Good morning.
Futures flat but the tape tells a different story. SPX and Nasdaq flat on the surface, but underneath we have a violent rotation. MU +15-18% (now ~$1.4T market cap, passing META and TSLA). Meanwhile AAPL -6% on price hikes, MSFT hitting a new 52-week low (down 28% month-to-date). The "Memory Wars" thesis is in full effect — hyperscalers are the payers, memory suppliers are the winners. Biotech and housing made new highs.
Key earnings reactions: MU printed blockbuster gross margins of 85%, operating margins 81%, and signed 16 take-or-pay contracts with $100B committed minimum. Half of forward revenue is locked under 5-year SCAs. AAPL raised MacBook Air +$200, iPad Pro +$200, blaming "extraordinary surge in demand for memory." Tim Cook called it a "hundred-year flood."
Asia / macro: Durable goods orders -4.5% (beat -5.0% consensus), but PCE hot at 4.1% y/y, core 3.4%. Inflation is sticky, rates stay restrictive. Taiwan wafer makers (GlobalWafers, Wafer Works) signaling price increases — that feeds directly into the memory inflation narrative.
Three themes frame the day:
1. MEMORY INFLATION AS STRUCTURAL TAX. MU is trading price torque for duration with LTAs — but the cost passthrough to Apple, Microsoft, and hyperscalers is now undeniable. The binding constraint is DRAM wafer capacity, not compute. Rubin Ultra already cut HBM from 1TB to 768GB because of it. MU’s CFO says they’ll return 100% of excess cash — that’s a signal these margins are sticky.
2. ASIC VS GPU BIFURCATION ACCELERATES. OpenAI’s Jalapeño project with Broadcom confirms custom silicon will absorb high-volume inference. Merchant GPU share (NVDA, AMD) still has training and flexible workloads, but the mix shift toward ASICs for predictable inference factory workloads is real. QCOM bringing four data center chips to China adds another vector.
3. ENTERPRISE AI DIGESTION RISK. The viral anecdote of a company going from $30M to $500M in AI spend before the CEO slammed on brakes is not just noise. It implies a non-zero chance AI lab revenue flatlines in H2 as enterprises optimize. That’s a headwind for the hyperscalers paying the memory tax.
We'll hit up MU, AAPL, and ORCL first, then get to semicap and networking names.
THE VERDICT: This isn't a memory cycle — it's a contract cycle. The 16 strategic customer agreements (take-or-pay, $100B floor revenue, $22B upfront deposits) fundamentally re-architect Micron's risk profile. Street is scrambling to keep up: PTs have exploded from a $900-1,500 cluster to $1,100-2,000 in a single night. The bull case says margins are structurally higher, the cycle lasts through 2028, and 10x forward P/E is cheap. The bear case (Goldman, basically alone) says the stock has already priced in peak — but they're the only Neutral in a sea of Overweights. The rate of change here is the business model itself, not just pricing.
14 firms raised targets post-print. Range: $1,100 (Goldman, Neutral) to $2,000 (DA Davidson, Susquehanna, Positive/Buy). Median around $1,500-1,600. Consensus is overwhelmingly bullish — 15 analyst estimate revisions upward in the last 24 hours.
The collective thesis: May quarter revenue of $41.5B (consensus $35.7B), EPS $25.11 ($20.49 expected). August guide: $50B revenue, $31.00 EPS — 16% and 22% above consensus. Gross margin guided to 86% (up 110bps QoQ). DRAM pricing up low-60% QoQ, NAND up mid-80% QoQ.
But the numbers are almost secondary. The real story is the 16 supply/capacity agreements — five-year, take-or-pay, with floor prices that guarantee gross margins well above Micron's prior peak of 61.4%. Estimated floor GM is ~70% or higher. $100B in cumulative revenue at minimum pricing. $22B in customer deposits to fund $45B+ capex in fiscal 2027.
"These supply agreements significantly alter price discovery, with Micron likely booked into the third month of each quarter by at least 70% versus a historical 30%." > — Cantor Fitzgerald (PT $1,500, top pick)
The bull case is structural. Micron is no longer a cyclical commodity DRAM/NAND producer. It's a contracted infrastructure supplier with downside protection on 40-50% of revenue. Floor margins above 80% mean trough EPS is much higher than any previous cycle. DRAM and NAND demand (driven by AI inference, HBM, and server units growing high-teens) outstrips supply through at least CY28 — the cleanroom buildout takes 3-5 years from construction to wafer out. Bull PTs of $1,600-2,000 are based on 9-10x FY27/FY28 EPS of $173-200. At 10x peak earnings, that's cheap relative to secular growth.
Key data point: FCF of $110B in FY27 (Susquehanna estimate). That's not a memory company — that's a tollbooth.
Goldman Sachs is the lone holdout at Neutral with a $1,100 PT. Their argument:
"We would consider being more constructive if supply growth discipline continues across the industry through 2028 and beyond." > — Goldman Sachs (PT $1,100, Neutral)
The bear case is valuation and timing. Stock is up 726% in one year. At $1,048, it trades at 49x trailing P/E. Even if you buy the structural story, the risk/reward is "balanced" per Goldman. They worry that the capex ramp ($45B+) will eventually overshoot demand, especially as competitors (Samsung, SK Hynix) also add capacity. The floor margins are great, but if demand softens, ceiling prices could compress. Also: the $100B floor revenue is minimum — actual could be higher, but the ceiling is set at current market pricing. Any pricing weakness beyond the floor doesn't hurt GM, but it caps upside.
Other subtle bears: some firms (Baird, KeyBanc) note that the stock is above their Fair Value estimates. The rate of change in upgrades is staggering — PTs doubling from $550 to $1,540 (JPMorgan). That kind of parabolic revision often marks a peak in sentiment.
"The 16 agreements transform Micron's business model from cyclical commodity producer to multi-year contracted supplier with downside protection on revenue and margins." > — JPMorgan (PT $1,540, Overweight)
"Memory tightness is expected to last into 2028. HBM4 is ramping twice as fast versus HBM3e, with HBM4E driving blended HBM ASP up 70-100% YoY in CY27." > — Mizuho (PT $1,375, Outperform)
"These agreements would not be signed unless customers expected shortages to persist for several years." > — TD Cowen (PT $1,600, Buy)
"The lack of cleanroom space over the next several years is what leaves us most bullish on the sector. From construction start to wafer out is a three-to-five-year process." > — Wolfe Research (PT $1,500, Outperform)
This is a sector-wide signal, not just a Micron story. The memory cycle is structurally longer because AI inference absorbs massive wafer capacity that can't be replicated quickly. Read-through to:
The take: Qualcomm finally gave the Street a data center narrative worth underwriting. The investor day was a game-changer — $5B DC rev by FY27, $15B by FY29, EPS >$18. The stock ripped 12% after-hours, and the analyst community went into full re-rating mode. But the $15B target is still a leap of faith. The bull case is valuation (3.5x FY29 EV/Sales vs peers at 6-11x). The bear case is execution risk against entrenched incumbents. Consensus PT cluster is $220-265, with the high-end outlier at $300 (Benchmark) and the low at $190 (Susquehanna — stock already above that).
Six firms moved post-investor day. Here's the range:
Everything that matters is incremental to the prior narrative:
Bull case: QCOM is trading at 11x FY29 P/E vs peers at 20-30x. Even if they only hit $12-14 EPS, the multiple expansion alone gets you to $250+. The handset business is stable (5% CAGR Android). The data center TAM is $1T+. They have two hyperscaler ASIC programs already locked for FY27 — that's de-risked revenue. Meta signing on for server CPUs validates the architecture. The shift from Apple-modem overhang to edge-to-cloud AI infrastructure is a fundamental re-rating story.
"Qualcomm shifted investor focus from handset-cycle and Apple-modem risks to edge-to-cloud AI infrastructure opportunities." — Benchmark
Bear case: The $15B DC target is aspirational, not earned. The two hyperscaler ASIC programs are likely Tier 2 or internal-facing — not competing head-to-head with NVIDIA for training. The Oryon CPU doesn't ship until 2028 — by then, AMD/Intel/NVIDIA's Arm-based CPUs will have market share locked. The AI accelerator is behind Cerebras and NVIDIA's LPUs (ramping later this year). Microsoft's testimonial was "vague." And mobile handset headwinds are still there — Susquehanna flags it.
"Revenue strength seems credible despite remaining questions about product roadmap. We remain somewhat skeptical but do not forecast underperformance given overall strength in AI." — Morgan Stanley
"Qualcomm moved past Apple-related headwinds that previously affected its financial model and valuation. The stock trades at 3.5x fiscal 2029 EV/Sales versus peers at 6 to 11x, and 11x fiscal 2029 P/E versus peers at 20 to 30x." — Rosenblatt
"Qualcomm is now targeting more than $18 in EPS in fiscal 2029, higher than the $14 outlined at the previous analyst day. We modeled $7.5B in fiscal 2029 data center revenues." — Cantor Fitzgerald
"The company expects its data center business to reach approximately $5B in fiscal 2027 driven by the ramp of two custom silicon hyperscaler programs." — Susquehanna (neutral, but acknowledging de-risked near-term)
QCOM's pivot is a clear read-through for custom silicon and AI infrastructure. The hyperscalers are diversifying away from NVIDIA — QCOM winning two ASIC programs and a Meta CPU deal validates that trend. Positive for MRVL (custom ASIC, interconnect), AVGO (VMware + custom chips), and AMD (CPU competition keeps them honest). Negative signal for NVDA in the sense that there's more contender shelf space, but NVIDIA's lead in training won't be threaten by a 2028 CPU.
Also thematic for edge AI: QCOM's "data center-to-edge platform" narrative is the same one ARM and MRVL are running. The $1T edge TAM is real, but QCOM needs to show software execution — the Modular acquisition and Hugging Face deal address that.
Bottom line: QCOM is in the early stages of a multiple re-rating. The data center narrative is no longer theoretical. But the stock now trades at 21x trailing earnings — the bull case only works if you believe the FY29 numbers are achievable. PMs should watch for hyperscaler field-programmable proof points, not just press releases. The next catalyst is Q3 earnings and any updates on those two ASIC customers.
TRUIST LIFTS PT TO $225 (FROM $208), MAINTAINS BUY. The call is straightforward: DLR's $1.6B+ in announced transactions hit all the right levers — hyperscale, colocation, and the Strategic Private Capital platform. Truist sees 15% upside from here, anchored by a >10% Core FFO CAGR over the next several years.
The broader analyst crowd is in lockstep — Stifel at $235, Scotiabank at $222 — all citing the same structural tailwinds: power-constrained supply, inference demand, and a vacancy rate that's practically a rounding error. The only discordant note is William Blair trimming their data center index score to 75, flagging power constraints and local opposition. That's a risk, not a thesis-killer.
"Digital Realty is one of our favorite names in communications infrastructure, with a favorable supply and demand backdrop for data centers and significant tailwinds from Inference."
— Truist Securities, Jun 24 2026
CLSA tweaked PT down $5 to $600 (still 57% upside from $382) but that’s noise — the real signal is they see 30% potential upside to AVGO’s own $100B AI semi revenue guide for FY27. The bear case is MediaTek’s encroachment at Google TPU, but CLSA frames it as a 60/40 volume split in Broadcom’s favor, 70/30 in rev, and calls the renewed Google agreement a negation of the sidelining narrative. Longer-duration deals + secured financing for ASICs (Apollo/Blackstone XPV stuff mentioned by Wolfe) means the revenue visibility is actually getting better, not worse.
“The firm sees possible 30% upside to Broadcom’s current $100 billion AI semiconductor revenue guidance for fiscal year 2027.”
JPM, UBS, Wolfe all still bullish — this isn’t a downgrade, it’s a PT trim on optics while estimates get raised. Buy the dip.
Verdict: Wedbush is doubling down on the thesis while the stock sits at a 52-week low ($116.70, $230 PT). The catalyst this morning is the Zeta partnership — a marketing AI infrastructure play that’s expected to generate $100M+ in revenue over multiple years. That’s small relative to PLTR’s $4B+ run rate, but it’s a narrative accelerant: enterprise value creation via agentic AI connecting operational and customer intelligence. Wedbush’s core argument hasn’t changed — Street still doesn’t fully grasp PLTR’s capabilities. The 84% gross margins and 68% revenue growth support that conviction.
"The technology and value Palantir provides reinforces our view that many do not fully understand the company’s capabilities." — Wedbush
Notable that UBS (Buy, $200) and Wolfe (upgrade to Peerperform) also chimed in recently, but no new price targets. The setup here is crowded on the bear side after a 40% drawdown — r/r is asymmetric if the thesis holds, but the Zeta deal alone won’t move the needle on revenue. Watch for any broader AI spending commentary from enterprise clients.
Stifel cuts MSFT to $400 from $415, Hold maintained — and the call is all about margin erosion. They see FY27 gross margin compressing ~450bps YoY to ~63%, a full 350bps below consensus of 66.5%. The culprit is the revenue mix shift toward Azure (growing 3x the rest of the business) plus accelerating capex crushing Azure margins. Even if Azure gross margin stabilizes at Q4 FY26 levels (~47.5%), Stifel still sees 300bps of total GM compression. Operx leverage (headcount down, mid-to-single-digit OpEx growth) will only partially offset. Stock at $365.46, skimming the 52-week low of $356.28 — feels like the market is already pricing in some of this, but the magnitude of the miss vs consensus is nasty.
"We model ~100-150bps QoQ Azure gross margin compression in FY27 due to accelerating capex. Even if Azure GM remains flat at ~47.5%, FY27 gross margins would still decline roughly 300bps YoY."
Backlog looks bulletproof, but the cash flow story is ugly. Evercore stays Outperform / $245 (stock ~$165, down 10% on the week after the 10-K hit). The headline: cRPO printed $77B — that’s 86% of NTM revenue vs. ~68% last Q. Hard to argue with that revenue visibility. But the price of that growth is brutal: capex exploded to $55.7B from $21.2B, flipping FCF to NEGATIVE $23.7B for FY26. They’re funding it with ~$20B in equity-linked financing over the next couple quarters.
"The filing disclosed Oracle’s current remaining performance obligation reached $77 billion, representing 86% of Evercore ISI’s next-twelve-month revenue estimate, up from approximately 68% in the fourth quarter." — Evercore ISI
The offset: headcount down 21k YoY (that’s real — largely offsets the gross margin mix shift from OCI). And 1GW of capacity coming online in FQ1’27. Evercore thinks the implied 1H outlook is reasonable. Demand environment? Still supportive.
KeyBanc and Mizuho also reiterated Overweight / Outperform — both pointing to improving expense clarity and strong IaaS growth. The rest of the noise (Theator, Baystate, OPERA Cloud Assistant) is nice-to-have but not moving the needle for PMs. The real trade here: is the market pricing in enough FCF pain vs. backlog certainty? Not sure. Watch the equity-linked dilution math.
CBRS PRINTED A 92% Y/Y REVENUE BEAT ($191M) AND JUNE GUIDANCE OF $194M VS $181M CONSENSUS — that's a 7% beat on the forward look. Needham keeps Buy at $300 (12.5x 2028 EV/Rev on a 63% CAGR). OpenAI GPT-5.4 is live on Cerebras hardware, GPT-5.5 porting underway, and the AWS disaggregated inference deal (Trainium3 prefill / CS-3 decode) starts ramping in 2027. That's not priced into 2026 numbers.
"Core gross margin is expected to improve faster than previously anticipated due to increasing cloud prices and capacity expansion."
The bull case is clean: demand is the bottleneck, not wafer supply. Cloud pricing is improving, gross margin accelerates, and the partnerships stack (OpenAI, AWS). Rosenblatt and Mizuho both at $300 too — no dissenters on the Street. At $226, that's 32% upside if they execute. Risk? Capacity constraints cap near-term revenue linearity. But this is a narrative stock riding AI inference traction, and the narrative is working.
Piper Sandler is leaning into the risk/reward. Reiterated Overweight at $60 – stock at $48.75 – on the thesis that FCF estimates can keep grinding higher. They see a wild range ($31-$95), so the asymmetry is the pitch. Not ignoring the near-term drags: indirect hardware exposure, priority issues, renewal base lumpiness. But the call is that the market is too fixated on AI growth stories to notice the compounding cash flow machine here.
The broader analyst community is constructive. Post-earnings, firms like KeyBanc, RBC, Needham, and BofA all lifted PTs to $58-65. Consensus view: execution is solid despite supply chain headwinds, and the VMware migration tail + annual billing shift provide a multi-year base.
"Piper Sandler sees a favorable risk-reward profile for the stock with a range of $31 to $95 per share, noting that free cash flow estimates could continue to move higher."
That spread captures the debate. NTXN is a real cash flow story in a market that only wants to pay for AI narratives. The bull case works if you believe the conversion cycle is accelerating. The bear case says hardware exposure and renewal lumpiness keep the multiple compressed. For now, Piper picks the former.
GF Securities throws cold water on the AI server darling. Downgraded to Hold from Buy after the 266% rally. The thesis is straightforward: you’re paying 20x FY28 consensus EPS for a business where the easy AI re-rating is now fully discounted. Jeff Pu sees “limited upside amid elevated expectations.”
At more than 20 times consensus fiscal 2028 EPS, or a sum-of-the-parts valuation of 25 times AI business and 15 times core business, GF Securities views the risk-reward as uncompelling.
He’s not bearish on the backlog — AI revenue upward revision to >$70B and near-term GB300/HGX orders are real. But the setup gets tougher in 2027: expects Super Micro to take share in SpaceX’s next-gen deployment, and both SpaceX and CoreWeave are evaluating ODM-direct. That’s a structural margin headwind PMs need to track.
Valuation now 33.7x trailing P/E, PEG 0.35. EPS revisions still up (21 analysts raising the last 90 days), but the rate of change is slowing. Hard to get excited from here without a reset.
BACKBLAZE IS THE SLEEPER AI INFRA PLAY. Needham doubles PT to $14 (from $8.50) after the CoreWeave deal — $335M over 5+ years, their largest contract ever. Stock ripped 44% to $11.66, now sitting just off the 52-week high ($11.85). Market cap ~$700M, which starts to look cheap given this is a guaranteed revenue stream (warrant strike $7.60 for CoreWeave).
The contract has two parts: white-label B2 Cloud Storage and a managed storage service where Backblaze embeds software + staff in CoreWeave DCs. The managed piece is new — opens up a higher-touch, higher-margin delivery model. Consensus FY26 rev guidance is $162.5M at midpoint — this deal is NOT in guidance yet (auditors still chewing on revenue recognition). So there’s built-in upside.
"The agreement validates Backblaze’s strategic positioning in AI infrastructure." — William Blair (upgraded to Market Perform from Underperform)
Other firms also chiming in: Citizens to $14, Craig-Hallum to $16 (upgraded to Buy). Three analyst upgrades in one day is a strong signal. The bear case? The warrants cheapen the effective economics, and managed storage delivery is unproven at scale. But rate of change is clearly positive — this is a small company landing a massive, credible counterparty. r/r looks good here.
BUY INITIATION FROM LAKE STREET — Tyler Burmeister slaps a $58 PT (~30% upside) on this power semi pure-play, calling the March quarter THE BOTTOM for both rev and GM. The narrative: AI data center expansion is accelerating, balance sheet got a $150M jolt from the Chongqing JV monetization (funding R&D), and the stock trades at a "material discount" to power semi peers. That's the bull case.
The bear side? They just printed a -$0.28 EPS miss vs -$0.13 consensus, though revenue beat by 7% ($163.8M vs $153.5M). The new Intel Panther Lake/Wildcat Lake controller family is a nice product cycle hook, but near-term profitability still sucks.
"The March quarter marked the bottom for both revenue and gross margins at the company." — Lake Street
Risk/reward skewed favorable if you believe the turn is real, but this is a small-cap name with a single initiating analyst — not much cover to support a crowded trade. Watch for follow-through on the AI DC ramp and GM recovery before piling in.
Take: Benchmark kicks off coverage with a Buy and $100 PT (~30% upside). They see JFrog as a top SMID-cap infrastructure pick riding the AI code generation wave. The core thesis is structural: as AI pumps out more code, the control point shifts from source code to artifact governance. JFrog’s platform becomes the system of record for binaries, packages, and AI artifacts — making it the “DevGovOps” control plane for the enterprise.
“Accelerating code generation shifts the primary control point from source code to artifact governance.” — Benchmark’s Yi Fu Lee
The bull case stacks: 77% gross margins, 25% revenue growth, a TAM north of $20B, and consistent beat-and-raise execution. Other firms are on board too — UBS ($92), TD Cowen ($100), Cantor ($80) — all pointing to the same AI tailwind. The recent Claude Code plugin (with Anthropic) adds a near-term catalyst by extending governance to AI coding agents.
Not every SMID-cap infra story earns a 45+ Rule of 45 target, but JFrog has the margin structure and the narrative. Risk? Valuation is already pricing in a lot of the AI shift (stock up 76% in the past year). But the rate of change on the artifact governance theme is still accelerating. Worth a look if you’re underweight DevSecOps exposure.
MONNESS GOES LONG ON THE WRECK. Upgraded WDAY to Buy from Neutral with a $150 PT after a 62% drawdown from early-2024 highs. The firm's thesis: valuation is washed out, margins are sticky (76% gross), and FCF yield is a fat 10%. They also see early traction in agentic AI ramping — a nice narrative hook if the market starts caring again.
"Workday is the worst performing stock in its coverage universe in 2026, falling 45%."
Other shops (Cantor, Oppenheimer) are already sitting at $160-165 PTs, staying constructive. But don't ignore the overhang: AI fear has crushed the whole sector, and WDAY's own AI hiring tool faces a discrimination lawsuit that just survived a motion to dismiss. Ugly combo of macro sector headwind + idiosyncratic legal risk.
Bottom line: cheap enough to get a tactical upgrade from one coverer, but the tape still hates it. Watch for any catalyst that breaks the 45% YTD stupor — maybe a decent FQ2 print or AI product announcement. Until then, this is a value trap in a sector the market is actively punishing.
AXON -20% YTD is ugly against the Russell+7%, but the bull case is still alive. Citizens reiterated Market Outperform with a $700 PT (53% upside from ~$457) after hosting a panel with public safety brass. The chatter was all about field-level AI adoption, 911 dispatch integration, and drone-as-first-responder – not just hardware. Q1 beat (EPS $1.61 vs $1.60, rev $807M vs $778M) and Piper Sandler at $674 reinforce the thesis. The stock is pricing in execution risk, not product cycle. If AI deployment accelerates from here, the r/r gets interesting.
VERDICT: Hated stock ($14.57, near 52wk low, -44% in six months) but KBW says the AI selloff is overdone. Management met with analysts and pushed back hard on the narrative that AI in banking is a threat — they're seeing acceleration on the agentic banking platform. KBW reiterates Outperform, pointing to platform pricing transition driving revenue growth reacceleration (Q1 rev $159.4M, +11% yoy vs +6% prior quarter). Valuation is the hook: 12x FY28 EBITDA and 16x FCF, both inclusive of SBC.
"Management highlighted a disconnect between market concerns around AI disruption and the strong engagement and pipeline activity the company is experiencing with customers and prospects."
JPMorgan raised PT to $17 (from $16), calling AI defensibility concerns overstated. Stock trades at a discount to growth — if the platform transition sticks and AI monetization materializes, the r/r is asymmetric here. Risk is the 44% drawdown reflects structural churn, not just sentiment. KBW is betting on the rate of change. Not a PM favorite, but at these levels it’s worth a look for the deep-value TMT trade.
Barclays stays at Equalweight / $360, but the call is a footnote. Real story: they’re modeling Q2 deliveries at ~418K units, well above consensus 396K and their own prior 385K. Momentum in Europe, solid China, domestic weakness. Doesn’t matter. The stock trades at 343x P/E and is driven entirely by narrative — Robotaxi, Optimus, AI. Fundamentals are an afterthought.
"Automotive volumes and fundamentals have become secondary to investor focus on future technologies."
Levy notes the auto business needs to fund ~$25B+ capex in 2026. Bull case rests on inflection points that may or may not hit. Baird sees a potential Tesla/SpaceX merger in 12-18 months; UBS has Q2 deliveries at 405K. PTs range $123-$600 — tells you everything about conviction dispersion. Not sure we can read much into any single estimate. Narrative stock, narrative world.
Verdict: Intra-cycle price hikes are a tell. Apple finally flexed pricing power on Mac/iPad/Apple TV (17-54% increases) to offset memory cost spikes. The take? This is NOT a sign of pricing power — it’s forced margin defense. Evercore stays Outperform ($365 PT, 31% upside), but the unusual move (Apple almost never hikes mid-cycle) signals cost pressure was too big to absorb. iPhones spared, so the real test is September’s iPhone 18 Pro refresh.
“The intra-cycle price hike suggests the magnitude and pace of cost pressure exceeded Apple’s ability to absorb it.”
Riskiest bit: demand friction. Macs and iPads are more discretionary than iPhones. If consumers balk, this backfires. UBS neutral on China iPhone data (-19% YoY in May). BofA bullish on pricing power, sees $100 hike in estimates. Net: margins get a Band-Aid, unit volumes now the swing factor. AAPL at $279, 33.6x P/E — you’re paying for the brand, not the growth.
AI CPUs are the new GPU narrative. TD Cowen slaps a $475 PT (from $265) on ARM — a 33% upside from yesterday’s close and a market cap nearing $500B if it gets there. The thesis: agentic AI shifts work from GPU “thinking” to CPU “doing,” and ARM’s v9 architecture is the core beneficiary. The firm sees the FY31 AGI CPU target of $15B as “reasonable,” and thinks Street models are mechanically applying a 15% share to Nvidia’s $200B CPU TAM — too conservative.
“The Doing Behind The Thinking: As agentic AI shifts more work from the thinking GPUs do to the doing CPUs handle, CPUs are becoming an AI beneficiary.”
This isn’t isolated — UBS ($470), Bernstein ($500), Mizuho ($500) all raised targets in the same week, converging on the CPU-attach-rate theme. The stock is off 12.5% in the last week despite a 235% YTD run — some noise around China export control comments from CEO Haas, but the structural thesis is intact. Key variables to watch: GPU-to-CPU attach rates and pricing per core. If the market re-rates ARM on CPU dollar content vs. GPU, the $381B current cap starts looking cheap.
Rosenblatt starts CRWV at Buy, PT $250 — that’s 2.5x upside from here. They’re leaning into the scale story: 1GW active, 3.5GW contracted, $100B backlog. Revenue doubling this year to $12.8B, 77% CAGR to 2030. The pitch is that CRWV’s value per compute unit stays stable even as electricity and depreciation costs fall — so pricing power holds.
“CoreWeave’s value per unit of work has remained relatively constant despite decreases in electricity and depreciation costs.”
But $35B of debt and a 7.4x D/E ratio is the elephant in the room. Rosenblatt sees 3.2x EV/Rev on 2027 numbers, which is cheap if the growth trajectory holds. But that leverage profile means any demand hiccup or capex miscalculation gets magnified. The BattleBots partnership is cute, the MLPerf win is nice — neither moves the needle on the $100B backlog thesis.
Bottom line: pure AI-infra play with massive convexity, but you’re signing up for a balance sheet that could snap in a downturn. Not a position for nervous PMs.
The NHanced acquisition is the story — and it’s a big swing. Rosenblatt stays Buy / $22 (reiterate), seeing 2027E revs more than double to $65.5M if NHanced hits its $35M target. Cantor sits Neutral / $10, Lake Street Buy / $16. The deal math: $68.1M cash plus $5M stock, with up to $72M in earn-outs tied to revenue milestones. Max cash needed is $140.1M — only 10% of Q1 cash pile. Balance sheet remains clean (current ratio 66.67, zero debt). The kicker: Rosenblatt’s base-case 2027 rev of $30.5M jumps to $65.5M with NHanced. That’s from a $4.3M LTM base. Transformative if the integration works, but Cantor’s $10 PT implies the market isn’t pricing in the full optionality yet.
“Rosenblatt’s 2027 and 2028 revenue forecasts of $30.5 million and $59.5 million would increase to $65.5 million and $109.5 million respectively if the NHanced revenue targets are included.”
Bull case: Cash-rich, debt-free, buying revenue scale. NHanced’s 2.5D/3D packaging and chiplet integration fits the quantum/photonics narrative (Luminar deal already contributing). The escrow structure aligns incentives — sellers only get the full $20M if they deliver $35M in 2027 and $50M in 2028. That’s a steep hurdle, but if hit, QUBT goes from microcap to real revenue story.
Bear case: $22 PT implies >2x current price — Rosenblatt is already the bull. Cantor’s neutral $10 is a reality check: quantum hardware revenue is lumpy, NHanced is a services business with thin margins historically, and the $375M 2035 market share projection is a decade out. Execution risk on integration is real. The earn-out reverting to QUBT if targets are missed is a backstop, but it also signals the seller isn’t confident in the ramp.
Bottom line: High r/r, binary on NHanced hitting its numbers. For PMs with a 12-18 month horizon, the $22 target offers ~2x on a clean balance sheet. Just don’t confuse revenue potential with profits — yet.
AMZN — Raising GPU instance prices 20% July 1st. Inverse Moore’s Law — compute costs going up. AWS passing memory inflation to customers. Validates “AI inflation” thesis but volume risk if enterprises balk. Signed Memory SCAs locking in multi-year floor prices — converts variable cost to semi-fixed obligation, negative for margins. Internal silicon (Trainium/Inferentia) partially insulates but memory cost is indiscriminate; advantage vs MSFT/META narrower than people think.
META — No internal silicon = pure undifferentiated compute consumer. SCAs locking in ripping component costs. AI capex massive, revenue ROI questionable. Near-term margins squeezed by memory and power. “Autodata” paper suggests synthetic data could reduce compute needs long-term, but META is still scaling. No custom silicon program means stuck with merchant GPU pricing.
NVDA — Rubin Ultra 660kW racks force transition to 800V HVDC. Vera CPU in production, “built for agents.” But HBM cut from 1TB to 768GB due to DRAM shortage — flagship product memory-constrained. $800B buyback authorized — signals confidence but dwarfs FCF. Power density is next bottleneck; 800V HVDC rack targeting 3Q26 readiness. OpenAI Jalapeño threat bifurcated: ASICs for inference, NVDA retains monopoly on frontier training and software-velocity deployments.
ON — Acquiring Synaptics ~$7B all-stock (1.35 exchange ratio). Pushes into Edge AI/Physical AI, adding $30B to TAM. Power electronics plus intelligent sensing. Core automotive soft but physical AI story sustains premium.
SYNA — Being acquired by Onsemi ~$7B all-stock, ~19% premium. Shareholders get stock in combined entity with broader AI reach. All-stock structure may trigger opposition if cash deal preferred.
AMAT — If memory growth is bit volume (not just pricing), semicap is the biggest beneficiary. Multi-year capacity expansions — Micron FY27 capex >$40B. Tool orders guaranteed. 2-3 year backlog visibility. 4F2 transition and EUV dose issues extend capex cycle.
ASML — EUV dose increasing per node — more tools per wafer. Monopoly pricing power. High NA EUV needed for sub-2nm. Dutch government lobbying US to limit export controls — protects pricing elsewhere. Memory capex cycle validates continued tool intensity.
LRCX — Benefits from memory capex in etch/deposition. More etch steps per wafer as 3D NAND layers increase and DRAM moves to 4F2. Bit volume vector positive. EUV-lite could reduce etch steps but not near-term.
KLAC — Process control leader. Yields critical in HBM and advanced packaging (CoPoS, glass substrates). Defect inspection demand rises with capacity. Revenue benefits from capacity additions.
ONTO — Smaller semicap with beta to memory and advanced packaging. HBM4 and agentic CPU DRAM need specialized process control. Mentioned as big winner if growth vector is bits not pricing. Higher volatility, higher beta.
FCEL — 380MW clean on-site power deal for data centers validates behind-the-meter generation thesis. Grid interconnection queues 3-7 years; fuel cells bypass them. Direct play on data center power bottlenecks. Behind-the-meter DC capacity projected to reach 40GW by 2028.
COHR — Strong 6-inch InP capacity. CPO lasers and 200G EML generate 80-90% gross margins. Supply chain checks confirm strength. Optical margin story bifurcated, but COHR on the right side.
MRVL — Targeting 20%+ custom ASIC revenue ($15-20B) but viewed as optimistic. CFO sold $632K of shares. Broadcom stronger with OpenAI. Hyperscaler internalization trend is nightmare if they lose capacity leverage.
TSM — Allocated next 2 years capacity including memory and packaging. CoWoS still constrained. Ultimate bottleneck for advanced silicon. Pricing power immense. Advanced packaging elongated — Feynman delayed, CoPoS/glass substrates maturing. Margin mix improving as TSMC sells more packaged solutions.
INTC — Using advanced packaging (EMIB, Foveros, hybrid bonding) as side-door attack on TSMC. HBM4 era memory base dies become battlefield. Foundry pivot still unprofitable but packaging differentiation buys time.
IBM — World’s first sub-1nm chip (0.7nm) using nanostack vertical architecture — research demo, not product. 50% more performance or 70% better energy efficiency vs 2nm. No near-term impact but validates roadmap extension by 10 years.
PANW — Part of cybersecurity rotation rallying with XBI, JETS. No specific AI catalyst but benefits from increasing AI attack surface. High-quality name in strong demand.
CRWD — $1,000 contribution per child of employees for opening Trump Accounts by Jan 1, 2027 — controversial but immaterial to earnings. Elsewhere benefitting from cybersecurity demand and rotation.
FTNT — Rotating with cybersecurity group. Benefits from enterprise network security and AI data center buildout. No specific new signals. Momentum favorable.
OKTA — Identity/access management crucial as AI agents proliferate. Should benefit from increased need for secure authentication/authorization. No specific catalyst.
WEN — Meme stock with 35% short interest. WSB-style squeeze. Not fundamental. High short interest but no earnings change.
MSTR — Down 78% total return in 12 months. Bitcoin position down $14B. Facing $10B option expiry. Bitcoin ETF flow-weighted average loss 36%. Crypto outflows since May — risk off in speculative assets.
SNDK — Citi raised target. Pure NAND play. Micron NAND ASPs up mid-80% Q/Q. YMTC at 13% market share — not collapsing pricing. Beneficiary of memory supercycle.
STX — Seagate up 8-10% on Micron sympathy. HDD relevant for bulk storage in AI data lakes. Memory shortage may accelerate HDD adoption for cold storage. Direct beneficiary of DC buildout.
WDC — Western Digital (SanDisk) effectively same as SNDK. Storage benefits from memory pricing and DC demand. Structural NAND shortage helps pricing.
HPE — Hardware ecosystem challenged per SCA framing debate. May face margin pressure from memory costs and demand uncertainty. Pass-through difficult in enterprise deals with long lead times.
HPQ — HP faces memory cost pressure in PCs/printers. Consumer price hikes (MacBook, iPad, Xbox) will eventually translate to PC price increases. Margin trajectory challenged.
LOGI — Logitech peripherals may see price increases or demand sensitivity. General hardware headwind applies. No specific data.
GLW — Corning GlassBridge glass-based fiber-to-PIC connector for CPO/NPO. GlassWorks platform with GlobalFoundries. Benefiting from DC connectivity and packaging innovation (glass substrates for advanced packaging).
GFS — Partnering with Corning on GlassWorks AI platform. Specialty nodes and photonics. Not leading node player but could drive growth.
STM — Power semiconductor price hikes announced for 2026. Demand from AI data centers and automotive. Benefits from 800V HVDC high-density racks.
IFNNY — Infineon also raising power semi prices. Dominant in IGBTs and SiC. DC power density expansion (660kW rack) drives demand for high-efficiency modules.
TXN — Texas Instruments raising power semi prices in 2026. Broad industrial/auto exposure. AI DC boom incremental demand for power management. Pricing firming.
KEYS — Completed acquisition of VPIphotonics for unified photonic design automation. Provides test equipment for DC components (optical, electrical). Indirect beneficiary of DC upgrade cycle.