that actually matter:

Are our eggs in it? And is this going to make eggs expensive again?

So I did what I always do. I pulled the FDA and CDC filings, the USDA price reports, and a calculator. The answer to the second question surprised me, and not in the direction the headlines suggest.

First: is it your eggs?

The recall is from Midwest Poultry Services, L.P., which recalled 1,589,577 dozen shell eggs on July 22, 2026. That’s about 19 million eggs.

Quick note, because I saw this in three different places: some outlets ran with “38 million eggs.” That looks like a doubling error. 1,589,577 × 12 = 19,074,924. It’s 19 million. Bad enough without inflating it.

Brands on the recall list:

• Kroger
• Simple Truth Cage Free
• Brookshire’s
• Country Morning
• Sunups (a Cal-Maine brand)
• Plus bulk Grade A/AA foodservice packs

One clarification, because a lot of coverage muddles it: Cal-Maine did not issue this recall. Midwest Poultry Services did. Sunups is a Cal-Maine label that appears in the recalled packaging, which is a very different thing from Cal-Maine’s own farms being implicated. Cal-Maine has said as much, and it matters if you’re looking at the stock later in this post.

They came in every size you can imagine — 6, 12, 18, 24, 30, 36 and 60-egg cartons — in both white shell and brown cage-free.

The identifying codes. This is the part that actually matters, because the brand name alone doesn’t tell you anything:

• Plant code P-1950 or 0840962
• Julian date 157 through 184
• Best-by / sell-by dates July 20 through August 17, 2026

That last line is why this is still urgent as I write it. The best-by window doesn’t close until August 17. These cartons are not a historical curiosity — they can still be sitting in a refrigerator right now.

Is my state affected?

Here’s where it gets genuinely confusing, because there are two different lists and most coverage blurs them together.

Where the recalled eggs were shipped (6 confirmed states): Texas, Oklahoma, Louisiana, Arkansas, Mississippi, New Mexico. Produced at Texas farms between June 6 and July 3, 2026. The FDA notes the product may have reached additional states, so treat this as a floor, not a boundary.

Where people actually got sick (17 states): Arizona, California, Colorado, Georgia, Illinois, Louisiana, Michigan, Missouri, Mississippi, North Carolina, New Mexico, Nevada, New York, Oklahoma, South Carolina, Texas and West Virginia.

Seventeen states of illness from a six-state distribution. That gap is normal — people travel, eat out, and the eggs went to foodservice accounts too — but it’s why “my state isn’t on the distribution list” is a weaker comfort than it sounds. The FDA has also said the products “could have reached additional states.”

Washington is on neither list. Neither is Oregon, or anywhere on the West Coast except California and Nevada on the illness side.

One local wrinkle worth naming, since I live here: Kroger owns QFC and Fred Meyer, and Kroger and Simple Truth are both recalled brands. That is not the same thing as saying recalled eggs came to Seattle — the recalled lots went to Kroger stores in Texas and Louisiana. But if you see “Kroger brand eggs recalled” in a headline and shop at Fred Meyer, I understand the jolt. Check the plant code, not the logo.

The part that made me nervous

The CDC counts 98 confirmed illnesses and 26 hospitalizations, with no deaths. Illness onset runs from November 21, 2025 all the way to June 30, 2026 — this has been quietly going on for months.

But here’s the detail that got my attention as a parent: whole genome sequencing of the patient samples predicted resistance to ciprofloxacin, one of the standard antibiotics for a severe Salmonella case. That doesn’t make you more likely to get sick. It makes the bad outcome harder to treat if you’re in the small group that ends up hospitalized — older adults, very young kids, anyone immunocompromised.

Symptoms usually show up 12 to 72 hours after exposure: diarrhea, fever, stomach cramps, typically 4 to 7 days. The CDC says get medical attention for diarrhea lasting more than two days, fever above 102°F, bloody stools, or signs of dehydration.

Confession: writing this taught me I’ve been doing two things wrong

I was going to write a smug paragraph telling you to go check your carton. Then I went to check mine and realized I couldn’t.

We throw the carton away the moment we get home. Eggs get rinsed under water and go into a plastic egg tray in the fridge. It looks tidy. I’ve been doing it for years.

Both halves of that are a mistake, and I’d rather tell you myself than have you find out later.

Mistake 1: the carton is the only place the recall code lives. Plant code, Julian date, best-by date — all of it is printed on the carton, none of it is on the egg. Toss the carton and you have permanently opted out of every future egg recall. You will never be able to answer “is it mine?” The FDA’s own egg safety guidance says plainly: “Store eggs in their original carton and use them within 3 weeks for best quality.” The carton also insulates the eggs and keeps them from absorbing fridge odors. I thought I was being neat. I was destroying my own receipt.

Mistake 2: washing store-bought eggs can make them less safe, not more. US commercial eggs are already washed and sanitized at the plant, which strips the shell’s natural protective cuticle — processors replace it with a thin food-grade mineral oil coating. When I rinse that off at home, I’m removing the barrier. Worse, eggshells are porous by design, so water sitting on the shell can carry surface bacteria inward through the pores. USDA guidance is that you don’t need to wash store-bought eggs, and that washing them can increase contamination risk.

Two habits, both of which felt responsible, both of which quietly made my family less safe. This is why I write these posts with a calculator and a source list instead of from memory. T_T sad.

New rule in our house: eggs stay in the carton until they’re cracked. If the tray really matters to someone aesthetically, at least cut the code panel off the carton and tape it inside the fridge door.

And whatever else you do — cook them properly. Firm yolk and white, or 160°F for anything custardy or casserole-ish. Heat is what actually kills Salmonella. No amount of rinsing does.

Now the money question: does a 19-million-egg recall raise prices?

This is the part everyone gets wrong, including a few finance sites that should know better. Let me put on my money-blogger hat.

19 million eggs sounds enormous. It is a rounding error.

The US produced 636.8 million dozen table eggs in June 2026 alone, per USDA NASS. (Total egg production including hatching eggs was 741.6 million dozen — I’m using the table-egg number, since that’s the eggs-you-eat category the recall belongs to.) The recall is 1.59 million dozen. Run the division:

1.59 ÷ 636.8 = 0.25% of a single month’s production.

Put another way: America produces the entire recalled quantity in roughly one hour and 48 minutes.

A supply shock that lasts under two hours does not move a national grocery price. If you allocated the entire recall’s cost across US egg spending and passed it straight through to my family, it would cost us somewhere around fifty cents a year. That’s an illustrative figure, not a forecast — but it’s the right order of magnitude, and the right order of magnitude here is “pocket change.”

So no. The recall is not why your eggs might cost more.

But prices ARE moving. Here’s the actual reason.

Wholesale large eggs jumped about 35 cents to $1.09 a dozen in the week ending July 24, with large-egg inventories down 8% week over week. So something is definitely happening.

A lot of coverage hung that jump on the recall. I went and pulled the weekly USDA AMS series, and the timeline doesn’t support it:

Week	National large loose eggs	Change
July 10	—	+$0.19
July 17	$0.74	+$0.28
July 22	← recall announced
July 24	$1.09	+$0.35
July 31	$1.08	−$0.01
August 7	$0.98	−$0.10

The rally started two full weeks before the recall. Prices were already climbing 19 and 28 cents a week, and Large-class inventories were already down 7% the week of July 17 — before anyone had heard of Midwest Poultry Services. The recall broke on July 22, roughly one day into the July 24 reporting week.

And then look at what happened after: prices flattened, then fell back to 98 cents by August 7. If a Class I recall of 19 million eggs were driving the market, that is not the shape the chart would make.

So what is? Bird flu. Again.

Producers have destroyed roughly 12.4 million commercial laying hens in 2026 year to date. For comparison, the entire second half of 2025 saw 5.4 million. APHIS confirmed commercial poultry detections across 12 states as of June. Every detection triggers mandatory depopulation of the whole flock, and depopulations historically show up at retail six to eight weeks later.

Unlike a recall, that’s a real supply loss. A recalled egg was already laid. A culled hen doesn’t lay tomorrow, or for the 18 months it takes to raise her replacement to laying age.

The USDA agrees. In its August 14, 2026 outlook, it raised its 2026 egg price forecast to 110.6 cents per dozen, up from 96.8 cents a month earlier. That’s a 14% upward revision in a single month — a big move for a government forecast. It also nudged first-half 2027 up to 111 cents from 108, and trimmed 2026 production to 9,109 million dozen.

The context nobody puts in the headline

Here’s the part that should calm you down: eggs are still dramatically cheaper than a year ago.

The USDA’s national wholesale index for large caged eggs on August 3, 2026 was 105.31¢. One year earlier: 285.55¢. Retail told the same story — the BLS average for a dozen Grade A large eggs was $2.19 in July 2026, down roughly 35% year over year. USDA still projects 2026 finishing about 30% below 2025 overall.

So the honest framing is not “eggs are spiking.” It’s “eggs crashed from an insane peak, spiked briefly in July, eased again in early August, and are expected to run modestly higher into next year from a very low base.”

Less exciting. More true.

What this actually costs my family

Enough theory. Real numbers, my actual grocery habit.

We pay about $7.50 for two dozen Kirkland eggs at Costco (call it the midpoint of $7–$8). We eat about a dozen a week, so 52 dozen a year — 26 cartons.

That’s 31 cents an egg, and about $195 a year on eggs.

A three-egg scramble costs me 94 cents.

Now let’s stress-test it. If the USDA’s 14% upward revision passed through to retail completely and immediately — which it never does, retail is stickier and Costco is stickier still — here’s the damage:

Scenario	Per dozen	Per egg	3-egg scramble	Our yearly egg bill
Today ($7.50 / 2 dozen)	$3.75	31¢	94¢	$195
USDA’s new forecast (+14%)	$4.28	36¢	$1.07	$223
My own 2025 peak ($10 / 2 dozen)	$5.00	42¢	$1.25	$260
The all-time national record (Mar 2025)	$6.23	52¢	$1.56	$324

The USDA scenario costs my family $28 a year. Fifty-four cents a week. Thirteen cents per scramble.

The nightmare scenario — a complete replay of what I personally paid at the worst of 2025 — costs us $65 a year. A dollar twenty-five a week.

And that bottom row is the true worst case in recorded American history. The BLS monthly national average for a dozen Grade A large eggs peaked at $6.23 in March 2025, and the daily national average touched an unbelievable $8.15 on March 4, 2025. If that exact peak had held for a full year, our egg bill would go from $195 to $324 — $129 more, or $2.48 a week.

Two side notes worth pulling out of that table.

First: even at the peak, my Costco price of $5.00 a dozen was running about 20% below the national average of $6.23. Buying the bigger pack at a warehouse store did more for me during the crisis than any strategy I could have layered on top.

Second, and this is the part I want to sit on: the worst egg crisis in modern American history — a year of headlines, a DOJ antitrust inquiry, genuine political fallout — would have cost a family eating a dozen a week about $129 over twelve months. Roughly one restaurant dinner.

That is not me saying it doesn’t matter. For a family eating four dozen a week on a tight budget, that $129 becomes $516, and that is very real money. But it explains why I’m not reorganizing my finances over an egg headline. Eggs are the most visible price in the grocery store — a single item, bought weekly, with a number you can’t help but memorize. Visibility isn’t the same as impact. The line on my budget that actually deserves this much attention is the mortgage, and it isn’t close.

Should I stockpile? (I wanted to. I can’t.)

My instinct was to grab an extra carton on the next Costco run. Buy at 31 cents an egg, laugh at 36 cents in October. Easy win.

Then I opened my refrigerator.

We don’t have the space. Two adults, young kids, a Costco-sized everything already fighting for shelf real estate. There is no world in which four dozen eggs live in my fridge. So I’m buying one carton, same as always.

Which is honestly the right answer anyway, and here’s the math that proves it. Eggs keep about 3 to 5 weeks refrigerated. Say I successfully stockpiled an extra month’s worth — four dozen, $15 — right before a 14% increase. I’d save 14% of $15, or $2.14. Once.

Two dollars and fourteen cents, in exchange for a jammed refrigerator, elevated spoilage risk, and the mental overhead of tracking best-by dates. And that’s the best case, where I correctly time a price move that professional commodity traders can’t reliably time.

The stockpiling play is not worth it at this scale. Not because it doesn’t work, but because eggs are too cheap for the win to matter. Sometimes the honest answer to “how do I optimize this” is “this isn’t worth optimizing.” There’s always a catch, and here the catch is that the prize is two bucks.

My fridge being too small saved me from a bad idea. I’ll take it.

If you want to actually reduce this line item: buy the larger Costco pack, don’t waste them (Americans throw out a shocking share of the eggs they buy), and if you order groceries online, run it through a card with a grocery bonus. My Sapphire Preferred earns 3x on online groceries — my in-store Costco run doesn’t qualify, so I’m not going to pretend it’s saving me anything here. But if delivery is how you shop, that one’s worth a look.

Which stocks get hit?

Now the part I find most interesting, because the market’s reaction was the exact opposite of what the headlines would predict.

On Friday, July 24 — the day the CDC and FDA announced the outbreak link — the egg-linked stocks went UP.

• Cal-Maine Foods (CALM) — the recalled Sunups brand is its label — closed $90.32, up 2.5%
• Kroger (KR) — the clearest retail link, via Kroger and Simple Truth labels — closed $56.87, up 2.0%
• Vital Farms (VITL) — no exposure to this recall whatsoever — closed $13.02, up 2.2%

(On July 22, the actual recall announcement day, CALM rose a barely-there 0.7% to $87.86. The market’s shrug was immediate.)

Why? Because traders ran the same division I did above. The recall is 0.25% of a month’s supply — for Cal-Maine specifically, the affected volume is roughly 0.13% of its annual shell-egg sales of about 1.2 billion dozen. Meanwhile wholesale prices were climbing hard. The market looked straight past the recall and priced the eggs. For a producer, an egg-price increase is revenue.

That’s the counterintuitive lesson worth keeping: a recall is a headline risk; a supply shortage is an earnings event. They can point in opposite directions, and the market usually cares about the second one.

Where they sit today

Checked on August 14, 2026, the day of the Class I upgrade:

Ticker	Company	Price	Day change
CALM	Cal-Maine Foods	$81.09	−1.8%
KR	Kroger	$56.69	−0.9%
VITL	Vital Farms	$10.90	−6.0%
POST	Post Holdings	$80.95	+0.5%
SFM	Sprouts Farmers Market	$82.50	−0.2%
TSN	Tyson Foods	$58.17	+3.2%

Mostly red. Worth noting that CALM has given back most of its year: it was up about 5% YTD in late July, and after today’s slide it’s hovering near roughly flat for 2026 — it also posted a surprise quarterly loss in July, which matters a lot more to that stock than any recall. Verify every quote here before acting on it; they’ll be stale by the time you read this.

For context on the sector, 2026 has been a brutal year for food-safety headlines generally — Chipotle dropped 9% on a jalapeño-linked salmonella investigation (a separate outbreak, 345 illnesses across 27 states, which several outlets have confusingly mixed into egg-recall coverage), and Taco Bell pulled shredded lettuce nationwide over a cyclosporiasis outbreak.

How I’d actually think about it

Cal-Maine (CALM) is the pure play, and it’s a genuinely strange asset. It’s a commodity producer whose earnings swing violently with egg prices, which in turn swing with a bird virus nobody can forecast. It printed extraordinary profits during the 2025 crisis and a surprise quarterly loss in July 2026 — same company, about a year and a half apart. It is not a quiet compounder. It’s a bet on avian influenza severity, whether you frame it that way or not. It also carries legal overhang from the post-2025 price scrutiny: on June 29, 2026 it settled with the DOJ and 17 state attorneys general over alleged egg-price antitrust claims for $1.5 million, plus a commitment to donate 30 million eggs.

Kroger (KR) carries the brand and liability exposure, but eggs are a rounding error against a company this size. The recall is a headline, not a thesis.

Vital Farms (VITL) had nothing to do with this recall and is down about 6% intraday today anyway. That’s the sympathy-move risk in this sector: the whole shelf gets repriced when one carton makes the news, whether or not the carton was yours.

The honest position: I’m not doing anything. Not because there’s no trade here, but because “buy an egg producer because a bird virus might get worse” requires me to forecast avian influenza, and I can’t. Neither can the USDA — that’s why they just revised their own forecast 14% in one month.

What I am doing is watching USDA’s weekly layer-flock depopulation numbers rather than recall headlines. Depopulations lead retail prices by six to eight weeks. If culls accelerate into fall — and fall is when wild-bird migration historically drives detections up — that’s the actual signal, and it will show up in the data long before it shows up in my Costco receipt.

What I’d do this week

1. Check your carton codes — P-1950 or 0840962, Julian 157–184, best-by July 20–Aug 17. Best-by dates run through Monday, so this is live right now.
2. If they match, don’t taste-test them. Salmonella doesn’t change how an egg looks or smells. Throw them out or return them for a refund, then wash the shelf and any container they touched with hot soapy water.
3. Keep the carton from now on. Costs nothing. It’s your only proof of identity in the next recall.
4. Don’t rinse store-bought eggs, and cook them until firm.
5. Don’t panic-buy. The recall isn’t a supply event, and the real driver — bird flu — moves on a six-to-eight-week lag you’ll have plenty of warning about.
6. If you’re watching the tickers, watch flock depopulation data, not recall headlines. The 2026 chart already told us which one the market cares about.

The bottom line

A 19-million-egg Class I recall is a serious food safety story — antibiotic-resistant strain, 26 people hospitalized, cartons still in fridges this weekend. Take that part seriously and go check your codes.

As a money story, it’s noise. The recall is 0.25% of one month’s supply — under two hours of national output, and the wholesale rally that got blamed on it started two weeks before it happened. What actually moves your egg bill is 12.4 million dead hens. And even a full replay of March 2025, the most expensive month for eggs in recorded American history, would cost my family $129 over a year.

Check your carton codes. Keep the carton. Don’t rinse them. Cook them until firm.

Then go worry about a line on your budget that costs more than one dinner out.

---

Recall details verified against FDA and CDC filings on August 14, 2026. This is an active investigation — codes, brands and affected states can expand, so confirm current status on FDA.gov before relying on anything here. Price figures come from USDA NASS, USDA AMS and BLS as of the dates cited; egg markets move weekly and these will age fast. The July 24 AMS weekly figure is secondary-sourced. Stock prices in the “where they sit today” table are intraday on August 14, 2026. The cost scenarios are illustrative arithmetic built on USDA’s own published forecast revision — they are not predictions, and nobody can forecast bird flu, the USDA included (they just revised themselves 14% in a month). I hold no position in any stock mentioned and this is not a recommendation to buy or sell anything. I’m a software developer who writes about my own money, not a financial advisor or a doctor, and nothing here is financial or medical advice. If you think you have a Salmonella infection, call a doctor, not a blogger.

Every few weeks the AI trade throws out a phrase that makes old finance people flinch, and lately the phrase is vendor financing. If you’ve been around markets long enough — or you’ve read enough dot-com post-mortems — that phrase comes with a scar attached, and the scar is named Cisco.

So let’s do the Hebojago thing: no hype, just math. What is vendor financing, what actually happened to Cisco in 2000, what is Nvidia doing with OpenAI right now, and are they the same movie? Short version: the shape rhymes, the scale is bananas, and the details are different in ways that matter in both directions.

One honest note up front: a lot of the biggest 2026 numbers in this story are reported, not confirmed. I’ve flagged every one. Nvidia and OpenAI have not confirmed the headline figures, and people close to the talks say the deal could still fall apart. Treat the rumored numbers as rumored.

What “vendor financing” even means

Strip out the jargon: vendor financing is when the company selling you something also lends you the money to buy it.

That’s it. A chip company helps its customer pay for the chips. A network-gear company floats a loan to the startup buying routers. On the way up it looks like genius — your sales explode, because you’ve removed the one thing stopping customers from buying (they’re broke). The catch is that you’ve quietly turned a chunk of your “revenue” into loans you’re hoping get paid back.

The uncomfortable question is always the same: are people actually demanding your product, or are they buying it with your own money? If it’s the second one, your revenue growth and the customer’s “demand” are the same dollar, counted twice.

Cisco, 2000: how this ended last time

Rewind to the peak of the dot-com and telecom boom. Equipment makers — Cisco, Lucent, Nortel — were extending billions in loans to cash-strapped internet providers and “CLECs” (competitive local phone carriers) so those startups could buy networking gear. Per the historical record: Lucent committed around $8.1B in vendor financing, Nortel extended roughly $3.1B, and Cisco promised about $2.4B in customer loans. By 2000, an estimated ~10% of Cisco’s ~$20B in revenue was tied to this kind of financing.

For a while it was magic. Cisco’s market cap hit roughly $450 billion in March 2000, briefly making it the most valuable company on earth, at something like 165x earnings.

Then the customers — companies with big dreams and no profits — couldn’t generate enough revenue to pay the loans back. Between 2000 and 2003, dozens of CLECs went bankrupt (Covad, NorthPoint, Focal, and 40-plus others). Cisco wrote off close to $900 million in bad customer loans by 2001. The stock fell more than 89% from its peak — and here’s the part that should stick with you: even growing earnings many times over in the two decades since, Cisco’s stock spent an astonishingly long time never getting back to its 2000 high.

The lesson wasn’t “vendor financing is fraud.” Most of it wasn’t. The lesson was: when you finance your own demand, you don’t find out it was fake until the loans come due.

Nvidia + OpenAI, 2025–2026: the same skeleton, way more zeros

Here’s the timeline, cleaned up:

• September 2025 — Nvidia and OpenAI announce a letter of intent for Nvidia to invest up to $100 billion in OpenAI as it deploys 10 gigawatts of Nvidia systems. Analysts immediately call it “circular.” (Bernstein’s Stacy Rasgon flagged it the day it dropped.)
• February 2026 — That $100B number quietly collapses. Instead, Nvidia takes a roughly $30 billion equity stake in OpenAI’s funding round (reported valuation in the $850B range). CEO Jensen Huang later reframes the original $100B as “never a commitment.”
• July 2026 (reported, not confirmed) — Nvidia is said to be in talks to guarantee ~$250 billion in data-center lease payments for a giant 10-GW OpenAI site in Ohio (developed by SoftBank’s SB Energy), plus separately finance up to ~$350 billion in chip purchases tied to the same site. Total project cost reportedly north of $500B — which would make it the largest data center ever proposed.

The structure is the tell. That Ohio backstop reportedly lets the developer borrow against Nvidia’s balance sheet rather than OpenAI’s — because OpenAI still doesn’t carry an investment-grade credit rating. So Nvidia’s financial strength becomes the collateral that lets a customer buy… Nvidia chips.

That’s the circle. Money leaves Nvidia (as an investment, a guarantee, or a backstop), and comes back as GPU revenue. Nvidia has consistently said it does not contractually require partners to spend the money on its chips — worth stating plainly and fairly. But critics like Michael Burry aren’t buying the spirit of it; Burry summed up the loop as “around and around we go.”

The differences that actually matter

This is where lazy takes go one of two ways — “it’s exactly Cisco, sell everything” or “it’s totally different, nothing to see here.” Both are wrong. Here’s the honest side-by-side.

1. Scale is not comparable — it’s a different planet. Cisco committed ~$2.4B against ~$20B of revenue. The Nvidia figures being floated ($250B guarantee, $350B chip financing) are two orders of magnitude larger. And here’s the number almost nobody quotes: Nvidia’s own Q1 FY2027 10-Q caps its total lease-guarantee exposure at just $3.5 billion. A $250B commitment would be roughly 71x its currently disclosed guarantee book. That gap is either “the filings will catch up to the ambition” or “the ambition is mostly a press release.” I genuinely don’t know which yet — and neither does anyone tweeting confidently about it.

2. The customer is not a no-name CLEC. Cisco lent to startups with dreams and no revenue. OpenAI reportedly generates around $2 billion a month. That’s real. That’s the strongest point for the bulls. The asterisk: OpenAI also runs enormous operating losses, and $2B/month of revenue does not cover a $500B+ buildout. Real revenue, still-negative math.

3. Cisco mostly lent. Nvidia has shape-shifted. Watch how Nvidia’s exposure keeps changing costume — equity stake, then loan guarantee, then lease backstop. Each version moves the risk around without removing it. That’s more sophisticated than Cisco’s straight customer loans. Whether “more sophisticated” means “safer” or just “harder to see on the balance sheet” is the whole debate.

4. Nvidia is absurdly healthy; Cisco’s customers weren’t the only fragile ones. Nvidia runs monster margins and a ~$5T-ish market cap. It can eat losses Cisco couldn’t dream of. The counterpoint: Cisco was also healthy relative to its customers in 1999, and still got dragged down 89% when the customers cracked. Being the strong one in a circle doesn’t save you if the circle breaks.

5. The bull’s ace: the infrastructure might be durable. The dot-com telecom bust laid a trillion dollars of “wasted” fiber that later became the literal backbone of the cloud boom. If AI compute turns out to be similarly durable and useful, a shakeout in who owns it doesn’t mean the stuff was worthless. Bubbles can vaporize equity and still leave useful pipe in the ground.

My take — I’m long, and I’m not going to pretend I’m not

Full disclosure, because that’s the whole point of this blog: I hold NVDA, and it’s about 50% of my stock portfolio. That’s a big, concentrated bet and I’m not going to dress it up as anything else — one name, half the book. If you’ve read my other stuff, you know where I land: I don’t think AI is a fad, I think it’s the biggest platform shift of my working life, and I’d rather own the company selling the shovels than try to guess which miner strikes gold. So my plan is simple and I’ll say it out loud — I keep buying NVDA on dips, with money I can actually afford to invest, whenever I’ve got cash to put to work.

So does the “circular financing” story change that? Honestly, no — but it sharpened what I’m watching. Here’s the distinction I keep coming back to: “circular financing” is a real risk, not a reason to panic-sell. It’s the bear case that could actually bite, which is exactly why I’d rather understand it than flinch at the word. Cisco taught me the word isn’t the danger — the loans coming due is the danger. So I watch the tells below, not the headlines.

Now the catch, and I have to say it to myself as much as to you, because it’s the same lesson as my leveraged-ETF panic piece: 50% in one stock is aggressive, and my conviction doesn’t make it diversified. If the AI trade takes a Cisco-style haircut, I take it too — no thesis talks you out of an 89% chart. I’m comfortable with that risk with my money and my timeline. You might not be, and you shouldn’t inherit my risk tolerance just because you like my writing. “Buy the dip” is a great slogan and a terrible reason to overextend — I only do it with money I won’t need, never borrowed, never a dip I feel like I have to catch. Know what you actually own and why, so a scary headline doesn’t get to make the decision for you.

What I’m actually watching (so I don’t have to guess)

Three concrete tells, in plain English:

1. Does the $250B show up in a filing? Rumors are rumors until they’re in a 10-Q or an 8-K. If the disclosed guarantee book jumps from $3.5B toward the reported figure, the story got real. If it doesn’t, a lot of this was vibes.
2. Can OpenAI (and peers) start paying with their own cash? The tell for “real demand” is customers funding purchases from operating cash flow, not from vendor-backed debt. Watch the mix.
3. Does anyone break the circle first? In 2000, it only took a few CLEC bankruptcies to expose the whole structure. Today’s circle has more players (Microsoft, AMD, SoftBank, CoreWeave, and more). More nodes = more resilient, or more places for a crack to start. Both are true until one wins.

None of this tells you what the stock does next quarter. It tells you what to watch so you’re reacting to the balance sheet instead of the headline.

---

Not financial advice — I’m a software developer who trades on the side, not a professional analyst. Full disclosure: I’m long NVDA and it’s a large, concentrated position (see my take above), so read my read with that bias in mind. The 2026 Nvidia–OpenAI figures ($250B guarantee, $350B chip financing, $500B+ project) are reported and unconfirmed as of publication and the deal may change or collapse; verify current terms before acting on anything. Historical Cisco/telecom figures are drawn from the public record and widely cited post-mortems. This is my read on public information, not a recommendation to buy or sell anything.

Hebojago is my journey to financial freedom, recorded live — real experiments, real numbers, wins and mistakes. Follow along here.

The semiconductor sector officially entered a bear market — down more than 20% from its June peak (SOXX, the iShares Semiconductor ETF, fell from around $655 to the low-$520s). Every time chips drop this hard, the same two sentences show up together: “It’s a buying opportunity.” and “Not yet.”

I felt the same itch you probably did — open the account, see red, want to average down, and somehow the finger won’t click. Turns out Wall Street was frozen in the exact same spot. Now they’ve answered — and they gave three different answers. Here’s the map.

Semiconductor stocks just entered a bear market — how it happened

No single crash. Just one brick pulled out of the wall every few days:

• A class-action lawsuit accused Samsung, SK Hynix and Micron of throttling DRAM supply to prop up prices — putting “memory prices aren’t normal” on the record.
• Cerebras got attention for building AI chips on SRAM instead of HBM — a hint that the HBM chokehold isn’t permanent.
• Meta was reported to be weighing renting out spare data-center capacity — read by the market as a whiff of over-investment.
• Raymond James and Evercore ISI both warned memory prices may be near a cycle peak. That’s the scary one: memory makers’ profits jumped mostly because prices rose. Peak price = peak profit. Not “earnings get bad” — just “earnings can’t get better.” Markets hate that even more.
• SK Hynix reportedly considered slowing HBM expansion — i.e., building less of the priciest product.
• Then a macro punch: Middle East tension pushed oil up, which revived inflation fears, which killed rate-cut hopes. The 2-year Treasury yield hit ~4.2%, its highest since February 2025. Rising rates hit long-duration “pay now, profit later” stocks first — chips sat right in the blast radius.

Korea got it worse: SK Hynix fell as much as ~43% intraday from its high — 2022-crash territory.

The part almost nobody notices: who was actually selling

Here’s the weird bit. In July, SOXX had one of its worst months in years — and yet billions flowed into the ETF the same month (around $6.1B, its biggest monthly inflow since 2018, with one single day near $5.4B). Sister fund SMH saw big inflows too.

So money’s pouring in… and the price still drops? Think of the price as the water level in a tub:

• The faucet = money flowing into ETFs. This is mostly unborrowed money — pensions, 401(k) contributions, retail. Nobody can force it to sell. When price drops, it actually flows harder, because things got cheaper.
• The drain = hedge funds, which had been pulling money out at a record pace. Goldman’s prime-brokerage desk (which sees what hedge funds trade in real time) flagged signs of capitulation — the last holders finally giving up.

And why do they give up? Not impatience — leverage. Hedge funds run on borrowed money against collateral. Price falls → collateral falls → they’re forced to sell to shrink the position. They’d love to buy the dip; the margin math won’t let them.

The drain was bigger than the faucet, so the water level dropped. But the two flows are completely different animals: one is forced, one is free.

The three Wall Street scenarios

① The bottom is already in — JP Morgan. Buyers’ money is locked in; supply can’t grow for years. Hyperscalers (Google, Microsoft, Amazon, Meta) have pledged roughly $725B to AI infrastructure — reportedly up ~77% year-over-year, and nobody’s cut. New fabs started today don’t ship until ~2028. No new supply + committed demand = little reason for prices to fall. JPM even raised its memory-market-size estimate and told clients to re-enter chips this summer. → If true: today’s screen is the cheapest you’ll see for a while.

② A bounce comes, but the leadership doesn’t — Morgan Stanley (Mike Wilson). Expect a rebound, but don’t count on chips leading the market again in the back half of the year. The AI capex cycle isn’t over, but spending has outrun revenue — and credit and CDS spreads are widening, the bond market flashing caution while stocks still party. → If true: your exit plan matters more than your entry. Ride the bounce, but decide in advance where you get off.

③ The bottom isn’t here yet — Evercore ISI. This is a mid-cycle correction, not the end of the run. Across the last 25 years, all six chip upcycles saw a correction within ~4 months of the revenue-growth peak (Evercore pegs this cycle’s peak at Q3 2026). Those corrections historically ran ~13% and ~6–7 weeks, and bottomed when chip valuations matched the S&P 500 — right now chips are still ~10% more expensive. So: maybe 2–3 more weeks and another 10–15% down (roughly SOXX $450–475). The kicker: after these corrections ended, chips historically rebounded ~36% over ~20 weeks, and the hardest-hit names led — Evercore flagged Micron, ARM, and Intel. → If true: don’t fire all your cash at once — there may be a cheaper window in a few weeks.

The thing all three agree on: none of them thinks the AI cycle is over. The fight isn’t buy vs. don’t — it’s when. And all three are staring at the same number: hyperscaler data-center spending. This earnings season (Alphabet kicks it off) settles who’s right.

⚠️ Very important to know before you touch this

This is the part I’d tell a friend, no hype:

1. “Down 20%” is not the same as “cheap.” By Evercore’s yardstick, chips were still ~10% more expensive than the S&P 500 at the lows. Fell a lot ≠ good value.
2. Forced selling ≠ your selling. Hedge funds sold because leverage made them. You (probably) have no margin call. Copying a forced seller with free money is a totally different decision that just looks the same.
3. Money flowing in doesn’t mean up — yet. The drain can outrun the faucet for a while. But leveraged sellers eventually run out of things to sell.
4. You’re not going to miss it by a day. Historical rebounds averaged ~20 weeks. This isn’t a lottery ticket you miss by sleeping in.
5. Two things to actually watch: (a) the data-center capex line in Big Tech earnings — held = bullish, trimmed = the ②/③ worry gets real; (b) whether DRAM/NAND selling prices actually roll over, or it’s just talk.
6. The real edge isn’t timing — it’s staying power. When someone else is forced to sell and your hand is free, the same drop is their problem and your opportunity. Only deploy money you can sit on.

My take

In my previous articles, I’ve been saying AI is still in early development and this is just a correction along the way. Even the U.S. government is backing it — they can’t afford to lose this AI race to other countries. My position hasn’t changed: wait patiently for the moment, and buy the dip on semiconductors — not all at once, just cost-average in, always.

---

Not financial advice — my read on someone else’s analysis plus my own, for a family software dev who trades on the side, not a professional. Semiconductor stocks can swing 10–20% in a single session; figures here reflect mid-July 2026 and change fast. Do your own homework and only risk money you can hold through the drop.

Hebojago is my journey to financial freedom, recorded live — real experiments, real numbers, wins and mistakes.

I want to be honest about something before I give any advice: staying calm in a crash is hard, and anyone who tells you it’s easy is either lying or isn’t actually in the trade. I’m in the trade. And the last few weeks have not been fun.

So this isn’t a “just relax and think long term 😌” post. This is what actually happened to my nervous system, what I got wrong about my own numbers, and the reframe that finally brought my heart rate down.

Let me start with my own greed

I bought SOXL — the Direxion 3x leveraged semiconductor ETF — below $100. A few weeks ago it was around $300.

I’d tripled my money, and I’ll tell you exactly what I was thinking, because it’s the sentence that should have warned me: this is going to the moon. Not “this is a good position.” To the moon.

Then July arrived and it fell off a cliff. And the same brain that had been planning the moon landing was suddenly screaming at me to sell everything before I lost it all.

I didn’t sell. Here’s the thinking that got me there — including the part where I realized my panic was doing arithmetic from the wrong number entirely.

What actually happened to semis

For the first half of 2026, semiconductors were untouchable. The Philadelphia Semiconductor Index (the SOX) was up roughly 65% in six months and posted what was, by most accounts, its best quarter on record. If you owned chips, you were a genius. Everyone was a genius.

Then the music stopped:

• The SOX has shed more than 20% from its June peak in about three weeks.
• Early data suggests July alone erased roughly 35% of the sector’s entire 2026 gain.
• Micron is down around 17% from its peak. Intel fell 21% in seven trading days. SK Hynix had its biggest single-day drop on record.
• Investors pulled roughly $7 billion out of growth funds in a single week, rotating into value.

The cause isn’t chip demand collapsing. It’s a valuation reset: Wall Street questioning whether the enormous AI infrastructure spending pays off, reports of SK Hynix slowing its high-bandwidth memory expansion, and a more hawkish Fed under new Chairman Kevin Warsh.

A 20% drop after a record quarter is not the same event as a business breaking. Those two things feel identical at 2am and are completely different in reality.

Why your brain is screaming (and why that’s normal)

Here’s the part the “stay calm” gurus skip: your panic is not a character flaw. It’s wiring.

Loss aversion means a dollar lost hurts roughly twice as much as a dollar gained feels good. So when a position I was up big on gives back a chunk, my brain doesn’t process it as “still up a lot.” It processes it as pain, right now, make it stop. That’s not weakness. That’s a 200,000-year-old survival circuit doing its job in an environment it was never built for.

Knowing that doesn’t make the feeling go away. But it does something useful: it lets me treat the urge to sell as information about my emotional state, not as information about my investment. Those are two different data streams, and the entire game is not confusing one for the other.

The $300 that was never mine

Here’s the trick my brain played on me, and I think it plays it on almost everyone.

I bought below $100. Even after this crash, I’m still up. I have not lost a single dollar of the money I put in. By any honest accounting, this is a winning position.

So why did it feel like a catastrophe?

Because the moment SOXL touched $300, my brain silently moved the goalposts. $300 became the new baseline — the “real” number, the money I had — and everything after felt like theft. But I never sold at $300. That number was never in my bank account. It was a screenshot, not a paycheck.

Panic usually isn’t caused by losing money. It’s caused by measuring from the peak.

Once I recalculated from my actual cost basis instead of my highest-ever screenshot, my heart rate dropped without me doing anything else. Same position, same market, completely different feeling. The only thing that changed was where I put the zero.

That’s not a trick to make yourself feel better about a bad position — if I were underwater on my cost basis, this section would say something very different. It’s about making sure the number you’re panicking about is a real one.

The honest part: I don’t own semiconductors

Now the section that costs me something to write.

I keep saying “AI is still growing, this is just noise,” and I believe that. But I have to be careful, because that’s a thesis about semiconductors, and I don’t own semiconductors. I own SOXL.

SOXL is built to deliver 3x the daily move of a semiconductor index — and it resets that leverage every single day. This selloff showed exactly what that means:

• On June 23, SOXL fell 23% in one session while unleveraged semi ETFs fell about 8%.
• On July 1, it dropped 16.4% — from $266.71 to $223.01 — while the underlying index fell 5.68%.
• Over the past month it’s down roughly 30%.

And the amplification isn’t the sneaky part. The daily reset is. In a choppy market, rebalancing every day bleeds value even if the index ends up back where it started — volatility decay, and it’s structural, not a fee line. The fund also carries billions in notional swap exposure whose financing costs erode value every trading day, up or down.

The long-run receipts are humbling: over the past five years SOXL returned roughly 479% while the unleveraged SMH returned about 404%. Three times the daily risk. Barely any extra return.

So here’s the thing I’d want you to take from this post more than anything else: with a normal chip ETF, being right about the destination is enough. With a 3x daily-reset product, the path can beat you even when you’re completely right. A long, choppy, violent, sideways stretch — which is exactly what a valuation reset often looks like — is the specific weather this product handles worst.

Being right slowly is a losing trade in this thing. That’s not a reason to panic. It is a reason my sell trigger can’t be vibes.

My rules (decided before, not during)

The most calming thing I do has nothing to do with the crash. It’s that I made my decisions in advance, while calm, so the terrified 2am version of me doesn’t get a vote.

1. With leverage, position sizing isn’t “how much can I lose” — it’s “how much volatility can I emotionally survive.” A 7% day for the sector is a 20%+ day for me. The right question was never just about the downside number; it’s whether I can watch that number and still behave like an adult.

2. I stopped checking hourly. Watching a leveraged position tick red all day is self-harm with extra steps. The information value of an intraday chart is roughly zero; the emotional damage is not.

3. I separate “the thesis broke” from “the price dropped.” Price dropping is noise. Thesis breaking — demand actually falling off a cliff, the story I bought being wrong — is the only valid reason to sell. But per the section above, for a leveraged product I need a second trigger too.

4. I keep cash on the sidelines on purpose. Not because I can time bottoms — I can’t — but because dry powder turns a scary red day from a pure threat into a possible opportunity. That reframe is worth more than the returns.

The trader’s trick that turns fear into getting paid to wait

Quick one, because it’s genuinely calming: I sell cash-secured puts.

Stripped of jargon, a cash-secured put is a way of saying “I’d happily buy this at a lower price, and I’ll take a premium today for committing to that.” When markets fall and everyone’s scared, those premiums fatten. So a crash flips, mechanically, from “the thing hurting me” into “the thing paying me to name my buy price.”

I’m not telling you to go trade options — done wrong they will hurt you, and this isn’t a how-to. [ADD INTERNAL LINK — your cash-secured puts explainer, e.g. /post/cash-secured-puts-explained/] The point here is psychological: owning a mechanism that gets better when things get scarier is the most effective anti-panic tool I have.

When “staying calm” is actually denial

A calm post that only says “hold” is a dangerous post, so here’s the counterweight.

Calm is a virtue right up until it becomes an excuse to ignore reality. Diamond-handing something whose story has genuinely fallen apart isn’t discipline — it’s the sunk-cost fallacy in a motivational t-shirt. Sometimes the disciplined, unemotional move is to sell.

The test I use: Am I holding because the thesis is intact, or because selling would mean admitting I was wrong? If it’s the second one, that’s not calm. That’s fear wearing a better outfit.

Before you copy me: please read this

I know how this post could be misread. “Guy turned <$100 into $300 on SOXL” is exactly the kind of sentence that makes people open a brokerage app and skip everything else.

So, plainly: 3x leveraged ETFs are not buy-and-hold investments. Direxion itself says these products aren’t suitable for all investors and are meant for people who understand leverage risk and actively manage their positions. They’re built as short-term tactical tools. I bought early into a historic run and got a big move in my favor — that outcome says more about timing and luck than about skill, and the same product on a different path could have cut my money in half instead.

If you want exposure to the AI/semiconductor story without the daily-reset decay, unleveraged funds exist and track the same companies. That’s not a recommendation — it’s just the part of the menu people skip when they’re excited.

What I’m actually doing

The boring truth is that most of “staying calm” is infrastructure you build before the storm: right position size, rules written down, cash set aside, the discipline to check less — and knowing precisely what you own, not the thing you wish you owned.

Do that work in the sunshine, and a crash becomes something you sit through instead of something that happens to you.

I’ll keep updating this as it plays out. That’s the whole point of doing this in public.

---

This is my personal experience and my own math, not financial advice, and I’m not a financial advisor. I hold a position in SOXL, so read me with a skeptic’s eye. Leveraged ETFs carry substantial risk including rapid and permanent loss; they are designed for daily exposure, not long-term holding, and are not suitable for every investor. Nothing here is a recommendation to buy or sell any security. All market figures cited were accurate around mid-to-late July 2026 and change constantly — verify current numbers and read the fund prospectus before acting on anything.

Hebojago is my journey to financial freedom, recorded live — real experiments, real numbers, wins and mistakes. Follow along here.

I spent the weekend refreshing news apps like everyone else. Korea’s market was closed for the holiday while the rest of the world sold semiconductors hard, and the jokes started immediately — the holiday saved the KOSPI, let’s make Monday a holiday too.

Funny. Also not funny, because everyone typing that is quietly dreading opening their brokerage app.

So let me do what I always do here: no hype, just math. Which of these headlines can actually change the direction of my holdings, and which ones are just loud?

Everything below is as reported in Korean and US coverage over the weekend. I have not independently verified the individual figures — check them before you act on any of it.

What actually happened

Thursday was already ugly in Seoul. SK Hynix dropped 11.53% to ₩1,842,000. Samsung Electronics fell 8.77% to ₩255,000. Then the market closed for the holiday.

The damage didn’t stop, it just moved somewhere Korean investors couldn’t touch it:

• SK Hynix’s US-listed ADR fell 13.69%, from about $176 to $152.
• Kioxia lost a US patent case and a $229 million judgment — down 15% Thursday, another 16% Friday. Roughly 29% in two sessions.
• The Nikkei was down more than 6% intraday.

The holiday didn’t erase any of that. It stacked three days of fear into a single Monday open. That’s the whole reason this weekend felt heavier than usual.

Why it fell: plumbing, not fundamentals

Here’s the part that matters more than any single number. The diagnosis most analysts landed on — domestic and abroad — is flows, not earnings.

The chain works like this. Semiconductors ran for months, so leveraged buyers piled in. Money started rotating out of the sector. Prices slipped. Leveraged positions got force-sold. The forced selling pushed prices lower, which triggered the next tier of liquidations. Repeat for several days.

Nothing in that sequence is a company problem. It’s a money problem.

We saw the identical movie days ago, when the KOSPI fell more than 8% in a session and tripped a circuit breaker. Earnings didn’t change that morning. Leverage unwound. And when the index shakes, foreign selling concentrates almost entirely in the same two names — Samsung and Hynix — because that’s where the liquidity is.

To be fair: this is a diagnosis, not a proven fact. But when several independent sources point at the same mechanism, it’s the working theory I’d start from.

Monday will open down. How far down, nobody knows.

Direction is the easy call — three days of offshore damage arrives at once, so a gap lower is likely.

Size is genuinely unknowable, and here’s the specific reason: SK Hynix’s ADR is trading at more than a 15% premium to the Seoul shares. So which one is the “real” price? Do the Seoul shares climb toward the expensive ADR, or does the ADR fall toward Seoul? Nobody can tell you, which means anyone quoting you a specific percentage drop is guessing with confidence.

One thing does argue against a repeat of the circuit-breaker cascade, though. In Friday’s US session, sellers hit hard early and then the market clawed most of it back by the close. Panic supply got absorbed. On the circuit-breaker day, selling pressure went one direction all day and nothing pushed back. Friday was two-sided — frightened sellers against bargain hunters.

Down open, yes. Automatic collapse, not necessarily.

And the flip side: “probably not a crash” is not the same sentence as “it’s going up.” Whatever number Monday opens at, that gap is a price, not a forecast.

Noise vs. signal

This is the section I’d read twice. Three headlines dominated the weekend. Two are noise. One got read backwards.

🔇 NOISE — Kioxia losing half its value

The fear is obvious: if a memory company halves, why not ours?

Because Kioxia’s collapse had two specific causes that don’t transfer. A lost US patent suit — a legal problem, not an industry problem. And near-total dependence on NAND, one product line with nowhere to hide.

Samsung and SK Hynix run DRAM, NAND and HBM. When one leg wobbles, the others carry weight. Scary chart, wrong company. This doesn’t change our direction.

🔇 NOISE (read backwards) — China’s Kimi K3

Moonshot unveiled Kimi K3 at an AI conference in Shanghai, reportedly near the top tier globally and closing the gap with the leading US models. China isn’t just competing on cheap-and-good anymore.

The market’s first instinct was the DeepSeek flashback: if China can do AI cheaply, expensive chips sell less.

Take it apart and it inverts. Frontier-scale models are enormous — K3 reportedly needs a full top-end NVIDIA server to run. Bigger, better models eat more HBM, not less. Efficiency gains have historically increased total consumption, not reduced it. More serious players in the frontier race means more memory demand, not less.

Logically bullish. Whether the market reads it that way this week is a separate question — logic and price often disagree for a while.

🔊 SIGNAL — the two stocks that went up on Friday

While semiconductors bled, two names rose: Western Digital and Seagate. Hard drive makers.

Why? Every bit of data AI generates has to physically land somewhere, and high-capacity drives are reportedly so tight that 2026 production is already spoken for. Several banks raised price targets on the storage names that same day.

That’s the market quietly buying the AI-infrastructure thesis in the corner of the market nobody was panicking about. If the warehouse is sold out, the question becomes whether the compute-side memory that Samsung and Hynix build follows the same path.

That question gets tested this week.

Why Google and SK Hynix earnings are the whole ballgame

Two dates: Google on the 22nd, SK Hynix on the 29th. Here’s the logic chain for why those two specifically decide this.

The drop was caused by flows, not fundamentals. Flow-driven drops reverse when real buyers show up — and real buyers only show up if the underlying demand is actually there. So the only question that matters is whether AI demand is still real. And demand doesn’t announce itself in headlines. It shows up as numbers, in the accounts of the people spending and the people selling.

Google (July 22) is the spending side. Google is one of the largest AI capex spenders on earth. If they confirm they’re maintaining planned AI investment, the demand feeding memory orders is intact and this selloff was a flow adjustment. If they trim capex, the entire thesis needs rewriting — that’s not noise, that’s the signal.

SK Hynix (July 29) is the selling side. Google can promise spending; Hynix’s results show whether it’s actually arriving as revenue. If the numbers meet expectations, demand is confirmed in real money rather than in guidance language.

Spending intent, then delivered revenue. Both green and this was a scare. Either one red and we’re talking about something else entirely.

What the people in the room are saying

Chey Tae-won, SK Group Chairman, at a business forum: AI is still a four-year-old, and for that child to grow up, it needs memory continuously. If AI is only four, the growth is ahead of it, and growing requires more memory every year. Long horizon, demand goes up and to the right. His advice was to hold rather than trade in and out.

But don’t skip the sentence right after it. He also said plainly that he has no idea what the stock does next month. Long-term demand and next month’s price are two different axes — one is demand, the other is flows.

Brokerages are split roughly in half: one camp calls a 10%+ drop a buying opportunity, the other warns of volatility all summer. The interesting part is that many of them raised price targets while the price fell. Price down, target up looks contradictory until you realize it’s them saying the same thing I’ve been saying: this is a flow problem, not an earnings problem.

Translation: the professionals don’t know what next week does either.

Quick answers

Will Samsung or Hynix halve like Kioxia? Kioxia’s fall came from a patent loss and NAND-only concentration. Our two run DRAM, NAND and HBM. Different situation.

Is Kimi K3 actually bullish? Logically yes — frontier models consume more HBM. Whether the market prices it that way this week is separate.

Is this the start of a second crash? The 22nd and 29th answer that, not Monday’s open.

Which one falls harder, Samsung or Hynix? Hynix has heavier HBM exposure, so it swings more with AI sentiment. Samsung’s business is broader, so it usually falls less. Thursday showed it exactly: Hynix −11.5%, Samsung −8.8%.

My take

I think AI demand is still high. That’s the part I’m confident about.

As more people use AI every month, the demand for the hardware underneath it goes up with them — memory, data centers, and the electricity to run all of it. How much? I don’t know. Nobody does, and I’m not going to pretend otherwise with a fake number. But the direction is not in doubt to me. More users means more compute means more memory means more power.

So I’m holding my shares, and I’m buying the dip.

With one condition I’d repeat to anyone: cost average. Don’t go all in at once. Nobody knows where Monday opens or where the bottom is — including me. Splitting your buys across weeks is how you participate in the upside without needing to be right about the timing. Going all in on a single scary morning is a bet on your own precision, and the last few sessions have shown how well precision works in this market.

Buy in pieces. Keep dry powder. Watch the 22nd and the 29th.

The short version

Monday probably opens lower, and nobody can tell you how much. This drop looks like a flow problem, not a fundamentals problem. Kioxia is noise, Kimi K3 is misread as bearish, and the storage stocks quietly rising are the real tell. The actual verdict lands on July 22 with Google and July 29 with SK Hynix.

Ten scary headlines this weekend. One of them can change the direction. The rest is volume.

If you want to run your own numbers before Monday, my cost-averaging math is the same logic I use on the mortgage and car payment calculators — small consistent inputs beat one heroic decision.

---

Not financial advice. I hold shares in the companies discussed, so read this with your skeptic hat on. All figures above are as reported by outside coverage and were not independently verified — market data, earnings dates and prices change, so confirm current numbers before acting. Your results depend on your own situation.

I don’t usually write about what billionaires are doing. It’s the laziest content on the internet — “Buffett bought X, so buy X” — and it’s made more people poor than rich.

But this one is different, because it’s a contradiction, and contradictions are where the actual lesson lives.

Here’s the setup. A 95-year-old man sat down with CNBC on July 15 and basically said there’s nothing worth buying in this market. His company is sitting on the biggest pile of cash it has ever held. And in the same interview, he volunteered that he personally started a $31 billion position in Google.

Both of those are true. Let’s do the math on why.

Fact 1: he really is sitting on the sidelines

Berkshire Hathaway ended Q1 2026 with $397.4 billion in cash and short-term Treasury bills — an all-time record, past the previous $381.7B high from Q3 2025. Roughly $52 billion of that is actual cash; the rest is parked in T-bills.

That detail matters more than the headline number. That money isn’t sitting in a mattress losing to inflation. It’s lending short to the U.S. government and collecting interest every single month. Which means waiting costs him almost nothing. He can say “no” for three straight years and still get paid to say it.

And he’s been saying no. In Q1 alone Berkshire sold $24.1 billion of stock and bought $16 billion — a net $8.1 billion of selling, in Greg Abel’s first quarter as CEO. Apple, once close to half the equity portfolio, is down to roughly 22% of it — still the largest holding by far, just half the weight it used to carry.

Fact 2: why he says there’s nothing to buy

Buffett’s language in the interview was blunt: it’s hard to find value when everybody around you would rather gamble. He’d used the same frame at the May shareholder meeting, describing the market as a church with a casino attached.

That metaphor is doing real work, so let me unpack it the way I’d explain it to my wife over coffee:

• The church crowd buys a business. They’re asking what this company will earn over the next ten years, and whether today’s price is cheap against that.
• The casino crowd doesn’t look at the business at all. They’re betting on whether the number goes up today.

His complaint isn’t that the casino exists. It’s that the casino got so attractive that church people keep wandering over. He singled out same-day options — contracts that settle by the closing bell — and said flatly that’s not investing and it isn’t even speculating. It’s gambling.

Here’s the mechanism that actually matters: money that only cares about today doesn’t care what a company earns in 2036. When enough of that money floods in, prices start moving independently of the underlying business. The company is the same as last year; the stock isn’t.

Run Buffett’s one ruler — is this price cheap against the next decade of earnings? — against a market priced by people who aren’t thinking about the next decade, and everything comes back expensive. Not some things. Everything. That’s how you end up with $397 billion and nothing to spend it on.

I feel this one personally, by the way. I sell cash-secured puts. I’m not going to pretend I’ve never been on the casino side of the room.

Fact 3: the one exception

Now the part that broke people’s brains.

Berkshire first showed an Alphabet position in its Q3 2025 filings, and it’s grown ever since. The stake is now north of $31 billion — behind only Apple and American Express. Less than a year in, and it’s already a top-three holding.

Everyone assumed this was Greg Abel’s fingerprint. New CEO, new era, tech bet, makes a nice story. Becky Quick asked him directly whose idea it was.

Buffett’s answer: “I initiated it.” He added that he isn’t doing anything Abel doesn’t approve of and vice versa — but that Abel is the decider going forward.

So the man who says the market is a casino personally started a $31 billion position in a company shoveling hundreds of billions into AI. That’s the contradiction. And it dissolves the second you look at when he bought.

The order of operations is the whole story

Most people only saw the $10 billion headline from June. But roughly $21 billion of that stake was accumulated before it, starting Q3 2025 — at open-market prices, no discount, no special terms. Same prices you and I could have paid.

Then in June, Alphabet announced an $84.75 billion equity raise (upsized from $80B after demand ran hot), and Berkshire took $10 billion of it as a private placement: Class A at $351.81, Class C at $348.20 — modestly below where the stock was trading.

Read that sequence again. He didn’t buy because he got a deal. He got a deal because he was already buying. The discount was a tip, not the meal.

Why Google needed the money at all

This part genuinely surprised me, so here are the numbers straight from Alphabet’s own filings and reporting:

• Operating cash flow for the year ending March 2026: about $174 billion
• Planned 2026 capital expenditures: $180–190 billion — and management says 2027 goes higher
• 2022 capex, for scale: about $31 billion

So one of the most profitable companies on earth is planning to spend more on data centers and AI chips in a single year than its entire operating cash flow. It had already issued over $85 billion of debt in the prior year. Still not enough — hence selling stock.

Issuing stock isn’t free money, by the way. There’s no interest and nothing to repay, but every new share thins out the slice existing owners hold. Same pie, more forks. Google accepted that trade anyway. That tells you how badly it wants the compute.

The 2008 comparison that changed my read

Here’s the detail that made this click for me.

When Buffett put $5 billion into Goldman Sachs during the 2008 crisis, he didn’t buy common stock. He took preferred shares paying a fat dividend, plus warrants to buy more later at a set price. He built himself an airbag. He wasn’t betting on Goldman so much as on his own terms.

This time? Plain common stock. No preferred, no warrants, no dividend guarantee — and notably, the same Alphabet raise did include mandatory convertible preferred shares, which went to public investors rather than to him.

He took the simplest, least-protected instrument available. You only do that when you think the business itself is going to be fine.

But he hedged, and the hedge is the important part

Buffett didn’t sell this as a slam dunk. Two caveats came straight out of his mouth:

One: he said he doesn’t like Alphabet as much as at least four or five other businesses Berkshire owns. That’s a ranking statement, not a rejection — he’s comparing it to wholly-owned operating businesses.

Two: on the AI arms race, he pointed out that everyone is pouring hundreds of billions in, and that’s real money.

That second one is the crux. The spending is certain. The return is not. Money leaves the account this year; whether it comes back as revenue is unknown. That asymmetry is the exact trade Buffett has spent seventy years avoiding.

So the honest summary of his position: high confidence in the company, open question on the AI math.

Now the pushback — because “Buffett bought it” is not a thesis

“His cash pile means a crash is coming.” I don’t buy it. He was at $381.6 billion back in Q3 2025 and the market went on to make new highs. Cash isn’t a prediction, it’s readiness. Treating it as a crash clock could keep you out of the market for years. [LINK — related Hebojago post on market timing]

“So I should just buy Google too.” Notice what actually happened: about two-thirds of his position was bought at ordinary market prices, like anyone could. His edge wasn’t access. It was being early and being willing. That’s not a thing you can copy after the news breaks.

“He’s not even CEO anymore.” Half true. Abel is the decider now, and Buffett said so plainly. But Buffett also said he started this one himself and that neither acts without the other’s approval.

“Buffett bought it, so it’s safe.” This is the one I want to be loud about: no. In 2011 he put about $10.7 billion into IBM — roughly 64 million shares near $170. He sold out over 2017–2018 in the $140s. Over that stretch IBM fell while the broad U.S. market roughly doubled. Someone who ignored him and bought an index fund did dramatically better. He is allowed to be wrong, and he has been, in this exact category.

The chart bit — take it lightly

The video I built this from made a technical argument I want to flag rather than assert: Alphabet had been capped by a descending trendline for months, broke above it on July 15 (the interview day) with a ~3% move, then gave it back on July 16 on Gemini-related news.

I checked directionally and the July 16 drop is real. The precise levels and moving-average claims I could not fully verify, so treat them as the video’s read, not as fact. If you care about that stuff, the only question worth asking is whether a broken level holds when it’s retested. Everything else is storytelling.

What I’m actually watching, in plain terms

Two numbers, both public, both free:

1. Does Berkshire’s cash pile start shrinking? It’s disclosed every quarter. If $397.4 billion starts dropping, it means he found something. That’s a far better signal than any prediction.
2. Does Alphabet’s AI spending show up as revenue? Cloud revenue growth in the quarterly report is the scoreboard. Buffett’s one reservation was that the outflow is certain and the inflow isn’t. Several quarters of that converting means his caution was misplaced. Several quarters of it not converting means it wasn’t.

The takeaway I keep coming back to isn’t about Google at all. It’s that investing isn’t a game of guessing better than everyone else. It’s a game of not buying when there’s nothing to buy, and buying when there is. A guy who held $397 billion for years and then wrote one $31 billion check is a live demonstration of both halves.

---

Not financial advice — I’m a software developer who writes about his own money, not an advisor. Every figure here was verified against CNBC’s July 15, 2026 interview coverage, Berkshire’s Q1 2026 disclosure, and Alphabet’s June 2026 SEC filings as of July 18, 2026, and is subject to change; anything I couldn’t confirm is labeled as unverified above. Copying a billionaire’s trade is not a strategy — his time horizon, tax situation, and downside tolerance are nothing like yours or mine.