The last trading day was Wednesday, September 9. All three indices fell: the Dow Jones Industrial closed at 52,380.66, down 0.77%; the Nasdaq Composite at 26,256.66, down 0.62%; the S&P 500 at 7,637.70, down 0.47%. The daily star was Marvell Technology (MRVL), up 4.26% at $235.01 — while the Nasdaq semiconductor group averaged just +0.98% the same day and Nvidia (NVDA), Broadcom (AVGO) and KLA (KLAC) all closed lower. After the US close tonight, Oracle (ORCL) reports the first quarter of its new fiscal year. Today's Academy covers asset-light versus asset-heavy: the same room, the same bottle, the same phone — the company whose name is on it and the company that built it are often two different businesses. All figures are as of the September 9, 2026 US close, and all times are UTC.
The September 9 Close: All Three Indices Fell, and the Dow Led
Not one of the three indices closed green on September 9. The Dow Jones Industrial fell 0.77% to 52,380.66, the largest drop of the three; the Nasdaq Composite fell 0.62% to 26,256.66; the S&P 500 fell 0.47% to 7,637.70, the smallest decline.
The push came from escalating geopolitical conflict lifting oil. Brent crude rose nearly 3% at one point to about $100.8 a barrel; the market read that as inflation risk, and the 30-year mortgage rate climbed to about 6.85%, its highest since June 2025.
Oil, inflation expectations and mortgage rates hang together because they point the same way: input costs move up, the discount rate follows, and the discount rate decides what future cash flows are worth today. So when crude jumps nearly 3% in a session, what falls is not only energy users but everything priced off the future.
August PPI and last week's jobless claims land at 12:30 tonight. But the week's real macro focus is August CPI at 12:30 tomorrow: July ran 4.7% at the producer end against just 3.4% at the consumer end, a gap of 1.3 points. That gap is itself a piece of information — production costs are rising faster than retail prices, and what gets squeezed in between is corporate gross margin rather than the consumer's wallet. Which end moves first decides how this round of inflation should be read.
Tonight's two protagonists were quiet yesterday: Oracle (ORCL) closed at $161.70, down 0.50%, already more than 50% below its 52-week high of $345.72; Adobe (ADBE) closed at $254.86, down 0.93%. Neither reported yesterday, and neither moved after the bell.
Daily Star Marvell (MRVL): What Moved Was One Interview
Marvell Technology (MRVL) makes custom silicon and interconnect for AI compute, at a market value of $205.8 billion. On September 9 it closed at $235.01, up 4.26%, adding about $8.4 billion of market value in a session. The price sits about 64% of the way up its 52-week range, still 40% below the high, after a cumulative gain of roughly 241% over the past year.
What moved it that day was not results. It was a television interview. The chief executive raised the two-year revenue target from $23.5 billion combined to $30 billion — $6.5 billion added in one sitting — with data centre alone guided above $15 billion for 2027, against about $2 billion in 2023.
The nature of that statement has to be stated first. It came from management on television. It is not published company guidance and it has not entered any filing; the source card says plainly that the line-item figures only get disclosed at the investor day and in later reported results. That does not make it untrue. It means the figure currently has no verifiable breakdown behind it — it is a direction, not a set of accounts. Reading news like this, asking who said it, where, and whether there are line items matters more than remembering the headline number.
The five-dimension score fills in the rest of the day. Peer ranking at 90 and valuation temperature at 85 hold up two corners; trend position at 64 and relative strength at 66 both sit mid-range; volatility control at 30 is the only dimension below a passing mark.
Low volatility control means this name swings wider than its peers by construction. It does not say anything is wrong with the day's gain. It says the same 4.26% carries a different meaning on a name scoring 30 than on one scoring 80. On the first it is closer to ordinary movement; only on the second would it be exceptional.
One Industry, 7.5 Points Between the Ends
Line up six semiconductor names from the same session and the picture is not one-directional: Marvell (MRVL) +4.26%, AMD (AMD) +3.04%, Micron (MU) +2.75%, while Nvidia (NVDA) fell 0.91%, Broadcom (AVGO) fell 1.13% and KLA (KLAC) fell 3.21%. The ends sit 7.5 points apart.
The Nasdaq semiconductor group averaged just +0.98% that day. That number matters, because it means this session cannot be written up as "semiconductors rallied". If the whole group had been rising, Marvell's 3.3-point outperformance would not exist; it is precisely because the group barely moved that the single-stock gain says something.
Three up and three down inside one industry says what got priced was not the label "semiconductor" but each company's own orders and news. Broadcom also makes custom silicon and still closed lower — so even "custom silicon rallied" does not hold. The accurate description is that names moved on their own orders and news.
Supporting name Micron (MU) at +2.75% ran a different line entirely. A broker judged that the worst of the memory downcycle may be behind it, and SK Hynix rose about 5% with SanDisk about 3% the same day — a whole chain reacting together, with nothing to do with Marvell's interview. Two gains on the same day, two unrelated causes, and they should be counted separately.
The One-Minute Concept: Reading a Warrant Comes Down to Three Numbers
When a large customer signs a long contract, the supplier sometimes attaches a warrant — the customer's right to buy the supplier's stock at a set price. It does not vest on signature; it unlocks in tranches as purchase thresholds are met.
The one Marvell disclosed in August is a good teaching case. Struck at $206.58 for about 59 million shares, it vests in full only after the customer's cumulative purchases reach $120 billion, in 240 tranches of $500 million each, running to January 2033.
Three numbers are enough to read terms like these.
First, how high is the threshold. $120 billion is the bar for full vesting, and splitting it into 240 tranches means each $500 million of purchases unlocks a small piece. The higher the bar, the further full vesting sits from reality; the finer the tranches, the more the terms vest gradually rather than as an all-or-nothing switch.
Second, how long is the term. January 2033 is more than seven years out. A long term gives the customer more time to build the volume, so the terms are likelier to be reached. Conversely, a warrant with a very high bar and a short term is worth close to nothing in practice.
Third, how much dilution. The new shares once fully vested amount to roughly 7% of the existing count. This is the item most often skipped: market value went up, but each share now represents a slightly smaller slice. When reading a company's market value, if terms like these sit on it, the dilution belongs in the arithmetic — otherwise the slice you compute for yourself comes out too large.
Put the three together and the nature of the instrument is clear: it is a long-term arrangement binding customer and supplier together, not a one-off equity payment. The more the customer buys, the more stock it earns, and the more aligned the two sides become — which is why terms like these have become more common across the compute supply chain.
What to Watch Tonight: Oracle (ORCL) After the Close
Oracle (ORCL) reports after 20:00 UTC, with the call at 21:00. Adobe (ADBE) also reports after the close. Tonight covers the first quarter of its new fiscal year, ended in August.
Its place in the AI chain is unusual: it does not make chips, it buys compute and packages it into cloud services to rent out, while still holding a decades-old database business. So it is neither a picks-and-shovels name like Nvidia nor a pure software company — it is the layer in between, converting capital expenditure into rental revenue.
The core metric is cloud infrastructure (IaaS) revenue per fiscal quarter, not total revenue. The four fiscal 2026 quarters ran $3.35B, $4.10B, $4.90B and $5.80B, growing 55%, 68%, 84% and 93% year over year. The scale nearly doubled and the slope is still steepening.
Why not total revenue? Because a third of it is still flat software licensing, which dilutes the AI slope. A quarter with 10% total revenue growth might be "IaaS up 90%, licensing down 2%", or it might be "both up 10%" — two situations that mean entirely different things for the company, and total revenue cannot tell them apart.
The first figure to check tonight is whether that line defends growth above 90%. Company guidance puts total cloud revenue growth at 57% to 63%, steeper than last quarter's 47%.
⚠️ One trap worth naming: 93% and 57%–63% are not the same measure. The 93% is last quarter's reported growth for the cloud infrastructure line alone; the 57%–63% is this quarter's guidance for total cloud revenue, IaaS plus SaaS. Different denominators — set them side by side and you reach the false conclusion that growth is decelerating.
The Largest Line Is the One That Is Shrinking
Break last quarter's $19.2 billion of revenue into five lines and the structure shows: cloud infrastructure $5.8B, up 93%; services $1.5B, up 13%; cloud applications $4.1B, up 10%; hardware $0.9B, up 9%; and software licence and support $6.8B, down 2%.
The largest line is the only one shrinking. Software licence and support at $6.8 billion is a billion larger than cloud infrastructure, and it is in decline. That means essentially all of the growth rests on cloud infrastructure alone — if that line slows, the company stalls immediately, because there is no second line to catch it.
So tonight is not about whether total revenue clears a bar; it is about whether that one line steps up again. A quarter that beats on total revenue while IaaS grows only 70% is in fact more worrying than a flat total with IaaS at 95%.
The second line to watch is where the money comes from. Remaining performance obligations — the signed but not yet recognised contract value, the backlog — stood at $638 billion at the end of last quarter, up $85 billion in that quarter alone, an order of magnitude larger than a year of revenue. Watch whether tonight adds a comparable amount, and whether the company frames how quickly those contracts convert into revenue.
Against that sits the spending side: capital expenditure ran $55.7 billion last fiscal year against negative $23.7 billion of free cash flow. That is the shape of the rented-compute business — revenue arrives in instalments while the data centres and the chips are paid for up front. So the third thing to check is whether the capex framing and the financing arrangements are revised upward alongside the orders. Orders growing without spending is a delivery problem; spending growing without orders is a returns problem.
Academy: What You Bought Was Probably Not Made by the Company on the Label
Thursday's theme is asset-light versus asset-heavy, and it opens with one line: the same room, the same bottle, the same phone — the company whose name is on it and the company that built it are often two different businesses.
A night in a hotel. Marriott (MAR) ended 2025 with 9,805 hotels and about 1.78 million rooms in its system, and owned or leased under 1% of them. What it sells is the brand, the booking system and the loyalty programme. The buildings belong to landlords like Host Hotels (HST), which owns 76 hotels and about 41,700 rooms, is barred by REIT rules from operating any of them, and drew 62.9% of its 2025 hotel revenue from properties carrying the Marriott name. One hotel, two companies, two businesses.
A bottle of Coke. Coca-Cola (KO) sells concentrate and syrup to franchised bottlers, and states in its filings that those bottlers are independent contractors, not its agents. The filling, warehousing and delivery belong to Coca-Cola Consolidated (COKE) — the largest independent Coca-Cola bottler in the US, with 13 plants and 80 distribution and warehouse sites covering 14 states and 65 million people.
A phone in your hand. Apple (AAPL) states in its filings that outsourcing partners do substantially all of its manufacturing, with production concentrated in Asia. The chips are made by TSMC (TSM), which only makes chips for others, spent about $40.6 billion on capital expenditure in 2025, and drew 78% of that year's revenue from its ten largest customers.
Three pairings, six companies, and each pairing is two halves of the same thing.
One Half Owns the Name, the Other Owns the Capacity
Compute property and equipment as a share of total assets for all six and the dividing line is remarkably clean: Marriott (MAR) 10.5%, Coca-Cola (KO) 10.8%, Apple (AAPL) 13.9% — the three names you touch daily all sit just above 10%; Coca-Cola Consolidated (COKE) 40.0%, TSMC (TSM) 47.1%, Host Hotels (HST) 85.8% — the three that actually make the thing run from 40% up to 85.8%. The ends are eight times apart.
First, what this figure is not: it is not leverage, not a valuation multiple, and not a measure of profitability. It answers one question — how much of what this company owns is physical plant, equipment and land.
Now why it splits this way. A name, a formula or a system is a "name-type" asset that costs almost nothing to copy again: Marriott adding one more hotel to its system does not require putting up a building. A plant, a machine or a property is a "capacity-type" asset where every additional unit has to be paid for first: TSMC producing another wafer needs the line to exist. Different marginal costs of expansion produce differently shaped balance sheets.
The two roles also break differently. Earning on the name means fearing a devalued brand and franchisees walking away — its moat is that others want to hang its sign. Earning on capacity means fearing weaker demand and idle plants that still take depreciation — its moat is that others cannot build the same line quickly. Neither is better; they simply fail in different ways.
From here, three questions for getting to know a company, asked in order. First, how does it charge you: on the name, or on capacity. Second, read the fixed-asset line: around 10% of total assets usually means the name, above 40% usually means capacity. Third, ask how it breaks: the name side fears trust, the capacity side fears idleness.
Put the three together and the transferable test is this: to know a company, do not stop at what it sells — ask one more question: does it actually make the thing?
Getting to Know One Company: Host Hotels (HST)
Host Hotels deserves its own paragraph because it pushes asset-heavy to the extreme: it is the largest hotel owner in the US and does not operate a single one.
It owns the buildings of 76 hotels and about 41,700 rooms, but REIT rules bar it from running them, so all of them are handed to brand operators like Marriott and Hyatt. Its relationship with Marriott is therefore not competition but division of labour: one supplies the building, the other the name and the management, and both share what the same guests pay.
It earns on room rates and occupancy, not on franchise fees. That shapes its return curve: in a strong travel market its upside is far greater than the brand's, because most of a rate increase lands with the owner; in a weak one, the empty rooms are its own loss to carry. The cost is written on the balance sheet — refurbishment and construction come out of its own pocket, and property and equipment make up 85.8% of total assets, the highest of the six.
One more piece of proportion: 62.9% of its 2025 hotel revenue came from properties carrying the Marriott name. In other words, this "largest hotel owner" relies on another company's sign and management team for close to two-thirds of its revenue. Asset-light and asset-heavy are not opposing ends so much as two halves that usually have to coexist inside the same business.
Frequently Asked Questions
Q: Can September 9 be described as "semiconductors rallied"?
A: No. The Nasdaq semiconductor group averaged just +0.98% that day, and three of the six names fell — Nvidia (NVDA) down 0.91%, Broadcom (AVGO) down 1.13%, KLA (KLAC) down 3.21%, with 7.5 points between the ends. What got priced was not the label but each company's own orders and news.
Q: Is Marvell's $30 billion target company guidance?
A: No. It came from the chief executive in a television interview; the line-item figures only get disclosed at the investor day and in later reported results, and it has not entered any filing. It is a direction, not a verifiable set of accounts, and that qualifier has to survive any rewrite.
Q: What should you look at in a warrant?
A: Three numbers. How high the threshold is (Marvell's vests in full only after $120B of cumulative purchases, in 240 tranches of $500M), how long the term runs (struck at $206.58 for about 59 million shares, to January 2033), and how much dilution results (roughly 7% of the share count at full vesting). The third is the one most often skipped.
Q: Can Oracle's 93% and 57%–63% be compared directly?
A: No — different denominators. The 93% is last quarter's reported growth for the cloud infrastructure (IaaS) line alone; the 57%–63% is guidance for total cloud revenue, IaaS plus SaaS. Placing them side by side produces the false conclusion that growth is decelerating.
Q: How should the property-and-equipment share of total assets be read?
A: It answers only how much of what a company owns is plant, equipment and land — not leverage, not a valuation multiple, not profitability. Around 10% of total assets usually means it earns on the name (Marriott at 10.5%); above 40% usually means it earns on capacity (Host Hotels at 85.8%). Neither is superior; the shape of the risk simply differs.
Disclaimer: This article is compiled and written by the MEXC RealStocks team. The data in this article is based on the closing of the US stock market on September 9, 2026. The content is a compilation of public market information, and individual stocks are publicly discussed targets, which do not represent the recommendation or opinion of MEXC and do not constitute any investment advice. More US stock content: @MEXC | @Alpha_MEXC | @MEXC_Research