Good morning.
Brent tagged $100 for the first time since July — five Iranian tankers sunk off Kharg — and U.S. futures barely flinch. No panic. Yesterday's rotation-not-risk-off call survives the headline; the real verdict lands today with CPI and Bessent's first Treasury buyback print. TAIWAN AUGUST EXPORTS AT RECORD. PHISON REVENUE +377% YOY, PCIe SSD CONTROLLERS +77% MOM. TSM FOUNDRY SHARE 72.5%. GOOGL DUMPING $15.1B INTO FINLAND — the capex-pullback narrative keeps losing to the tape.
Memory is the connective tissue now. DRAM costs 4x YoY, iPhone Pro +10-20%, the $2,000 folding phone debuts tonight — the shortage just moved from B2B P&Ls onto the retail shelf. Watch Kioxia's CEO publicly tapping the brakes on NAND spikes: supply discipline reads cycle-extending, but it also says no blow-off coming. GLW's Verizon 80M fiber-mile deal is visibility, not EPS — optical demand was already strong. AMKR's $12B Arizona expansion checks the strategic boxes (TSMC anchor, NVIDIA prepay) but stays in the show-me bucket with that FCF-negative core.
Asia backdrop: Taiwan export print is the cleanest demand proof yet, and DeepSeek files for a STAR Market IPO via CITIC with ~$500M ARR — Chinese frontier AI is maturing with a real P&L. Bessent telling yen traders "I am the house now" keeps the carry unwind risk on the table.
Three themes. One: memory pass-through gets two verdicts this week — Apple's preorder weekend (9/12-13) on elasticity, the October Samsung/SK contract round on trajectory. Two: AI credit supply broadens fast — Amazon's £3B+ sterling bond, Zankore's $3.1B GPU-backed loan — circular financing in a new coat, no credit event priced until the first one prints. Three: Google's TPUv7 SparseCore claims 50% better perf per dollar than Blackwell Ultra on MoE inference, and word is Google may replace NVDA as TSM's largest customer — while OpenAI hitches on-device silicon to Samsung Foundry. A two-supplier foundry world is forming in 2027.
We'll hit up MU, GOOGL, and AAPL first, then get to the semis — TSM, QCOM, AMD, NVDA — and the storage chain.
Braze beat and raised, and the stock still got clocked ~12% in the aftermarket. That disconnect is the whole story.
The headline math is unambiguous: REVENUE $227.2M VS $220.2M CONSENSUS (+26% Y/Y, +24.4% ORGANIC) — nearly 3% ABOVE THE HIGH END of guidance. Op margin EXPANDED 630BPS TO 9.7% (vs 8.1% consensus, 5.0% last qtr). FCF HIT A RECORD $24.2M, almost 4x the prior year. Management RAISED FY27 revenue guidance to ~23% growth from ~21% — an implied 2H add of $7-8M, more than twice the size of the quarterly beat — and raised the margin guide too.
Then they guided Q3 op margin below consensus. Forge conference costs plus a fresh layer of sales capacity hired ahead of FY28. On a name already down ~8% in the week before the print on a deceleration narrative, that was all the confirmation the sellers needed.
THE STREET IS NOT RUNNING. Zero downgrades across five desks. Two PT raises — Canaccord $35→$37, Stephens $31→$34 — plus Needham ($50), Citizens ($35), and Mizuho ($32) all reiterating. The cluster sits at $34-37 (avg ~$37.60 across these five). The stock trades at $30.31 — BELOW THE LOWEST PRICE TARGET ON THE STREET. That's a positioning problem, not a fundamentals problem.
The headline beat-and-raise was known by lunch. What matters:
BULL: This is a beat-and-raise where the raise was MORE THAN TWICE THE BEAT. Management sees pipeline, and they're putting bodies behind it. Margins up 630bps to ~10%, FCF at a record, profitability this year. The $500K+ cohort NRR at 112% for two straight quarters is the vendor-consolidation thesis printing in the most important cohort. Decisioning Studio (+136%) and AWS co-sell give the model an AI kicker that isn't fully in numbers yet. At 3.7x forward revenue with durable 20% organic growth, the risk/reward tilts long — the Q3 guide noise is exactly the kind of entry that works a year out.
BEAR: The deceleration narrative has teeth. Organic growth from 26.7% to 24.4%; cRPO from ~26% to 24% — every leading indicator points lower, not flat. The Q3 margin guide below consensus means the beautiful operating leverage this quarter does NOT linear-extend, because management is choosing to reinvest into decel. TTM NRR flat. The stock is DOWN 7% YTD vs the S&P UP ~12% — the market has been voting on this story for months and the after-hours tape just validated it. At $30.31, you'd be catching a knife that needs a narrative reversal, and those rarely happen in one quarter.
"The second quarter was strong despite being caught up in a deceleration narrative. Revenue, subscription revenue, billings, and pro forma EBIT all beat Street expectations, and management raised both third quarter and fiscal year 2027 revenue and pro forma EBIT guidance." > — Stephens
"The firm views Braze as a long-term artificial intelligence winner and near-term proof point for AI return on investment, helped by a new Amazon Web Services co-sell motion." > — Stephens
"Needham expects the company to sustain 20% organic revenue growth through at least late fiscal year 2028." > — Needham
This is a cautionary tale for the whole app-software tape. A beat-and-raise with record FCF gets sold because the Q3 margin guide doesn't keep pace — that tells you the marginal buyer in this sector demands acceleration, not patience. Braze sits in the "AI ROI proof point" bucket with the martech/CPaaS cohort (think HubSpot-type platforms and adjacent customer engagement names), and that bucket is being held to a HIGHER bar than the AI infrastructure complex. If a 24% organic grower approaching profitability gets no credit for the guide, PMs should expect the same treatment for similar names into prints.
And the sleeper for the multi-manager crowd: management is ramping sales capacity "ahead of next year" into improving enterprise demand — competitive takeaways, upsell momentum, 112% NRR in the large cohort. That's a tell. The bear move would have been under-spending into a window. The over-spend into Q3 is the bull signal wearing bear clothes.
The after-hours tape disagrees with me today. It's been wrong before.
Three fresh notes across this event cycle. Three Overweight/Buy ratings. Zero bearish takes. Yet IONQ still trades BELOW the low end of the Street's $49-$100 target range — stock at $40.47. The three newest prints: $52 (Mizuho, CUT from $61), $70 (Cantor), $100 (B.Riley). Positive event, massive guide raise, and the sell-side can't agree on price by nearly 2x. That's the whole setup in one paragraph.
The event did its job. FY26 revenue guidance goes to a $455M midpoint — up $170M from prior, versus Street consensus at $288M. Pre-orders are open for the 256-qubit Superion. SkyWater closed July 31 and now gives IonQ manufacturing capacity nobody else in quantum has. The stock popped 11% Tuesday after the pre-market announcements — after a 27% Q3 drawdown. From the $84.64 high, SHARES ARE 52% OFF.
Mizuho is the note to read first. They raised FY27/FY28 estimates to $780M and $1.17B — roughly 2x the Street in FY27 and 50% above in FY28 — AND CUT the PT to $52 from $61. That's a de-rating wearing Outperform clothing. When numbers go up and the target goes down, the analyst is telling you the stock already front-ran the fundamentals. The story doesn't need to improve; the price does.
Bull: B.Riley frames the guide as a floor, not a ceiling:
"Revenue guidance could see an additional 5% to 7% upside by year-end 2026 from both organic IonQ and SkyWater revenues... annualized revenue could approach $1 billion in the next four quarters." > — B.Riley, reiterating Buy with a $100 PT
That sits on top of 371% LTM revenue growth and a fifth straight record quarter (Q2: $80M revenue vs $65M consensus, -$0.33 EPS vs -$0.54 expected). Then add the foundry math: cycle time cut ~80% (nine months to two months), 12x capacity, 330x cost-per-qubit reduction, ~400x qubit increase. Superion systems with 99.99% two-qubit fidelity ship early 2027. And FY26 carries just four months of SkyWater — full-year 2027 adds another mechanical gear of revenue on top of organic growth.
Bear: A $16B company guiding to $455M is ~35x forward sales — and a meaningful slice of that guide is acquired foundry revenue, not core quantum. Mizuho's cut is the honest version of this argument. Add the 1.96M-share resale prospectus (not new issuance, but an overhang is an overhang). With beta at 3.29, this name doesn't need bad news to drop 20% — sector air pockets do the job.
Management pitched a $215B TAM by 2040 — a moon shot, ignore it. What matters: vertical integration, a real manufacturing moat, and guidance that obliterated consensus. Mizuho's $52 PT is the valuation conscience. B.Riley's $100 is the bull case. Cantor's $70 is the referee. Stock at $40 sits below all three. Quantum remains a narrative business, and IONQ is the highest-quality narrative in the space. Size it like an option, not a compounder.
The Amazon deal is a confidence signal, not a number-changer. All three houses out on it this morning land at the same conclusion: management kept the FY27 $5B and FY29 $15B data center targets intact, so the headline "up to $60B over 10 years" is a ceiling on a relationship that was already in the model. Stock at $174 — up 27% in six months — and the PTs are still catching up to the tape.
The 8-K is the interesting part, not the press release. Qualcomm granted Amazon 25M warrants at $161.26, vesting in tranches tied to purchase commitments that could total $60B through 2036. That's a big paper number. But the structure says what the company won't say outright: the warrants only vest if Amazon actually buys. It's an option, not a contract.
Bernstein (Market Perform, $165) has the cleanest framing:
The deal should give Qualcomm high confidence to meet its numbers.
RBC was the most constructive — raised PT to $180 from $160, Sector Perform — and they nailed the nuance: the deal reduces risk around FY26/FY27 data center targets and looks mostly like a formalization of what was already announced at analyst day. But they flagged that Amazon runs multiple silicon suppliers (Marvell, Alchip, Astera Labs are all in the mix), so allocation share is unclear. They want to see volume ramps before doing anything more than a PT bump.
Barclays is the stubborn bear — reiterated Underweight at $180 — and their math is the most useful piece of the morning: the initial 3.75M shares vesting on first purchase commitments implies ~$9B of orders. Fine. But Barclays calls it back-end loaded. Their phrase: "capacity driving optionality in future years at hyperscaler ramps." Optionality is not revenue. And they're right that the numbers were already embedded from analyst day.
Bull case: The AWS design win gets QCOM into hyperscale AI datacenter — genuinely new territory. The accelerated Apple exit timeline removes the single most volatile relationship from the model, which is what Baird's $400 PT is celebrating. (That number is an outlier and a half — the other bulls are at $235.)
Bear case: Cantor says it best — crowded accelerator and CPU markets. QCOM is showing up late to a party Marvell and Broadcom are already hosting. And the warrant structure means Amazon only pays if products ship. Miss a ramp, and the $60B is a mirage.
Three firms, three holds (UW/MP/SP). Nobody is upgrading to Buy on this. That's the tell. The Street reads this as derisking, not inflection. The FQ3 print — $9.9B revenue beat, but $2.21 EPS missing at $2.23 — reinforces the margin story: top line grows, gross margin gets squeezed. Good morning for the narrative. Not yet a reason to chase.
StoneX walked out of the fireside with the same model and longer conviction — that's the whole ballgame for a name UP 243% IN TWELVE MONTHS. Reiterated Buy, stuck the $335 PT (call it ~50% upside from $223.14), only modest FY26-28 estimate bumps. Acree's framing is the honest version: the thesis didn't break, the probability it lands just went up.
The substance underneath: cleaner line-of-sight on demand quality, the 200MM-TO-300MM CAPACITY BRIDGE, and next-gen architecture timing. The recent record print ($2.68 EPS vs $2.21 est; revenue climbed 24% YoY to $460M) gives the fundamental side teeth. Street is consolidating on one narrative — contracted backlog, silicon photonics for AI data centers. Stifel and BofA both started coverage constructive; Benchmark pushed back on the China optical transceiver overhang.
"The fireside increased our thesis conviction without requiring a wholesale model reset. Management sharpened visibility into demand quality, the 200mm-to-300mm capacity bridge and next-generation architectures. We modestly lift FY26-FY28 revenue and EPS; the larger change is greater confidence in the path to FY28."
Watch item isn't execution — it's the multiple on that confidence. At 243% in a year, TSEM trades on derisking milestones (silicon photonics ramps, capacity transitions), not FY28 promises.
Mizuho hosted CEO Jay Chaudhry and walked away comfortable. Reiterates Outperform and $210 PT. Bull case is straightforward: Zero Trust Everywhere underappreciated, data/AI security greenfield, new logo activity improving into FY27. That's an inflection story, not a stable compounder story — but at this valuation, you don't need perfect execution.
The numbers give the patience cushion. Stock trades under 5x CY27 ARR and low-20s CY27 FCF, against 77% gross margins, ~25% revenue growth, and a Rule-of-~50 score. That's a decent r/r for a platform fighting a noisy competitive battle — and Mizuho acknowledges the noise, just doesn't think it's lethal.
"The CEO expressed enthusiasm for Zscaler's opportunities in data and AI security. The company's Zero Trust Everywhere opportunity remains underappreciated."
The post-print analyst cluster reads the same way — PTs bunched at $190-225: FBN $190, Stifel $200, JPM and Needham $215, Stephens $225. Q4 REVENUE $898M (+25% Y/Y, 2.4% BEAT), ARR $3.77B (+25%, ORGANIC HIGH-TEENS TO 20%). Demand isn't the debate. The debate is whether SASE competitive intensity caps the multiple — and at under 5x ARR, the market is already paying for a lot of that risk.
CANTOR IS BULLISH ANALOG — BUT TXN IS THEIR FOURTH FAVORITE. That's the whole story in one line. The firm reiterated Neutral with a $340 PT, and the logic isn't hard to follow: they see the analog group as a bottleneck with estimate upside and washed-out sentiment, but TXN has already run +52% YTD and now sits at 39x earnings. That's not "attractive valuation" territory — that's "priced for perfection" territory.
Cantor ranks covered analog names ADI > NXPI > MCHP > TXN > ON and sees ~57% average upside across the group to CY2028 earnings power. In context, TXN is the laggard of the bullish bunch. They'd be buyers of the group, expect incremental positive commentary over coming months, and clearly think the best risk/reward sits elsewhere.
The Q2 print supports the operational story — EPS $2.09 vs $1.92 est, revenue $5.46B vs $5.24B, +23% YoY, September guide of $5.90B above consensus at $5.61B. Mizuho went to $305 (Neutral), Rosenblatt to $350 (Buy), Benchmark to $360. But the bulls and the "yeah but" camp both see the same numbers — the divergence is what you pay for them.
"The analog semiconductor space is becoming a bottleneck, with upside to estimates paired with investor sentiment that has declined across the group. This creates opportunity at attractive valuations."
TXN is a good business with a rich tape. If you want analog beta, ADI/MCHP offer better entry points. TXN needs to grow into this multiple — the next few quarters of data center commentary are the swing factor. Neutral tinged with "not here, not at this price."
Three Buys and a $10 PT spread — that's the whole debate in miniature. Citi doubles down at $36 (~99% implied upside from $18.11). Needham sits at $30 on early AI adoption. TD Cowen just trimmed to $26 on margin pressure. The stock's down 43% YTD on a 26%-growing business with 74% gross margins. That's a margin-trust problem, not a demand problem.
Management's answer, per Citi's TMT conf meeting with co-founder/CSO Ed Hallen: pass SMS costs through and let multi-product adoption dilute the drag. The building blocks are there — Composer (AI agent optimizer) at ~138K users, 3+ product attach now >20% of ARR, $50K+ ARR customers at ~40% of total. Enterprise push runs through global system integrators plus a deeper Shopify partnership.
The Agency acquisition isn't a product story. It's a talent grab — Agency co-founder Elias Torres in as CPO. Don't model accretion from it.
The bull case: software that compounds at ~20%+ with this much AI attach for sub-3x revenue. The bear case: SMS pass-through wrecks near-term margins, and the softer H2 guide was the tell. Both are right — depends on your horizon.
Jefferies doubles down on MP at $55.34 — $85 PT, Buy. This is a 2027 story trading in 2026 clothes. The entire bull case rests on scale: Q3 NDPr PRODUCTION GUIDANCE >1,000 METRIC TONS and Jefferies' projected ramp to $300-400M magnet EBITDA from $40.68M trailing. That's a 10x step-function. Sales volumes flat QoQ on shipment timing and metallization lead times — that's logistics, not demand destruction. Customer engagement strong, government contracts shoring up cash flow visibility.
Q2 was the setup. Revenue $108.49M beat the $95.73M consensus on NdPr sales volumes +13.33%, but EPS printed -$0.01 vs +$0.01 expected — bleeding a little while scaling. $1.45B cash makes balance sheet risk a non-issue. Independence magnetics plant and 10X facility both progressing; that's where the multiple gets earned.
"The firm projects a path to $300 million to $400 million in magnet EBITDA for MP Materials."
Off-take announcements over the next few quarters are the tell. Recycling revenue and higher NdPr prices are the accelerants. Consensus at 1.28 (strong buy) means the Street already sees the arc — the stock needs contract signatures, not more PowerPoints, to close the gap to $85.
Bears would say the government is the customer of last resort and commercial demand remains unproven at scale. Fair. But at $55 with an $85 PT and capacity doubling into a structural rare earth supply crunch, the r/r skews favorably. Positioned like a call option on domestic supply chain urgency — just mind the volatility.
Canaccord cuts TTAN to $90 from $105 but HANGS ONTO BUY — the deceleration is real, but the multiple already reflects it. Stock sits at $81.58, DOWN 11.5% ON THE WEEK, and the market is punishing the GTV miss harder than the revenue beat.
Revenue +21% to $292.8M, beat by ~$8M — fine, but a tick below the usual $9-10M beat cadence. The real bogey: GTV +17%, ROUGHLY 200BPS BELOW RECENT NORMALIZED PACE. That's the forward-looking number, and it's rolling over. EPS -$0.26 vs -$0.25 expected — marginal miss, not the story.
HVAC is the weak spot. Industry lead and job growth softened, and the demand picture got muddier:
Customers didn't offset lower volume with higher ticket sizes during the period.
Seasonality did no favors either — the July 3 holiday behaved like a weekend (ate the business-day benefit), and an early cooling season pulled demand forward into Q1. So Q2 ate tomorrow's lunch AND had fewer working days. Not a great combo.
The bull case isn't dead, just deferred. Lead trends IMPROVED through the quarter and STABILIZED in July. That's the green shoot. But management stopped short of calling the all-clear — they said "stabilized," not "inflected." For a high-multiple software name, stabilization is a base, not a catalyst.
At $81.58, the stock has already repriced a lot of the bad news (down 11.5% in a week). Canaccord's $90 PT implies ~10% upside, and the Buy says the risk/reward works IF July stabilization holds through the fall print. We need evidence of sequential GTV reacceleration — not management confidence — before stepping in front of this one. Watch HVAC lead data as the tell for next quarter.
Cantor isn't blinking: Overweight reiterated, $125 PT, stock at $73.38. That's a 70% implied upside and a screaming gap between what the sell-side models and what the tape has already digested. The firm's fresh industry checks show analog cyclical trends strengthening — and the key nuance is that AI is the driver of that recovery, even in soft non-AI auto/industrial/consumer end markets. Cantor expects above-seasonal growth for the analog group regardless of the demand backdrop.
The FQ1 beat supports the bull case ($0.76 vs $0.70 est; rev $1.48B, +38% y/y), and 16 analysts have revised estimates upward with FY27 revenue forecast +36%. But UBS just trimmed its PT to $120 on inventory concerns — so the offsetting bear point is channel stuffing, not end-demand destruction. At $73 vs. a $125–$135 PT cluster, you're basically long the idea that one clean guide quarter collapses the gap.
"The analog semiconductor group is well-positioned for above-seasonal growth ahead regardless of broad-based end-demand conditions."
That's the whole debate in one line. Stock's already given back a lot from the $105.91 high — the easy downside's been taken. The r/r shifts hard if the next print shows inventory normalizing alongside that 36% FY27 growth.
Roth/MKM says Buy at $20 — PSQL's response? DOWN 34% ON THE WEEK at $8.04. That's a ~150% upside target on a $1.71B de-SPAC, which tells you this call is about the 2028 quantum payoff, not the next quarter. The bull case: neutral atom processors already deployed commercially, traction across materials/energy/finserv/logistics, and a tech base "built on Nobel Prize-winning research and France's sovereign national commitment to quantum innovation." Roth/MKM frames the real catalyst as hybrid classical-AI/quantum — the idea that massive cloud/AI infrastructure spend becomes the on-ramp for quantum workloads.
Everything else in the news flow is runway prep: new CFO, a commercial lead poached from HPE's HPC/AI division, a Quebec photonics packaging center, and a Credit Agricole CIB partnership with INITIAL PRODUCTION USE CASES TARGETED BY 2028. That date is the whole ballgame. At $8.04, you're buying a story stock with real technology, real sovereign backing, and a multi-year cash-burn runway — the tape is simply pricing in the wait. The initiation gives it legitimacy, not momentum.
"Pasqal has a strong base of technology built on Nobel Prize-winning research and France's sovereign national commitment to quantum innovation."
JMP/Citizens' Walravens stays Market Outperform with a $285 PT into Thursday's F1Q27 print — and he's not just throwing darts. The bull case is OCI inflecting to +111% (from +93% last quarter) with total revenue accelerating to +29%. He models $19B revenue and $7.1B OCI. That's the acceleration math that gets the story back on track.
The setup is clean: stock's down 17% YTD while the Russell 3000 is +12%. Sentiment is trash. That's how this name usually gets interesting. The +11.5% pop over the past week tells you PMs are starting to front-run the print.
His due-diligence checks improved — 83% positive vs 76% last quarter. Modest, but directionally right.
Street's converging on the same narrative: Mizuho sees an OCI-led beat, Guggenheim wants to see the $20B ATM equity raise actually deployed, and Oppenheimer leans on broader enterprise software demand as a tailwind. Morgan Stanley is the odd man out — $210 PT and Equalweight — though they acknowledge better GPUaaS gross margins. That's the one to watch if the growth story wobbles.
Options market implies a big move Thursday. The bears have had their window; a clean OCI acceleration and the weight comes off. PEG of 0.8 on this growth rate means valuation stops being the excuse.
THE TAKE: ADI is no longer just the analog cyclical — the market is paying up for the datacenter power/optics story, and TD Cowen (Buy, $460 PT) just spent an hour with ADI's Head of Datacenter & Energy plus sales lead to confirm it. Stock at $363.10, +35% YTD, and the Street has collectively moved PTs into the $425-465 range (Bernstein at $465, Needham at $450, BMO initiating at $430, Seaport upgrading to Buy at $425). The bull case is system-level selling — not cherry-picked components — across power, optics, and energy infrastructure.
THE NARRATIVE: The FQ3 print ($4.02B, +11% q/q, beat Stifel's $3.90B est.) was the catalyst that forced the re-rating, but the sustained story is AI infrastructure. Management is making the case that ADI's content opportunity is broader than the market gives credit for — datacenter power delivery and optical interconnect are becoming meaningful growth vectors, not just industrial recovery plays. That's why you're seeing 21 upward earnings revisions and a cluster of PT hikes on relatively thin incremental news.
The call itself didn't drop a new number — no "management raised guidance" headline here — but the optics of hosting datacenter leadership on an investor call tells you where the company wants the narrative focused.
"The call highlighted ADI's broad product portfolio and ability to drive system-level sales across datacenter power, optics, and energy infrastructure."
CAVEAT: Stock's up 47% over the past year and InvestingPro fair value flags it as slightly overvalued. The AI datacenter narrative is now largely in the price — the r/r from here depends on execution translating into order visibility, not just management presenting a good slide deck. Still, with the multiple expanding and analyst momentum clearly positive, fighting this tape is a losing game until we see a datacenter capex air pocket.
CLSA drew the line in the sand: PT to $710 from $575, Outperform, FY27-28 EPS up 25-29%. The whole call is MI-455 — MORE UNITS, HIGHER ASPs — with CLSA now 30% ABOVE CONSENSUS on 2027. They've got AMD printing $20.25 non-GAAP EPS in 2027, hitting the company's 3-5 year target on the early side.
The model underneath: $5M rack ASP vs $3M BOM. AMD invoices ~70% of the rack as revenue, which spits out late-40s to early-50s gross margins and ~$50K implied per-GPU pricing. CLSA also flags peak FLOPS claims now clear Vera Rubin and beat ASIC rivals on price. Caveat they own: NO INDEPENDENT TESTING YET. So you're trusting AMD's spec sheet in a market that's compute-starved enough to not care.
CLSA calls 2027 A YEAR OF RECKONING for AMD's GPU ambitions — and the rest of the Street is lining up behind that framing. Raymond James went Strong Buy, BMO started Outperform on the "full AI infrastructure" angle. Phillip is the outlier: Buy maintained, but FY26 forecast cut 8% on client/gaming weakness. That's the bear in one line — consumer drag persists even as the accelerators narrative runs hot.
$505.74, $826B cap, +224% over the last year. At 35x CLSA's CY27 number, you're paying up and then some — but in this tape, scarcity of compute trumps price. Risk/reward works as long as MI-455 ships and the spec sheet holds up under real testing.
Evercore ISI nudges PT to $14 from $13, holds In Line — and the stock trades at $14.14. PT below spot. That's the whole r/r story in one number.
The Sept 3 restructuring is the only real catalyst: 15% HEADCOUNT CUT, ~$150M annualized savings (Evercore models $125-175M). They bump FY26 EBITDA +3%, FY27 +14%, trim SBC forecasts ~10%. Restructuring charges land at $39-51M in Q3, partially offset by a $4-5M SBC reversal. Balance sheet is clean — more cash than debt.
But cost cuts don't fix the demand tape. Rosenblatt cuts PT to $12, Neutral, citing weak ad trends and a negative Q3 guide. UBS holds Buy at $16 but admits MACRO PRESSURES HIT 25% OF GROSS SPEND. KeyBanc stays Sector Weight post-announcement. The street collectively says: nice opex discipline, still no ad recovery.
72% DRAWdown in twelve months. One Buy, one Neutral, one Sector Weight, one In Line — nobody's swinging. The bull case hinges on whether the savings drop to the bottom line or get reinvested into Kokai shipping speed, which UBS flags as the real fix. That's a 2027 narrative.
Not much juice at $14.14 until the macro ad spend bogey turns. Cost cuts make the stock less bad, not good.
Pick your price target: $55-58 across four buys. Needham to $55 from $50 today, Piper Sandler leading at $58 from $47, Cantor and D.A. Davidson holding $55. All converge on the same driver — SaaS ARR ex-conversions +25% YoY to $598.1M, a $2.6M beat vs guide midpoint, and management raising FY26 ARR ex-conv guidance another $5M.
Thesis is a clean up-market AI-security tailwind. Enterprise and Strategic customers rethinking data security to deploy AI safely, Varonis catching that spend with 77% gross margins and multi-year sales productivity gains from a simplified comp plan. Needham thinks the company keeps outperforming its own 2026 guidance and flags Q3 upside from deals delayed out of the final week of June.
"Varonis is positioned to drive durable 20%-plus year-over-year growth in 2027 and beyond."
17 analysts revised EPS upward. Stock already +87% in six months, so the easy money is made. Add M&A speculation — reports of acquisition talks with Proofpoint (Thoma Bravo) — and you've got headline risk plus a potential takeout bid embedded in the tape. Net: strong fundamental story, increasingly crowded trade. Watch the Q3 deal-closure rate.
Lynx Equity downgraded AAPL into the iPhone 18 launch event — PT stays at $250 against a $316.22 tape (~21% downside). The call isn't about demand. It's supply: Apple FAILED TO LOCK UP ADEQUATE MEMORY AND FLASH FOR FY2027, and suppliers already allocated the global 2027 supply. New CEO needs a year to untangle that, per Lynx's checks.
The China bleed accelerates too. Huawei owns the high end; Oppo/Vivo already secured DRAM from Taiwan suppliers and flash from Kioxia. Apple didn't. Add Samsung pricing pressure on the low end (S26 FE at $700/$800) and high end (S26 Ultra at $1,200) and today's launch tape gets messy. Fold slips to 2027 as well — Samsung foldable display capacity runs <3M units, and most of that feeds Samsung's own Fold.
"The global supply of memory in 2027 has already been allocated."
Not all bears out there. BofA reiterates Buy at $380 (App Store Q4 FY26 revenue +YoY) and HSBC stays Buy at $366 — services momentum still carries the bull case. But 13 analysts have revised forward estimates lower and KeyBanc already occupies the Underweight $250 seat. One downgrade is one print. But when supply allocation becomes the binding constraint and a CEO transition adds friction, FY27 numbers become the next bogey. Not the last we hear of this.
Needham starts Buy with $8 PT, Oppenheimer already Outperform $9. Both banking on the same catalyst: Kodiak removes the safety observer by year-end, flipping the narrative from tech validation to business scalability. THE STOCK SITS AT $3.89 — so the street sees 2x+ here, but let's not get ahead of ourselves.
Consensus is Strong Buy with PTs $8.50–$13, which means Needham's $8 actually sits BELOW the range. They call it conservative — 5x FY2030 adj EBITDA discounted back, modest versus the industry leader. That's the disciplined version of the bull case. The aggressive version prices in an OEM partnership, which Needham flags as supported by recent industry developments and Kodiak's trucking partner relationships.
Needham views this milestone as a shift in the company's narrative from technology validation to business scalability.
The numbers tell a different tension. Q2 revenue $3.5M, +91% q/q — great trajectory off a tiny base. But full-year FCF outlook got NARROWED, which is management's polite way of saying the cash burn is real and they know it. This is a story stock with a fundamental clock ticking.
Oppenheimer's angle adds texture: Kodiak isn't just long-haul trucking. They frame it as multi-domain — industrial trucking and defense too — which gives the platform optionality beyond the core. For PMs, the r/r here is binary-ish: either the safety-removal milestone lands and the OEM partnership follows, or this stays a sub-$5 story fighting GAAP losses with no clear path to scale. Light coverage today, but two analyst initiations in the same window tells you the sell-side is starting to take the autonomous freight narrative seriously again.
AEHR — AEHR IS BACK IN HIGH-BETA SEMICAP BASKETS. That is a risk-appetite read, not a fundamental datapoint. Test/burn-in became a crowd trade again. If semis hold today, AEHR rides the flow; if CPI breaks risk, AEHR gets sold before the large caps even blink.
AKAM — CONSUMER AGENTS ARE AN AKAM EVENT, NOT A DATACENTER-CAPEX EVENT. Agentic web usage drives CDN hits, security hooks and observability. Market still prices AKAM as legacy delivery. Watch consumer-agent launches for usage hockey-sticks into the network — that is the repricing catalyst.
AMKR — DEMAND IS CUSTOMER-VALIDATED, NOT SPECULATIVE: TSMC signed a 10-year advanced packaging framework, NVIDIA PREPAID $1.5B, and Apple anchors the packaging/testing volume. Total Arizona campus goes to ~$12B — roughly $5B incremental capex on top of $7B — but Phase 2 construction starts late 2027, AFTER the Dec 31 2026 cutoff for the federal 35% investment credit. Without the credit, AMKR needs ~$3.25B more equity or debt; 1H26 operating cash flow less capex already runs NEGATIVE ~$307M, and Arizona startup costs plus depreciation hit 2027 op margin by 100–200bps. Q2 revenue hit ~$1.898B (+26%), 16.8% gross margin, $0.70 EPS; Q3 guides to $1.95–2.05B and 18.5–19.5% — Street wants proof the FCF bridge improves before 2028.
AMZN — ENERGY PROCUREMENT IS THE NEW AWS MOAT. Massive on-site gas projects bypass grid queues and push the AI capex cycle even longer. AMZN priced a multi-part sterling bond (at least £3B per the tape) as AI credit supply broadens from hyperscalers to neoclouds and labs. The credit stack runs ahead of underwriting; the first AI credit event reprices all this leverage for everyone.
ARES — PRIVATE SOFTWARE IS BIFURCATING: AI eats some assets, strengthens others. 2021–22 PE software vintages cannot hold standalone valuations under AI compression, so they consolidate. Ares sees distressed-debt and buyout flow. Treat European PE software restructurings as the honest real-economy read-through for AI disruption.
ASML — HIGH-NA EUV JUST GAINED A MEMORY DEMAND LEG. Intel extends the High-NA lead with mature solutions for 2028/29; TSMC and Samsung confirm production insertion by end of decade; Samsung and SK hynix target 2028 adoption. That adds memory to logic/foundry — three customers plus two more. ASML gates all lithography; this broadens the TAM, not just the order book.
AVGO — BACKLOG: LOCKED. PRICING POWER: LESS LOCKED. Broadcom signed late-2025/early-2026 LTAs across foundry, substrate and memory — capacity lands in 2028, the same year new merchant competitor capacity arrives. QCOM's ASIC push is now a credible second source, so the attack is on AVGO's margin, not its backlog.
COHR — THE NEXT OPTICS TAM LEG IS SCALE-IN: interconnect moves inside the tray to a memory-bandwidth tier roughly 10x Scale-Up per GPU. Coherent has named exposure alongside Lumentum across optical components and OCS. This is a 2027–28 architecture shift, but design wins start now. Market still thinks optics ends at Scale-Up.
CPNG — COUPANG'S MONTHLY CARD SPEND BROKE W5 TRILLION AFTER THE DATA LEAK. User-loss fears are dead; execution beat headline risk. A logistics and fulfillment moat absorbs regulatory and security shocks and keeps spending growing. Single-country datapoint, but a useful sentiment marker for Asian e-commerce.
CSCO — CSCO JUST QUANTIFIED THE AGENT DEMAND GAP: WE ARE ALREADY COMPUTE-CONSTRAINED WITH SUB-2% OF HUMANS USING AGENTS IN POWER-USER MODE. Enterprise agent traffic will not be kind to old capacity assumptions. Installed base plus campus and enterprise networking exposure makes CSCO a levered way to play agent adoption.
FIGR — BOFA UPGRADED FIGURE TO NEUTRAL, PT $49 from $31, AFTER 2Q VOLUME +132% Y/Y. Figure Connect drives 65% of volumes. Early proof humanoid commercialization works; the Neutral rating says valuation still does not discount execution. China tightens robot IPO scrutiny, so the US-listed leader benefits from a cleaner listing path.
FORM — FORM IS THE EARLIEST PRODUCTION-CLOCK BET ON CPO. SAM SUNG IS USING FORMFACTOR TEST TOOLS FOR CPO/PIC — the same tools TSMC uses — and plans to bring photonic testing in-house by year-end ahead of silicon photonics foundry services in 2027. That puts a dated revenue catalyst on FORM. Test capacity, not just transceivers, is the first CPO bottleneck.
FSLY — FASTLY IS AN AGENTIC-TRAFFIC STORY, NOT A CAPEX STORY. Consumer agents like Instinct and Astra drive edge delivery, security and observability. Investor day Sept 22 is the near-term catalyst; bull case requires management framing agent acceleration and moving toward the prior 30 handles. Watch usage data as the proof point.
GFS — MATURE-NODE TIGHTNESS RUNS THROUGH AT LEAST 2028; SOME CALL 2030. AI demand plus consumer silicon content created a capacity scramble, and GFS is one of the few pure-play mature-node foundries. Pricing firms, buyers accept volume-based deals. Capacity-constrained seller's market.
GLW — MULTI-YEAR VISIBILITY WITH MODEST HEADLINE MATH. Verizon's 80M fiber-mile deal stacks on Meta (up to $6B) and Amazon multiyear agreements; Q2 optical comm revenue grew 32% with segment net margin ~21.1% vs 15.8%. Each $1B of agreement value spread over six years adds only ~$0.015–0.019 annual EPS — pay for manufacturing scale and technology lead, not structurally higher fiber prices (Prysmian's >$1B US capacity doubling sees to that). Second option: multi-core fiber content inside the tray as Scale-In moves optics inside the box.
GOOGL — TPUV7 EXTERNALIZATION IS REAL AND THE FIRST CREDIBLE MERCHANT THREAT TO NVIDIA'S INFERENCE PRICING. Claims on InferenceX: up to 50% better perf/$ vs Blackwell Ultra, plus ~12% SparseCore throughput on MoE; Taiwan supply-chain chatter says Google could REPLACE NVIDIA AS TSMC'S LARGEST CUSTOMER. GOOGL committed ~€13B to Finland — largest single European investment — plus a 22-year Fortum nuclear deal for 50% of Loviisa output. But it just printed its FIRST NEGATIVE FCF QUARTER as a public company; debt/equity-funded AI flips the bull case from earnings power to balance-sheet capacity.
INTC — EMIB-T HAS ITS FIRST EXTERNAL CUSTOMER: GOOGLE'S NEXT-GEN TPU. Intel's silicon-bridge packaging with TSVs for power delivery just moved from internal story to merchant IP. UBS hears positive yield and execution feedback; MediaTek TPU v9 alone could need >2M substrates in 2028. INTC keeps the High-NA EUV timing lead with mature solutions for 2028/29, but the $20B equity raise and government-stake overhang cap the re-rate.
LITE — SCALE-IN IS ON THE CLOCK: LUMENTUM CEO SAYS 12–18 MONTHS. Interconnect moves inside the tray to connect memory and GPUs at ~10x the Scale-Up bandwidth tier. External memory pool at 20–40% of local HBM bandwidth means 35–70Tb/s per GPU — several times today's Scale-Up. LITE has light-source, optical engine and OCS exposure across all three. TAM expansion, not share grab.
META — MUSE IS A PRIVACY-WRAPPED DATA MOAT WEDGE. Each user gets a confidential cloud VM even Meta can't see, and personalization comes from Instagram. Early users already do real-life chores like TSA PreCheck with it. Meanwhile POST-WATERMELON IS TRAINING ON PROMETHEUS, META'S 1-GW CLUSTER — Watermelon itself ran ~10x the compute of the prior flagship — so AI expense grows even with consumer monetization unproven.
MRVL — MRVL IS THE POOLED-MEMORY INTERCONNECT PLAY. Scale-In external memory pools need 35–70Tb/s per GPU of connectivity; Marvell's optical interconnect and pooled-memory platform sit directly on that spec. New architecture TAM on top of custom silicon. The bear case: QCOM lands the next large custom compute award and shrinks MRVL's ASIC upside.
MSFT — MSFT KEEPS ACCESS TO ALL OPENAI IP WITH NO REVENUE SHARE THROUGH END OF DECADE. That is an underpriced strategic put if OpenAI remains the frontier leader. The ex-Azure decline and capex horror narrative ignores it. Bearish MSFT because of capex = bearish on AI itself.
MU — 2027 SHAPES UP AS THE MOST SUPPLY-CONSTRAINED YEAR IN SEMICONDUCTOR HISTORY. Broadcom and NVIDIA signed long LTAs in late 2025/early 2026; new capacity lands in 2028, not 2027, and even the 2028 wave doesn't fully clear the deficit. CXMT HBM yields sit at ~25% versus an ~80% golden yield; the TSV bottleneck (CXMT ~3,000 per die vs SK hynix >8,000 on HBM3) pushes credible China HBM supply years out. Spot bifurcates: DDR4 +1.78% to $45.36, NAND 512GB TLC -2.71% to $20.146 — AI/server DRAM tight, consumer NAND soft.
NBIS — NEBIUS REFUSED TO LOCK 2026-27 CONTRACTS, BETTING SURGE PRICES PRINT ABOVE CONTRACT RATES. That is a seller's-market signal from an Nvidia-backed neocloud. If demand stays hot, revenue beats; if it turns, no backlog cushion. Street currently sides with Nebius.
NET — NET IS THE LARGER-CAP PURE AGENTIC-INTERNET TRADE. Consumer agents push traffic, security and observability demand to Cloudflare's edge. It monetizes usage, not datacenter capex. Whatever FSLY frames at its Sept 22 investor day applies to NET with more liquidity attached.
NOW — BERNSTEIN GOES OUTPERFORM, PT $248, MODELLING AI REVENUE >$9B BY 2030: $7B workflow bundle uplift plus $2B+ consumption credits. Total revenue reaches $30–32B by 2030. Agentic workflow is the monetization vehicle, and it charges per outcome, not per seat. Positive read-through for any enterprise software that can charge for agents.
NVDA — NVDA IS LOCKED THROUGH 2027: >70% REVENUE GROWTH GUIDED NEXT YEAR, DATA CENTER +117%, WAFER AND HBM SUPPLY SECURED. The demand-cut rumor is dead. The squeeze now hits inference pricing — Google TPUv7 claims up to 50% better perf/$ than Blackwell Ultra — but NVDA owns the ecosystem lane; the price umbrella is just smaller. Supply chain as strategy: $1.5B AMKR prepay, $3.5B MediaTek investment, Groq stake, and Firmus at $10.5B+ with 900MW+ contracted.
SMCI — SMCI IS A CONTENT-PER-RACK STORY IN THE LIQUID-COOLING CYCLE. Rack power moves toward 600kW–1MW, and cold-plate plus quick-disconnect content per rack keeps rising. ODM status gives unit growth but commoditized margins — watch mix, not just revenue.
SNDK — SNDK IS THE AGENT-MEMORY STOCK, NOT THE HBM TRADE. NAND demand estimates keep climbing as agent KV-cache and context-length storage grow, and SNDK is the purest US-listed NAND. Spot NAND stays weak today, but the demand curve shifts when agent context storage becomes a capacity driver. HBF plus aiDaptiv/MEXT family attacks cold-DRAM economics and re-rates NAND into a near-memory tier.
TSM — TSM HIT RECORD 72.5% FOUNDRY SHARE IN Q2. Process leadership is settled; ENGINEERS ARE THE BOTTLENECK for 2027 supply. If talent caps advanced capacity, TSM keeps pricing and allocation power. Google could REPLACE NVIDIA AS TSMC'S LARGEST CUSTOMER — OpenAI, AMD, Apple and Amazon all route through the same foundry; it does not care which ASIC wins.
UBER — UBER TAPPED ITS FIRST-EVER EURO BOND — FIVE PARTS, PAYING UP INTO THE AI LIQUIDITY WAVE. That has little to do with UBER's P&L; it is a credit-cycle tell. Consumer platforms can borrow on AI spreads until the first AI credit event reprices all this leverage together.
UMC — UMC GETS MATURE-NODE SCARCITY WITHOUT THE HYPE. The scramble runs through at least 2028, possibly 2030; buyers sign volume-based deals and pricing power shifts to suppliers. Longer tightness converts capacity into structural pricing.
PANW — NO FRESH PANW DATAPOINT IN TODAY'S FEED. Security demand compounds as agents multiply, but the stock needs enterprise usage data, not another platform launch. No position change until that prints.
HUT — NO FRESH HUT DATAPOINT IN TODAY'S FEED. Pure BTC beta plus equity issuance overhang; neither resolves before the tape picks a direction. Sitting on hands.
ABNB — NO FRESH ABNB DATAPOINT IN TODAY'S FEED. Rates do the talking; the multiple waits on the long end. Core CPI print decides.
SHOP — NO FRESH SHOP DATAPOINT IN TODAY'S FEED. Rate-driven multiple plus consumer health via CPI. No position change.
TTWO — NO FRESH TTWO DATAPOINT IN TODAY'S FEED. Calendar-driven tape until the next release-cycle datapoint. Nothing to underwrite.
T — NO FRESH T DATAPOINT IN TODAY'S FEED. Only telecom datapoint this morning belongs to VZ/GLW fiber. T stays a dividend and rates story until management changes the narrative.
TMUS — NO FRESH TMUS DATAPOINT IN TODAY'S FEED. Fixed-wireless competition and rates dominate the group; no AI-network optionality priced without a management datapoint.
VZ — VZ IS THE DEMAND SIDE OF THE GLW FIBER DEAL: 80M FIBER-MILES COMMITS VERIZON TO A MULTI-YEAR NETWORK BUILD. That validates the AI/telecom optical cycle beyond hyperscaler capex. The stock still trades like a yield vehicle, not a growth build. If management extends capex at the next guide, the market may finally re-rate it.
CRCL — NO FRESH CRCL DATAPOINT IN TODAY'S FEED. Stablecoin narrative stays policy-driven; DC headlines matter more than volume prints until the regulatory calendar clears. No position change.
BTBT — NO FRESH BTBT DATAPOINT IN TODAY'S FEED. Same setup as HUT — BTC beta plus issuance overhang. No edge until new hash/capex data prints.
DOCS — NO FRESH DOCS DATAPOINT IN TODAY'S FEED. Specialty software names need usage data, not AI narrative, to re-rate. No print, no position change.
NXPI — NO FRESH NXPI DATAPOINT IN TODAY'S FEED. Mature-node tightness helps auto/industrial supply chains on pricing, but end-demand still decides. Wait for the October print cycle.