U.S. equities closed lower on Sept. 8, but AI compute leasing names moved the other way. Palantir named Nebius as its preferred sovereign AI infrastructure partner, yet CoreWeave (CRWV) ended the day up 11.72% at $99.83, outpacing Nebius (NBIS), which rose 7.73% despite being the company directly tied to the announcement. On the same AI compute chain, Micron (MU) and Nvidia (NVDA) both closed down.
After 20:00 UTC, AeroVironment (AVAV) is due to report earnings, followed by its earnings call at 20:30. The source article frames this release as the company’s first year-over-year comparison on a like-for-like basis since it completed the BlueHalo acquisition in May 2025 and passed the one-year consolidation mark.
The figures in the article are based on the Sept. 8, 2026 U.S. market close, and all times are in UTC. The piece was written by the MEXC Crypto Pulse research team.
Indexes fell across the board, led by the Dow
Sept. 8 was the first trading session after the Labor Day long weekend, and all three major U.S. indexes finished lower. The Dow Jones Industrial Average closed at 52,786.07, down 1.18%. The S&P 500 ended at 7,674.13, down 0.58%. The Nasdaq Composite closed at 26,423.69, down 0.31%, the mildest decline of the three.
The article highlighted three forces running through the session: oil prices nearing $100 a barrel, the 10-year U.S. Treasury yield climbing to around 4.8%, and an escalation in the U.S.-Canada tariff dispute. They were not the same event, but they pointed in the same direction, with cost pressure and discount rates both moving higher. In that setup, indexes with heavier weight in large components were hit harder, which the article uses to explain why the Dow led losses and the Nasdaq held up better.
Beneath the headline index moves, sector dispersion was sharper. Chip stocks showed strength in places. Intel rose 9.05% to $104.47 as the market digested tighter server-chip supply and the return of pricing power. Qualcomm gained 3.17% to $174.09 on cooperation tied to Amazon’s custom AI chips. Software names weakened the same day: ServiceNow fell 4.99% to $134.21, while Salesforce and Intuit each dropped about 4%. The article said investors were worried that newer general-purpose models could take direct aim at vertical software businesses.
GameStop closed down 1.41% at $18.89, then reported second-quarter results after the bell. Revenue came in at $790.2 million, and operating profit reached $160.2 million, a same-period high. The article noted that after-hours figures should not be mixed with the regular-session close.
No U.S. macro data was released that day. The week’s heavier macro schedule sits in the final two sessions: August PPI and weekly jobless claims are due Thursday at 12:30 UTC, and August CPI is due Friday at 12:30 UTC. The Federal Reserve’s September policy meeting is scheduled for next Tuesday through Wednesday. For that reason, the source article said the market’s immediate attention was on earnings rather than macro prints.
CoreWeave rallied even though the contract went elsewhere
CoreWeave is described in the article as an AI cloud compute company with a market value of $54.5 billion. On Sept. 8, the stock closed at $99.83, up 11.72% on the day. That move added about $5.7 billion in market value. Volume reached 50.88 million shares, or 1.79 times its average daily volume, and the stock sat at roughly 40% of its 52-week range.
The news setup was unusual. Palantir named Nebius as its preferred sovereign AI infrastructure partner, so the contract was not awarded to CoreWeave. Even so, the market treated the partnership as a demand signal for the broader compute-rental trade. The result was a 7.73% gain for Nebius and an even larger 11.72% rise for CoreWeave.
The article cited a five-factor scoring model to characterize the move. Peer relative strength and peer ranking both scored 100, the maximum. Industry valuation temperature was 49, trend position was 42, and volatility control was 0.
That zero, the piece said, was not an error. It reflected a beta of 7.41, meaning the stock moved about 7.4% for every 1% move in the broader market. By dividing the day’s 11.72% gain by 7.41, the article translated the move into roughly 1.6% on a market-adjusted basis. The point was simple: high-beta names naturally amplify gains and losses, so raw one-day moves are not directly comparable with lower-beta stocks.
The article also pointed to what was missing from the Palantir-Nebius announcement. No dollar amount was disclosed, and no committed capacity was disclosed either. Without those two figures, the partnership reads more like a qualification or endorsement than an order that can be quantified. That does not make it negative news, but it does affect how long the market can lean on the announcement for valuation support.
The AI compute chain split in two
Looking across six companies on the same AI compute chain, the split was clear. CoreWeave rose 11.72%, Nebius gained 7.73%, IREN climbed 5.04%, and Super Micro Computer (SMCI) added 1.69%. On the other side, Micron fell 1.61% and Nvidia dropped 2.01%.
According to the article, the first four names sit closer to rented compute capacity and server assembly, while the last two sell chips. Capacity-rental names rose; chip sellers fell. The article explicitly argued that this was not a sign of weaker chip demand. The day’s catalyst pointed to demand for compute leasing, not to fresh commitments for chip purchases, and the market repriced the segment closest to that message.
That is also why the article rejected the description that “AI stocks broadly moved higher” on Sept. 8. CoreWeave’s peer group was up only 0.26% on average. If the whole group had rallied together, the stock would not have outperformed it by more than 11 percentage points.
IREN was the supporting name in the trade, up 5.04%. Its business model involves converting mining data centers into facilities leased for AI training, so it traded with AI cloud themes rather than crypto prices that day. In the same group, Coinbase (COIN) fell 3.09% and MicroStrategy (MSTR) dropped 4.40%. The article used that contrast to show that sector classification and same-day pricing drivers are not always the same thing.
How the article measures “read-through” value
The piece defines a “read-through” as a case where news tied to one company is treated by the market as evidence of demand for an entire lane. CoreWeave’s move on Sept. 8 was presented as a textbook example: the news belonged to Nebius, but part of the repricing landed on CoreWeave.
To judge whether that read-through reflects real incremental demand or only emotional spillover, the article laid out three checks:
- First, whether the announcement includes dollars and capacity. A deal with contract value, committed capacity, and delivery timing can be used to estimate incremental demand. A partnership framed only as a “preferred partner” cannot. On Sept. 8, those details were not disclosed.
- Second, whether the second company’s trading volume expands. CoreWeave traded 50.88 million shares, equal to 1.79 times its average volume. Using that ratio, the article backs out average daily volume at roughly 28.42 million shares.
- Third, whether the stock outperforms its peer-group average. CoreWeave beat the group by 11.46 percentage points that day.
Two of the three boxes were checked, and one remained blank. That, in the article’s framing, is the full picture of the Sept. 8 read-through: fresh money clearly came in, but the partnership behind the trade still lacks quantifiable content. What matters next is whether later earnings reports show measurable usage.
AVAV heads into its first clean year-over-year comparison
AeroVironment is set to report after 20:00 UTC, with a call scheduled for 20:30. Chewy (CHWY) and Signet (SIG) also have earnings on the calendar before the open, but the article put AVAV at the center of the evening.
AVAV is described as a major U.S. supplier of small military drones. After completing the BlueHalo acquisition in May 2025, it added exposure to space, cyber, and directed energy. The article places the company on the consumable side of the defense chain: it sells lower-ticket equipment replenished in batches and tied to operational tempo, rather than giant platform programs booked once over a decade. That positioning, it argues, naturally produces more quarter-to-quarter revenue swings.
The article urged readers to start with absolute revenue rather than year-over-year growth. In fiscal 2026, the company posted quarterly revenue of $455 million, $473 million, $408 million, and $642 million. The sequence was not a straight line. Revenue dropped to $408 million in the third quarter, then surged to a record $642 million in the fourth, showing how uneven delivery timing can be from one quarter to the next.
The quarter being reported now ended on Aug. 1, and the comparable quarter last year was $455 million. The article says holding that line matters more than producing a headline growth rate.
For fiscal 2027, company guidance stands at $2.125 billion to $2.225 billion in annual revenue, which works out to about 10% growth. On the low end of the range, that implies 7.5%; on the high end, 12.5%.
Why 133% and 10% are not the same kind of growth figure
On the surface, the gap looks dramatic. The article lists fiscal 2026 quarterly year-over-year revenue growth at 140%, 151%, 143%, and 133%, while the midpoint of fiscal 2027 guidance implies only 10% growth.
Its explanation is straightforward: consolidation base effects. The BlueHalo acquisition closed on May 1, 2025, and only from that date did the acquired business enter the consolidated statements. That means each fiscal 2026 quarterly growth number included the new business in the numerator but not in the prior-year base. Triple-digit growth followed from that accounting setup rather than from pure organic expansion.
The article describes this as a standard post-acquisition base effect that lasts for exactly one year. The quarter ending Aug. 1 is the first comparison after the acquisition has been fully annualized, meaning both numerator and denominator now include the acquired business. In that sense, this is the first cleaner read on underlying growth, and the company’s own full-year guide places that closer to 10%.
For that reason, the article says 133% and 10% should not be lined up and read as a collapse in growth. The correct reading, in its wording, is that the base effect has ended, not that growth has disappeared.
It then narrowed the earnings watch to two lines. First, whether revenue can stay above $455 million, the prior-year level and the denominator for the first clean comparison. Second, whether two operating markers change: gross margin after reaching 32% last quarter, versus 21% and 22% in the two quarters before that, and management’s earlier indication that free cash flow would be negative this fiscal year while growth would lean toward the second half.
Rare earths: the choke points are refining and magnets, not ore
In the article’s market-education section, the focus turned to rare-earth magnets. It opened with three figures: China mines 60% of the world’s magnetic rare earths, but refines 91% of them and produces 94% of sintered magnets.
The article’s conclusion was that the scarce element in the chain is not ore itself, but the two steps that turn ore into magnets. Rare-earth elements are not especially rare in the earth’s crust and can be mined in many places. The harder part is separating 17 chemically similar elements one by one, then compressing the separated metals into sintered magnets that can be used in motors.
It placed five U.S.-linked companies along that chain:
- MP Materials (MP), with a market value of about $9.9 billion, sits upstream in mining. It is the only producing rare-earth mine in the U.S. The company produced 50,700 tons of rare-earth concentrate in 2025 and 2,599 tons of neodymium-praseodymium oxide, about double the prior year’s roughly 1,300 tons. It also produced its first batch of NdFeB magnets in Texas.
- Energy Fuels (UUUU), with a market value of about $3.7 billion, sits in midstream separation and refining. Uranium is still its main business, but its White Mesa mill in Utah separates rare earths while processing monazite and last year produced dysprosium oxide at 99.9% purity.
- USA Rare Earth (USAR), valued at about $2.3 billion, sits downstream in magnet manufacturing. It focuses only on the final step, pressing rare-earth metal into sintered NdFeB magnets. Its Oklahoma line started production in March, with a year-end target of 600 tons of annual capacity.
- Ramaco Resources (METC), with a market value of about $700 million, represents a future source of new supply. Its main business is metallurgical coal, and its Brook mine in Wyoming extracts rare earths from coal seams. Trial production is not expected until 2027, so this part of the business generates no revenue today.
- General Motors (GM), with a market value of about $77.6 billion, sits at the end of the chain. Electric-vehicle drive motors need magnets, and GM is described as one of the earliest long-term buyers tied to this domestic magnet buildout.
Range width says more about uncertainty than direction
The article also compared each company’s “52-week high divided by 52-week low.” The figures came out in clean order: GM at 1.69x, Energy Fuels at 2.61x, MP Materials at 2.65x, USA Rare Earth at 3.84x, and Ramaco Resources at 6.75x.
It was careful to say what this measure is not. It is not a 52-week return, not a valuation multiple, and not beta. It only measures how wide the price range has been over the past year. The wider the range, the more often and more aggressively the market has revised its pricing.
The ranking follows business visibility. GM’s 1.69x range reflects a company selling vehicles today, so investors can extend from revenue that already exists. Ramaco’s 6.75x range reflects a mine that will not enter trial production until 2027, leaving valuation exposed to shifting assumptions on timeline, costs, and policy.
From there, the article generalized a rule that it says can be applied to other sectors as well: the closer a company is to business already being delivered, the narrower the range; the closer it is to capacity that has not yet been built, the wider the range. That is not a judgment on which position is better. It is a statement about the kind of volatility investors should expect.
The article then suggested three questions for reading an industrial chain in sequence: whether a company sells ore or process capacity; whether its revenue is already happening today; and who ultimately has to buy the output. Its final framework was to begin with the least replaceable link in the chain, not automatically with the most upstream one.
MP Materials stands between price support and future capacity
MP Materials received a separate section because the article sees it as the only company on this chain trying to move from mine all the way to magnets. In 2025, it produced 50,700 tons of rare-earth concentrate and 2,599 tons of neodymium-praseodymium oxide, double the year before, and in the fourth quarter it produced its first magnets in Texas using commercial equipment.
The company also has an unusual setup: a floor price has already been contracted, but capacity still lies ahead. Under a 10-year agreement with the U.S. Department of Defense, neodymium-praseodymium oxide has a price floor of $110 per kilogram. In a volatile upstream commodity, that effectively locks in part of the downside for revenue.
At the same time, once the new magnet plant is built, total U.S. magnet capacity would be about 10,000 tons, but production is not expected to begin until 2028. The article argues that this mix explains why MP’s one-year price range is 2.65x: wider than GM’s, because future buildout still matters, but narrower than Ramaco’s, because part of the economics is already anchored.
It added one more scale comparison: USA Rare Earth’s target of 600 tons of annual capacity by year-end would amount to 6% of the roughly 10,000 tons of total U.S. magnet capacity expected after the new plant comes online. In other words, much of the capacity discussed in this space still belongs to the future tense.
Key takeaways from the article’s Q&A section
Why did CoreWeave rise more than Nebius without winning the contract?
Because the market treated the partnership as evidence of demand across the compute-rental trade rather than as a single-company order. The article says the three checks are whether the announcement disclosed dollars and capacity, which it did not; whether the second company’s volume expanded, which it did at 1.79x average volume; and whether the move beat the peer-group average, which it did by 11.46 percentage points.
Can Sept. 8 be described as a broad rally in AI stocks?
No. CoreWeave’s peer group was up only 0.26% on average, and on the same AI compute chain Micron fell 1.61% while Nvidia dropped 2.01%. The part that rose was compute leasing, not the entire AI complex.
How should beta 7.41 be used?
The article treats beta as a simple scaling tool: if the market moves 1%, this stock tends to move 7.41%. On that basis, CoreWeave’s 11.72% gain is roughly 1.6% after a market-style adjustment. Without that extra step, comparisons with lower-beta stocks can mislead.
Did AVAV’s growth collapse from 133% to about 10%?
The article’s answer is no. The 133% figure was fiscal 2026 fourth-quarter reported growth when the prior-year base did not yet include the acquired business. The roughly 10% figure is the midpoint of fiscal 2027 guidance, with both numerator and denominator including that business. The gap comes from consolidation base effects lasting one year.
How should investors read the 52-week high-to-low multiple?
As a measure of range width, not return, valuation, or beta. A larger figure means the market changed its pricing more often and by greater amounts over the year. In the article’s framework, companies closer to current deliveries tend to have narrower ranges, while those closer to unbuilt capacity tend to have wider ones.