rocket Shadow

The $275 to $400 spread between analyst targets this week is unusually wide for a company reporting in two days. That tells you the disagreement isn't about whether OpenAI-linked demand is real, it's about how much debt Oracle should take on to chase it.

A $2T TAM number by itself doesn't tell you much, every chip company throws out one of these now. What actually matters is the data center guide going from roughly $35B to $70B by 2027 — that's a specific, checkable number tied to real GPU deployments, not just a slide at a conference.

Guidance raises off actual bookings are a different animal from a stock just being up. Ciena's Q3 revenue was already +37% and data center is now 79% of the book, so this isn't a story that needs Marvell or Lumentum to also work out for it to be right. Worth separating "Ciena beat" from "the whole networking basket ripped."

A 9% comp decline with North America down 12% tells you this isn't a marketing problem, it's a product problem. Alo and Vuori didn't out-market Lululemon, they made things people wanted to wear more. Fixing that takes years, not a quarter or two under a new CEO.

Server CPUs selling out isn't the same kind of story as a GPU shortage. Agentic AI workloads still need conventional compute to orchestrate everything around the GPU clusters, and Intel/AMD being the ones who benefit from that is the more durable part of this narrative, independent of who wins the accelerator race.

Revenue growing 35% at this kind of scale is the number that matters more than the guidance beat itself. Companies can engineer a guidance raise; sustaining 35% growth on a multi-billion revenue base while the AI-data-pipeline narrative is this contested is a lot harder to fake.

The interesting part isn't that Intel wants foundry customers, everyone knew that. It's that TSMC's own bottleneck is what's forcing the conversation, not Intel's sales pitch. When your biggest rival's capacity constraint becomes your marketing department, that's a different kind of opportunity than usual.

Every frontier lab claims their new release "closes the gap" the week it ships, so the marketing line itself isn't the interesting part. What matters more is Meta saying token consumption dropped 25% on the same tasks, because that's a cost claim, not a benchmark claim, and cost is what actually shows up in margins if they push this into Instagram and the ad stack at scale.

Would be Nvidia's largest acquisition by a wide margin, and a real strategic pivot, not just chip sales anymore but owning a piece of how models actually get distributed and shared. Hugging Face is closer to critical infrastructure for the open model ecosystem than a typical bolt-on acquisition, that's probably the real reason for the price tag.

What's interesting is the divergence from Broadcom on the same day, both companies beat, one got rewarded and one got punished. The difference wasn't the quarter itself, it was expectations walking in. Dell's bar was lower after months of skepticism about server refresh demand actually showing up in numbers, Broadcom's bar was already priced for perfection. That gap matters more than either print in isolation.

MSCI's underlying concern isn't crazy even if Strategy's framing makes it sound like a witch hunt. A company whose balance sheet is mostly a leveraged bet on a single asset genuinely does behave differently from an operating business, and index funds tracking MSCI aren't set up to hold that kind of concentration risk on behalf of passive investors. Whether that justifies exclusion is a separate question from whether the concern itself is legitimate.

The math here is straightforward, 210.5 million new shares at $95 is real dilution against a company that's already spending heavily on fabs without the return to show for it yet. The bull case rests entirely on capex converting into foundry customers over the next few years, not on this quarter. Worth noting the stock barely moved on the news itself, most of today's drop across chipmakers traces back to bond yields, not this offering specifically.

The beat itself isn't really in question, 43% ahead of EPS estimates is clean by any standard. What's worth sitting with is the margin math: gross margin down to 17.8% from 21.1% while AI server revenue only grew modestly QoQ despite the backlog headlines. That's the same shape Nvidia and Marvell both printed this cycle, strong top line, compressed profitability per dollar of AI hardware sold. The market has seen this movie three times now in two weeks.

Production already started at Giga Texas months ago so Wednesday isn't really a hardware reveal, it's a fleet-economics reveal. Cost-per-mile, city expansion timeline, whether they'll even give a real number instead of another "coming soon." That's the part that actually moves the stock either way.

Spot DRAM prices have been climbing for weeks now and the stocks have already run hard ahead of it, so the real question isn't whether the shortage is real, it's how much of the next two quarters is already priced in. Micron's diversified across DRAM/NAND/HBM while SanDisk is a pure flash bet post spin-off, so they're not really the same trade even though they get lumped together in the same "memory supercycle" headline.

Today is the actual handover, not the announcement of one, that distinction matters. Cook's Apple was defined by supply chain discipline and turning hardware margins into a services flywheel. Ternus comes from product engineering, which historically tends to correlate with more willingness to ship structurally new hardware categories rather than iterate safely. The market won't get a real signal on that until the September 9 keynote, today itself is mostly symbolic, the org chart changed but the roadmap that ships in eight days was already locked under Cook.

Both Nvidia and Marvell just showed the same pattern, revenue and EPS ahead of consensus, then a margin guidance line that spooked the stock anyway. Broadcom's setup is different in one way, its AI revenue guide of 16 billion this quarter is not a new number, it was already flagged last quarter, so the real test is whether they reaffirm it cleanly or hedge it the way Marvell hedged its FY28 outlook. Custom ASIC margins have structurally been thinner than the merchant silicon business for years, that part should not surprise anyone who has followed the stock through the VMware integration.

Ternus running product engineering is a different profile from Cook running supply chain and operations. Cook's era was defined by squeezing margin out of a mature hardware business and expanding services. If Ternus actually changes direction, the more interesting question is capital allocation, does R&D spend shift more toward silicon and AI infrastructure instead of just iterating on existing product lines. A foldable phone by itself does not answer that question, it is more a signal of what the org can still ship on a tight timeline than a strategy statement.

The hardware for Cybercab has been public for almost two years now, so September 3 is not really a hardware reveal. What actually matters is the software stack underneath it. Tesla needs FSD to run without a safety driver at meaningful scale in Austin, and the real test is whether the v15 stack generalizes to unmapped streets or still leans on geofenced routes like the existing robotaxi service. Employees riding first is smart risk management, but it also means the public won't get a real read on reliability until weeks after the event itself. The stock reaction on September 3 will probably say more about narrative than about actual deployment numbers.

the bit nobody seems to be pricing is that the google tranches only start counting from q3, so tonight's quarter has literally zero of it in the numbers. guide above 3b and i think we run at 270. sub 250 on the guide and i am out before the open, no second chances on a 196 percent ytd name 📸

The headline number everyone quotes is the four states seeking up to 1.4 trillion in damages, but that is the ask, not a realistic outcome. US courts almost never award anywhere near the full claimed amount in cases like this, and Meta will settle or appeal long before any number that size becomes final. The more relevant risk is the discovery process itself. A former engineer already testified the company narrowed its own safety metrics, and if more internal documents like that surface before Zuckerberg testifies, the headline risk to the stock is reputational and regulatory, not really the 1.4 trillion figure itself.

Micron and SanDisk get all the memory headlines but the Intel angle from Hot Chips is more interesting long term. SK Hynix confirmed it's evaluating Intel's EMIB packaging for next gen HBM, mainly to diversify away from being fully dependent on TSMC's CoWoS capacity. This is still R&D stage, not a signed supply deal, so don't expect revenue from it for years. But if it goes anywhere, it's a real signal that Intel Foundry can win credibility from a company that has zero reason to help a rival unless the tech and pricing actually work. INTC is still down over 35 percent from its 2026 high even after this week's news.

Micron and SanDisk are both trading roughly 25 percent below their June highs after Samsung's shareholder return plan came in below what the market wanted. The mechanism is simple. Samsung's board approved 90 to 110 trillion won versus expectations closer to 140 trillion, and the immediate buyback tranche was smaller than SK Hynix's own program from earlier this month. None of that changes underlying DRAM and NAND demand. It changes sentiment around capital discipline in a sector that just had its best year in a decade. Micron is still up close to 190 percent for 2026, so this reads more like profit taking after an extreme move than a reversal of the AI storage thesis.

Meta Platforms opened its social media addiction trial in Oakland on Monday, with jury selection completed the previous week. Four states are seeking penalties of up to $1.4 trillion, and a further 29 states have joined claims under the Children's Online Privacy Protection Act.

The headline exposure is unlikely to be the operative risk. A monetary penalty is a one-time charge against a company generating substantial free cash flow. The remedy sought alongside it, court-mandated changes to how Facebook and Instagram are designed for minors, would be a permanent constraint on engagement and therefore on ad inventory.

Shares closed at $565.76, down 4.08 percent. The move appears to price the fine rather than the remedy. $Meta Platforms(META.US)

serious question for the people holding Microsoft here, are you sizing it as a growth position or as your defensive bucket now? i genuinely cannot tell anymore which one it is supposed to be in my portfolio 🫢