Pinned Tweet
Most people see a headline, check the stock price, and move on.
I like digging a little deeper.
Reading filings, questioning the numbers, trying to understand what the market is pricing in and what it might be missing.
Sometimes I'll share something interesting buried in a 10-K.
Other times, just my thoughts on a company, an investment, or whatever catches my attention in the markets.
I won't always be right. That's part of the fun.
I'm also here to connect with people who are genuinely interested in capital markets, investing, and understanding how businesses work.
Whether you're a student, an investor, or someone who's been in the industry for years, I'd love to connect to exchange ideas, challenge each other's thinking, and learn from different perspectives.
That's what my account is about. Let's connect 👇
1
4
182
Something you're not realizing about Appaloosa's $172M $ORCL stake: it's a photo from Sep 30.
The 13F filed Nov 16 lists 1,253,932 Oracle shares valued at $172,164,864. That line wasn't in the June 30 filing.
• Common stock, no put or call attached
• Oracle was one of four new names, with Applied Materials, Modine and PG&E
• Total 13F book: $7.56B across 24 lines
A 13F shows quarter-end holdings, not trades or today's position.
What would change the read: the next 13F, for Dec 31.
29
A year ago one customer was 67% of $CRWV's revenue. Here's what last quarter looks like.
CoreWeave's Q3 10-Q lists every customer at 10% or more of revenue:
• Q3 2026: Customer B 30%, Customer A 24%
• Q3 2025: Customer A 67%, no other customer at 10%
• Nine months 2026: Customer A 33%, Customer B 26%
So two customers were still 54% of the quarter (my math). The 10-K names Microsoft as the top customer at about 67% of 2025 revenue, but the 10-Q doesn't name A or B, and the letters can change between periods. In receivables, Customer B alone is 33%.
What would change the read: the FY2026 10-K showing a third customer above 10%.
20
You're missing who actually pays $NVDA. Four buyers were about 48% of its nine-month revenue.
Nvidia's Q3 supplement lists four unnamed direct customers at 13%, 12%, 12% and 11% of total revenue for the first nine months of fiscal 2027.
• Q3 alone: three at 14%, 11% and 11%, about 36% of $115,844M (my math)
• Two customers were 18% and 17% of receivables
• Direct customers include distributors and ODMs, so the end buyer can sit behind them
What would change the read: the Q3 10-Q, which isn't filed yet.
23
The $1.68T 2028 capex forecast going around isn't in any filing. What's in there is one quarter of spending.
The latest 10-Qs, cash spent on property and equipment:
• $AMZN $77,259M and $GOOGL $51,355M
• $MSFT $39,367M and $META $33,942M
• $ORCL $28,499M (quarter ended Aug 31; the other four Sep 30)
That's $230,422M across the five (my math, Amazon gross; Alphabet and Meta are nine months minus six). It isn't one shared quarter, and it leaves out leases signed but not yet started: $364.6B at Microsoft alone, some still subject to conditions.
What would change the read: the next round of 10-Qs showing a different run-rate.
50
Everyone's quoting $STM's 17.9% growth. One segment grew 46.2%.
STMicro's Q3 revenue was $3,757M, up from $3,187M a year earlier. The growth wasn't spread evenly:
• RFOC (space, ranging and connectivity, audio, optical and RF): $504M, +46.2%
• EMP: $1,212M, +24.1%
• AM&S, the biggest segment: $1,560M, +8.8%
The CEO names AI datacenters and LEO satellites as growth drivers. Q4 guide: about +10.5% sequentially.
What would change the read: a filing that splits out data-center revenue. ST doesn't disclose one.
14
Think $ANT lost $104B running its business? Look at what's inside that number.
Anthropic's prospectus says most of it is a remeasurement: its convertible notes rose in value as the company's valuation went up, and that change runs through the income statement.
• Net loss, nine months to Sep 30: $104,039M
• Fair-value change on notes, derivatives and SAFEs: $94,324M, 90.7% of it (my math)
• Operating loss: $1,260M
What would change the read: a quarter where the operating line, not the remeasurement, drives the loss.
1
104
You're watching the wrong number on $IREN. Three customers were 70% of last quarter's revenue.
The 10-Q for the quarter ended Sep 30 lists Customers A, B and C at 28%, 24% and 18%, all AI Cloud. None of them hit 10% a year earlier.
• The names aren't disclosed
• Microsoft's $9.7B deal runs through 2031 and lets it terminate, subject to cure, if delivery dates are missed
• Microsoft asked for a design change, and dates extend for delays it causes
What would change the read: Horizon 2–4 delivering on schedule.
Everyone's quoting Oracle's $664B. Almost nobody reads the timeline printed next to it.
$ORCL's 10-Q splits that backlog by when it expects to book it as revenue:
• Next 12 months: about 13%
• Months 13 to 36: 37%
• Months 37 to 60: 34%, the rest after year five
So half lands within three years and about 16% comes after year five (my math). That's expected revenue recognition, not cash, and variable consideration is left out under an optional exemption. A year earlier the backlog was $455B; the 10-Q puts the jump mostly on "certain significant cloud contracts."
What would change the read: the next 10-Q moving a bigger share into the 12-month bucket.
Micron's "75% of 2027 sold out" isn't in its 10-K. What is in there tells you how binding those deals are.
$MU has multi-year take-or-pay agreements with binding volumes, and customers put up cash to secure supply. But that cash goes back to them.
• Customer deposits on the balance sheet: $12,895M, up from $142M a year earlier
• Nearly all of it is scheduled to be repaid between 2029 and 2031 if customers meet their minimums
• Remaining performance obligations: about $134B, roughly a quarter expected as revenue in the next 12 months
Micron's own note says that $134B is "not expected to be indicative of future revenue under these contracts." And most of the agreements price inside min/max bands, so pricing isn't fully locked either.
What would change the read: a filing that puts a number on how much of 2027 supply is committed.
18
Something you're not realizing about $NBIS: customers prepaid it $3.4B last quarter, and operating cash flow still went negative.
Nebius only prints nine-month cash flows, so Q3 is nine months minus six (my math):
• Operating cash flow: -$273.0M
• Deferred-revenue inflow: $3,434.0M. Without it, OCF was -$3,707.0M
• Cash capex: $7,300.4M
Financing brought in $16,618.0M over the nine months.
What would change the read: a quarter where OCF after prepayments covers capex.
1
105
You missed the line next to $CRWV's debt. It's another $19.3B.
CoreWeave's debt principal was $47,505M at Sep 30. Its lease liabilities sit outside that total: $18,697M operating plus $649M finance.
• Debt plus leases: $66.9B (my math)
• Cash and equivalents: $4,122M
• Signed leases not yet commenced: another $47.4B, undiscounted, off balance sheet
What would change the read: how much of that $47.4B starts in 2026 versus 2029.
1
19
$411M. That's all Intel Foundry sold to outside customers last quarter.
The rest of $INTC's foundry revenue is Intel selling wafers and packaging to itself.
• Foundry revenue, Q3 2026: $6,697M
• From external customers: $411M, up from $32M a year ago
• Foundry operating loss: -$1,853M
And the jump isn't a run rate. The 10-Q says it was "primarily due to revenue recognized under a non-recurring foundry contract." Year to date, most of the outside growth is Altera, which became an external customer after Intel deconsolidated it.
What would change the read: a quarter where external revenue grows without a one-time contract.
1
30
You're reading $NVDA's old Singapore number wrong. It was $10,156M.
That's what customers were billed in Singapore in the quarter ended Jul 27, 2025, 22% of $46,743M in revenue. The same 10-Q said the products were "almost always shipped elsewhere."
• Over 99% of controlled Data Center compute billed there was ordered by U.S.-based customers
• The next 10-Q switched to customer HQ and dropped the Singapore line
• Recast, that quarter's U.S. revenue is $32,897M, not $23,470M
What would change the read: a filing showing how much actually shipped to Singapore.
1
2
68
Quoting $AMD's 22.4% China share? Check what it measures first.
It's FY2025, not 2026: $7,751M of $34,639M in sales, by where the customer is billed, with Hong Kong included. It's not end use, and it doesn't split GPUs from CPUs.
• FY2024: 24.2%. FY2023: 15.1%
• The latest 10-Q gives no China figure at all
• 2025 also carried ~$440M of net MI308 export-control charges
What would change the read: AMD's FY2026 10-K geography table.
1
1
62
Everyone's quoting $CSCO's 24% growth. The next line goes the other way.
Cisco's Q1 FY27 revenue was $18,517M, up from $14,883M a year earlier. Its Q2 guide is $18.0–$18.2B, so even the top of the range sits below the quarter it just reported.
• Low end: $517M below Q1. High end: $317M below
• GAAP product gross margin: 64.5% → 61.8%
• Full-year FY27 revenue guide: $74.8–$75.8B
The release doesn't say why Q2 steps down.
What would change the read: a revised Q2 range, or Q2's actual print.
13
Something you're not realizing about $MU: it now spends 13 cents to make $1 of sales.
In Q4 FY26 (ended Sep 3), Micron's GAAP cost of goods sold was $7,182M on $54,229M of revenue. At the FY23 trough, costs were above sales.
• Q2 FY23 (ended Mar 2, 2023): $4,899M of costs on $3,693M of revenue, $1.33 per $1
• That quarter included a $1.43B inventory write-down
• Same quarter three years earlier, Q4 FY23: $1.11 per $1
What would change the read: the Q1 FY27 10-Q.
Today, $NVDA reports earnings.
Last quarter:
• Revenue: $96.2B (+106% YoY)
• Gross margin: 75%
• Data Center revenue: $89B
The company now expects $108B in quarterly revenue.
But there's another number worth watching.
In the first six months of FY27, receivables used $24.6B of cash. NVIDIA attributes part of the increase to extended payment terms for certain large customers.
That's not proof of weak demand. Operating cash flow still reached $74.4B.
But it raises a useful question:
As NVIDIA sells more AI infrastructure, how quickly does that revenue turn into cash?
Tomorrow, I'll watch receivables, cash flow and gross margins alongside the headline revenue number.
Growth matters. So does the quality of that growth.
2
166
Microsoft made $10 billion in revenue from OpenAI in one quarter.
OpenAI also owed Microsoft $7.2 billion at quarter-end.
Three months earlier, that figure was $6 billion.
Now consider the relationship.
Microsoft invests billions into OpenAI.
OpenAI spends billions on Microsoft's infrastructure.
Microsoft records the revenue.
There's nothing inherently wrong with this.
But there's a difference between recognizing revenue and collecting cash.
And when your customer is also a company you've invested billions into, that distinction becomes worth paying attention to.
The question isn't whether the revenue is real.
It's whether the cash eventually follows.
That's what I'll be watching in the next filing.
$MSFT
2
33


