Sentiment score (0–100)
A temperature for the whole AI stock group, not a count of companies. 0 = people are scared and selling. 50 = a normal day. 100 = people will pay almost any price. Today's 68 is warm, not a heat wave.
Plain-English meanings for the jargon in the daily note. The analysis itself stays in the original voice. Last updated 2026-08-13.
The score
A temperature for the whole AI stock group, not a count of companies. 0 = people are scared and selling. 50 = a normal day. 100 = people will pay almost any price. Today's 68 is warm, not a heat wave.
Buyers are eager and already paying high prices. It is not a moral word. It means the good news is getting bid into the stock.
The opposite of greed. People are dumping first and asking questions later.
The actual hardware in a data center: lasers, cables, memory chips, hard drives, power, and the buildings that hold the computers. Not the chatbot. Not NVIDIA's software.
The hardware names are being bid hard. NVIDIA itself and the giant buyers (Google, Meta) are not joining the party the same way.
How prices talk
The live stream of stock prices. Old word from the paper ticker tape.
The current price has already paid up for the good news. Buyers have bid as if the bullish outcome is done.
Investors have already moved the stock for that piece of news. Buying now does not get you that news for free.
The theme may be real, but this particular stock has not thrown a party yet. Look at the lagging list in mean reversion, not the fast-movers list.
A fast, almost vertical rise in a short time, driven more by chasing than by a slow grind of new facts. This week's 20–40% five-day jumps in Super Micro, Nebius, CoreWeave.
The smaller stocks that orbit a big one. Around NVIDIA that is CoreWeave, Super Micro, the laser companies, and so on. NVIDIA is the core. They are the satellites.
Guessing what one company's news means for another. Memory stocks jumping today because Cisco complained that parts are getting expensive is a read-through, not Micron reporting earnings.
The company reported better numbers, and the stock still fell. Often means the good news was already in the price. Coherent this morning.
Results that match what people already expected. Not a blowout, not a miss. An in-line NVIDIA print can still knock down stocks that already ran 30% expecting a blowout.
Not "buy this, sell the market." It is "if you are in this theme, I would rather own A than B." Today's: NVIDIA/Broadcom/TSMC/Astera/Micron versus Super Micro/Nebius/CoreWeave/Everpure/Credo.
Stretch, snap-back, swing
How hard a stock usually swings versus the S&P 500. 1.0 moves with the S&P. 2.0 historically moved about twice as hard. The compact table uses Yahoo's 5-year monthly beta versus the S&P 500, not the Nasdaq.
The same idea, measured only over the last month of actual daily moves versus SPY (the S&P 500 fund). A recent snapshot. CoreWeave and Nebius have no long Yahoo beta yet, so this is what I use.
The actual bet you are taking by owning the thing that swings that hard. High beta is the size of the swing. Underwriting is the story you are on the hook for (power and construction for Iris Energy, funding a huge buildout for CoreWeave, not simply "more NVIDIA").
Stocks that jump and dump much harder than the market. Useful as a warning label. Not a buy signal.
When a price has run far from its recent average, it often snaps part of the way back. Not a law. A tendency, especially after a news spike.
The stock is well above its recent average (I use the 50-day average). Super Micro +36% versus that average is a short-term stretch. It can keep going. It is extended.
The average closing price over the last 50 or 200 trading days. A simple "where has this been lately" line. Far above the 50-day = short-term stretch. Far above the 200-day = a longer bull-market run.
This stock has not gone up with the rest of the AI group. That can mean "not yet in the price" or "the market is worried about this one for a reason." Read the sentence next to the name.
Names that just had a huge percentage move, usually on news everyone is already talking about, often with wild day-to-day swings. Easy to get hurt chasing them.
Too many people already on the same side of the trade. If the news is only okay, there is a line at the exit.
Where investors already sit: piled into one name, or still on the sidelines. The 5-day melt-up is positioning in the satellites.
A known date that can move the stock a lot. NVIDIA's August 26 earnings is event risk.
Investors pay a lower price for each dollar of sales or profit than they did before. Astera Labs grew fast and the stock still sat below its 50-day: the multiple de-rated into the print.
How far a stock falls from a recent high. A 40% drawdown means it gave back 40% of that peak.
How much of the recent good news is already in this stock's price. Built from distance versus the 50-day average (50%), the five-day move (30%), and the one-month move (20%). High means the stock already celebrated. Low means it has not thrown the same party. A setup score, not a buy signal.
Labels on the price-in score. Lagging is under 38 (not fully in the price). Mixed is 38–58. Stretched is above 58 (already paid). Read the sentence next to the name: Corning can lag because it is the wrong part of the stack, not because it is a bargain.
The AI stack
The group of public companies I cover: the giant platforms, the chipmakers, the laser and cable companies, storage, memory, and the firms that rent AI computers.
The giant cloud companies that buy most of the AI gear: Microsoft, Google, Amazon, Meta.
Newer companies whose main business is renting NVIDIA computers to other people. CoreWeave and Nebius.
The specialized chips that train and run AI models. GPU is NVIDIA's version. XPU is a catch-all for GPU plus rivals (AMD, custom chips from Google/Amazon/Broadcom).
High Bandwidth Memory. Super-fast memory stacked on or next to the AI chip. A bottleneck this cycle. Micron is the listed US name most tied to it.
Sending data with light instead of copper once the cables get short and the speeds get high. CPO/NPO are ways of putting lasers closer to the chip. Lumentum, Coherent, Credo live here. Semtech (optical DSPs) and MACOM (analog / photonic semiconductors) sit in the same chain; Fabrinet assembles the modules.
The shopping list of parts that go into a machine. A "physical BOM confirmation" means real parts are selling (lasers, networking gear), not just a chatbot headline.
The connections between chips and between servers. Astera Labs makes a lot of that traffic-cop silicon. Arista and Cisco make a lot of the switches.
Capital expenditure. Money spent on buildings, chips, and gear rather than day-to-day costs. CoreWeave guiding $35–39 billion of 2026 capex is a huge buildout plan.
Free cash flow. Cash left after running the business and spending on gear. The $500 billion financing plan exists because big tech's cash timing and GPU useful life do not match.
The subset of the AI stack that moves data with light and high-speed electrical links: lasers (Lumentum), optical engines (Coherent), systems (Ciena), fabric silicon (Astera Labs), SerDes (Credo), transceivers (Applied Optoelectronics), glass (Corning), substrates (AXT), assembly (Fabrinet), optical DSPs (Semtech), analog/photonic semis (MACOM).
The contract manufacturer that builds optical modules for the laser and engine companies. Fabrinet (FN) is the listed name: it assembles transceivers and engines for Lumentum, Coherent, and others. It is not the laser print itself; it is the factory layer.
Company reports and financing
The company reported numbers. A "hard print" is an official result, not a rumor.
What the company says it expects next quarter or next year.
Profit left after the cost of the parts, before other expenses. Cisco's margin shrank even as AI orders grew: they are selling more, keeping less per dollar.
One hundredth of a percent. 210 bps is 2.10 percentage points. Cisco's margin compressed 210 bps.
Memorandum of understanding. A handshake agreement, not a final contract. NVIDIA's $500 billion figure is MoUs, not money already sent.
Work already contracted but not yet delivered or billed. CoreWeave's ~$104 billion backlog is future contracted revenue, not this quarter's sales.
The seller helps the buyer pay, or promises to cover part of a loan. NVIDIA saying it may backstop ~25% of that financing is closer to helping customers borrow than to customers paying cash.
What a used GPU is worth in a few years. Lenders worry about this if they treat chips as collateral for long loans.
How long the money is tied up. A GPU may be useful for 2–3 years. A data-center loan may last 7–10. That mismatch is the financing story.
How expensive the stock is versus its sales or profits. A high multiple means you are paying a lot for each dollar of business.
Market shorthand
The index of 500 big US companies, and the fund (SPY) that tracks it. This is what beta is measured against in the table.
The tech-heavy index, and the fund that tracks it. Not what the Beta column uses.
Baskets of semiconductor stocks. A quick read on chip companies as a group.
Consumer and producer inflation reports from the US government. This week's numbers were calm, so the Fed was not the main story.
The US Federal Reserve. Hawkish means they sound like they will keep interest rates higher. That usually hurts expensive growth stocks.
The Fed's late-August meeting in Wyoming. Markets hang on the speech. This year it sits right after NVIDIA's report.
Specific to that company, not the whole market. Cisco falling on its own margin news while chip stocks rose is idiosyncratic.
Start from how the industry actually works (parts, power, loans, who writes the checks), not from the stock chart alone.