Substack 研究号「TMT Breakout」发布高盛/花旗会议第三日 EOD 综述，混杂 ORCL/ADBE 盘后财报、微软 2032 年数据中心 38GW 扩容传闻及约 30 只个股会议点评，无单一可交易事件

Good afternoon. **QQQs -1%** led lower by **semis -2.6%** as negative headlines continue to come out of the middle east driving a **+7% move in Crude**. **Yields spiked 7-14bps** across the curve breaking out with 10year almost touching 5% and 30year almost at 5.4%, both near levels seen back in 2007. Fed expects shifted in a hawkish direction now pricing in 45bps worth of hikes this year and odds of a rate increase next week is now 75%. CPI comes tomorrow morning.

Day 3 at Citi and GS was busy. We’ll cover snippets below, among a few other items.

*Post-close:*

### **ORCL +2.5% solid print with IaaS accelerating and the $20B ATM overhang removed**

IaaS growth of 121% Y/Y, or 120% cc, cleared the 116–118% buyside bar , with 850MW of capacity delivered showing actual backlog conversion.. Sw was soft, but stronger operating margins and substantially less cash burn made this a better-quality quarter than the headline revenue upside alone suggests. Biggest relief for investors: ATM is finished, the $90B revenue outlook is intact, and more than $30B of new AI contracts require no incremental change to capital-raising plans. Stock was up +7% but fading a bit on the call…

**ORCL Key KPIs vs. Street:** Revenue $19.35B vs. $19.14B; IaaS +121% Y/Y (+120% cc) vs. ~117% Street and 116–118% buyside bogey; Cloud rev $11.6B, +62% Y/Y vs. ~$11.5B; op margin 42.1% vs. 40.7%; EPS $1.92 vs. $1.74; FCF ~-$5B vs. ~-$10B. RPO reached $664B, +$26B Q/Q vs. Street modeling only ~$2B of net adds.

**ORCL Guide vs. Street:** FY27 rev at least $90B vs. $89.6B, with EPS $8.10 vs. $8.07. Q2 total rev growth 30–34% cc vs. 31.9%, Cloud growth 64–70% cc vs. 65.2%, and EPS $1.85–1.93 vs. $1.90.

### **ADBE -2.5% small headline upside, but still no convincing growth inflection. Unlikely to change bull vs bear debate.**

Revenue and EPS came in modestly ahead, but ARR growth slowed to 11.2% from 12.5% last quarter, while 44% operating margins left little to get excited about on earnings leverage. AI-first ARR growth above 150% and Firefly app/credit-pack ARR up 40% Q/Q solid, but the 10.2% full-year ARR growth target still didn’t signal a broader reacceleration. FY rev guide actually moved slightly higher, rather than being cut, although the Q4 outlook offers little evidence of an imminent inflection.

**ADBE Key KPIs vs. Street:** Revenue $6.76B vs. $6.70B; EPS $6.13 vs. $6.09; ending ARR $27.5B, +11.2% Y/Y; op margin 44.0%, modestly softer than expected. AI-first ARR grew >150% Y/Y, while Firefly app/credit-pack ARR grew ~40% Q/Q.

**ADBE Guide:** Q4 rev $6.80–6.85B with EPS $6.30–6.35 and ~44% op margin. FY26 rev nudged to $26.576–26.626B and EPS to $24.45–24.50

Let’s get to the good stuff…


# **AI/SEMIS**

**Headline after the close: **MICROSOFT PLANS TO TRIPLE DATA CENTER CAPACITY TO 38 GIGAWATTS BY 2032, UP FROM 12 GW NOW.


-

**SWKS +10%: **Hey a non-AI semi leads the way on positive read-through from AAPL Duo launch…


-

**NVDA -2.4%** as CEO Jensen Huang at Goldman opened with ‘I was right’ on the $3-4tn call, pitched NVIDIA as ‘the world’s first and only growth value stock,’ and reaffirmed 70% growth next year. New: per-system content of ~$18K in Hopper, $25K in Blackwell and ~$40K in Vera Rubin; GB NVLink72 rack shipments up 27% month over month; the Australia 2GW for 2027 sized at $80bn; per-gigawatt math of ~$60bn of cost against ~$50bn a year of rental revenue; and compute turned into an asset-backed, loan-securing asset.

-

**LRCX -5.6%** as CFO Doug Bettinger at Goldman, a day after Citi, again promised no new statements and called the industry ‘fundamentally sold out,’ with leadership ‘more concerned about missing the upside’ than a downturn and 2027 WFE still DRAM-led with foundry closing the gap. New: Lam is ‘deeply engaged already’ with SpaceX’s Terafab, sized as a ‘very high’ opportunity; a $3bn US lab and engineering investment including a new Oregon lab; China WFE ‘slightly up’ this year but shrinking as a share; a company-wide no-M&A stance beyond tuck-ins.

-

**DELL -5.3%** as CFO David Kennedy at Citi called the FY27 guide ‘realistic’ where 90 days ago it was ‘prudent,’ on ~$61bn of Q2 AI server orders and a $95bn non-cancelable backlog. New: customers now negotiate supply access for years two through four (’guarantee supply does not guarantee price’), next year looks tighter than this one with DRAM and NAND the big constraint, OpEx runs 8% of revenue versus 20% six years ago, storage adds $2.5bn this year, and the buyback stepped up from $1.6bn to $3.9bn.

-

**SPCX +40bps** as CFO Bret Johnsen at Goldman said SpaceX has ‘even more conviction’ in the $100bn ARR target for year-end after closing another hosting deal this month worth about $1.11bn a month from Dec 1, with 5-10GW deployed next year versus a little over 2GW at year-end. Hosting deals are 90-day commits with 90-day outs because the company expects its own Grok products to monetize above the $30-50 per-watt range; the first orbital compute satellites fly next year; direct-to-device 5G service turns on in 1H28.

-

**ANET -2.0%** as CFO Chantelle Breithaupt at Citi kept the $12.6bn, 40% guide and named the four signals to judge the model on (P&L growth, the deferred trend, purchase commitments, RPO), with commitments tripled to $9.6bn and already reaching into chips for 2H27. Scale-up carries ‘zero in our guidance and zero revenue’ for 2026 with revenue in 2028; XPO early deployments come 2H27 with CPO about 12 months behind; a third and possibly fourth 10% customer is expected, and the $105bn TAM gets revisited later this year.

-

**STX -2.7%** as CFO Gianluca Romano at Goldman replaced the old 50% incremental-margin frame: incremental gross margin has run ‘70+%’ and ‘for the next few quarters, that is a good way to look at the business.’ New: 80-90% of data-center volume on HAMR within a couple of years; edge pricing changed as high NAND prices let HDD-SSD overlap segments take price; storage is low-to-mid single digits of customer capex, leaving room to raise price ‘for a very, very long period’; a dividend increase is likely after an Oct-Nov review.

-

**SMCI -4.0%** as Mike Staiger at Goldman said ‘I think we are number 2. We clearly have sights on number 1’ and answered the ODM-migration bear case with ‘that concern was around at $7 billion, at $15 billion, at $22, at $39, and it is going to exist at $68.’ New: software and services finished FY26 at $538m; sovereigns are 10-25% of the market; capacity is expanding ‘under the covers’ beyond 6,000 racks a month; the Q1 gross-margin midpoint carried contracts dragged from the prior quarter.

-

**CRDO -4.5%** as CEO Bill Brennan and CFO Dan Fleming at Goldman gave the back half its shape: about 20% sequential growth into Q3 and 30% into Q4, putting ‘a billion-dollar quarter within our sights within early fiscal 2028.’ New: the ZeroFlap histories, xAI bringing clusters up in five days versus six to eight weeks and Oracle’s long links driving a check-engine-light telemetry DSP; ALCs at 30 meters in a cable 75% smaller by volume, demoed at OCP next month; copper still above 95% of data-center connections.

-

**AVGO -1%**

-

**Memory** weaker (**MU/SKHY/SNDK** -4-5%), potentially on the back of some Deepseek V4.1 improvements:


-

**SNDK -4.1%** as CEO David Goeckeler and CFO Luis Visoso at Goldman signed off on the simple downside math yesterday at GS: in a down scenario two-thirds of the business sits at the ~80% gross-margin floor and one-third at prevailing prices (’an easy way to think about it. Totally agree’). The eight long-term agreements are backstopped by third-party financial-institution guarantees that pay if a customer walks; Goeckeler refused to game out competitors adding supply (’should you do something uneconomic first?’), and said HBF is ‘just taping out’ with samples next year.

-

**LITE -5.4%** as President Wupen Yuen and IR head Kathy Ta at Citi yesterday attributed the $40 FY28 EPS target to OCS upside at the largest customer and said shortages are ‘still extending... we are still short everywhere.’ New: the optical scale-up market splits roughly 50/50 CPO and NPO, CPO is exclusively external light source and NPO is drifting that way, all of it needs 120-150mW-plus lasers, and the industry needs a second high-power source (Coherent) by 2H27 that ‘we frankly don’t know’ will be there.

# **INTERNET**

-

**META -1.4%** giving some back after the big Muse rally yesterday despite the JPM upgrade. Oppenheimer was a bit more mixed on the launch saying a 20% EPS uplift would require ~$27.5B of agent revenue at 80% incremental margins, equivalent to ~115M subscribers paying $20/month, which it views as a high bar. Oppenheimer also sees improved Siri/Apple Intelligence via Gemini as a competitive risk. Wells Fargo was more positive saying Muse could ultimately drive $90B+ of revenue, including ~$33B of subscriptions at 7% penetration plus ~$60B of advertising, while taking a somewhat more cautious read-through to SHOP as agents currently favor larger marketplaces over direct merchants.


#2 now:


-

**RDDT +6%** as Piper said RDDT users rose 8% m/m in August and audience growth accelerated to 18% y/y. Cleveland was also positive saying advertising momentum sounds favorable and they haven’t seen lower LLM citations hurting spend or budget plans.

-

**GOOGL +60bps** as Oppenheimer said GOOG could sell ~6.6GW of TPUs externally through 2028, worth ~$170B cumulatively, with annual TPU revenue ramping from ~$19B in 2026 to ~$50B/$101B in 2027/28.

-

**UBER +2.1%** as CEO Dara Khosrowshahi bought 141K shares for $10M at $71. He was also at Goldman disclosing a restructuring of about 10% of full-time staff and laid out the AV roadmap: nine partners live, 15 by year-end, no-vehicle-operator partnerships demonstrated in the US next year, ‘2028, 2029 is all about scaling,’ with access secured to 120,000-plus AV-ready vehicles and a claimed 30% utilization edge over first-party operators. Uber One is past 50m members at ~50% of bookings and was cast as a moat in price wars; Eats is close to Rides on bookings; no consumer trade-down.

-

**Z -2.8%** as COO and CFO Jeremy Hofmann at Goldman quantified the Preferred shift: revenue per connection grew 23% more than the legacy ad model in 2025 and should be ‘roughly 35% more’ in 2026, with mid-teens growth reiterated and macro now assumed down for the year. New: ShowingTime powers 90% of US tours, Follow Up Boss has ~138K monthly actives up 20%-plus, AI mode is live for 20% of logged-in users, and the FTC settlement leaves the Redfin partnership intact. Zillow Home Loans is now profitable.

-

**SPOT -0.2%** as co-CEO Gustav Soderstrom at Goldman yesterday argued the internet is shifting from a fixed-cost to a marginal-cost business, which Spotify answers with add-ons: the audiobooks add-on is ‘already on EUR 100 million ARR after only about a year,’ inference add-ons follow the template and ‘the margin is a managed outcome.’ Spotify is priced above competitors in every market with no churn, is training its own generative music model with artist opt-in and full royalty pass-through, and the long-term ambition is 1bn subscribers.

-

**CART +1.3%** as CEO Chris Rogers at Goldman yesterday launched Clementine, the AI grocery assistant, across North America the same day, saying its orders carry more items and a higher AOV than the $115 average, and drew the agentic line: third-party surfaces (OpenAI, Gemini and Claude integrations are live) are lead gen and ‘really quite small’ next to agents on Instacart itself. GTV growth has stepped up three straight quarters to 14% with 14% guided for Q3, Grocery Outlet went to nationwide price parity, and Costco Spain and France ‘exceeded our expectations.’

-

**EXPE +1.6%** as CEO Ariane Gorin at Goldman yesterday said July trends matched Q2 and travelers keep prioritizing travel despite oil and air fares, with Labor Day searches up 35% for vacation destinations. New: natural-language search captures 60% more intent data but ‘is not yet delivering better conversion’; product and tech cycle times are 40% faster on AI; paid ads in ChatGPT are being tested; agentic voice resolves 10% of Vrbo partner queries; B2B investments and mix ‘could put short-term pressure on margins’ while company-level expansion stands.

# **SOFTWARE**

-

**CRWD +0.5%** as CFO Burt Podbere at Citi ran the session through the April ‘Claude Mythos moment’ that became a record $333m of net new ARR (+51%), and said SafeMind, the security frontier model, is not in this year’s financials and is contemplated for next year on token-pack pricing. New: Exposure Management is nearing standalone disclosure, Project QuiltWorks (25 partners) has $400m of pipeline TCV, competition is ‘Microsoft. It is not even close,’ and Flex is about 40% of ARR with a 53% uplift on multi re-Flex.

-

**OKTA -0.9%** as Okta’s management at Goldman yesterday priced its agent product as a simple per-user uplift on mostly one-year deals, conceded the model ‘will likely evolve,’ and pushed back on the new-entrant claim that agent identity needs a different architecture: ‘we do not agree.’ Dozens of transactions in the first 120 days since the April 30 GA, new products at 30% of bookings, Microsoft’s Agent 365 ‘we don’t see them a lot yet,’ CrowdStrike framed as complementary, and the cross-app access spec now adopted as an MCP extension.

-

**SNOW -0.5%** as CFO Brian Robins at Citi argued the raise from 31% to 36% growth is structural, with Cortex adopters consuming more of the core, and lifted the hood on guidance: a data-science team forecasts every customer and product line nightly, while new products are guided off launch curves with haircuts (CoCo only now carries real confidence). The SBC lever is headcount: ~330 year-to-date hires versus ~950 last year, about 180 of them from Observe, with engineering restaffed toward ‘AI-pilled’ hires one to two years out of college.

-

**VEEV +0.1%** as CFO Brian Van Wagener at Citi described Falcon as a standalone platform outside Vault acting as ‘a power user’ of the applications, priced per business transaction so Veeva is not ‘marked down’ to model cost, targeting ~70% of manual MLR labor industry-wide. CRM: 12 of the top 20 with two to go by year-end, and ‘not aware of any customers live on Salesforce CRM products.’ Commercial subscriptions grew ~13% versus a high-single-digit plan; Q3 is where the R&D step-down shows; Investor Day is Nov 5.

-

**DOCN -1.2%** as CEO Paddy Srinivasan and CFO Matt Steinfort at Citi put numbers on the unit economics: ARR per megawatt of $22m in Q2 versus $9-10m for neoclouds’ embedded base, incremental economics 30-40% higher, and token serving at ~60% utilization today with a 70-80% ambition via time-zone routing and overnight batch. The open-weight gap has closed to two to three months from six to nine, and the Inference Engine has over 6,000 customers, 99% product-led. No 2027 update until next earnings; H100/H200 pricing is ‘going up, not down.’

-

**MDB +4.3%** as CEO CJ Desai and CFO Mike Berry at Goldman, a day after Citi, said a Fortune 100 customer has made MDB the default ‘unless proven otherwise,’ ElevenLabs is now above $500m ARR on Atlas, and a frontier lab uses Atlas as conversational memory across two products. Self-managed is structural, not episodic: EA ARR is above 10% for three straight quarters, with the EA cohort’s Atlas consumption growing faster than the company. FY27 carries a ~30% engineering step-up; Voyage adds come ‘via Claude Code and Codex, period.’

-

**DDOG -1.6%** as CFO David Obstler at Goldman said the largest customer renewed and is ‘growing rapidly and spending above the commitment,’ with guidance de-risked to include only the commitment, and that hyperscalers have come asking to use Datadog for training and post-training, ‘a surprise to us.’ New: Federated Logs (logs staying in ClickHouse or Databricks) and Bring Your Own Cloud open the platform to data it never ingests; customers are being moved to the Infinite Cardinality Metrics SKU, called retentive and margin-enhancing; R&D at 30% of revenue is ‘out-investing everybody.’

-

**HUBS -2.9%** as CEO Yamini Rangan at Goldman yesterday disclosed that the OpenAI, Anthropic and Gemini connectors each have 350,000-plus weekly users, with connector cohorts retaining about 12 points higher, and that several use cases moved to open-weight models in the past six weeks at 80% of the cost, helping the gross-margin trajectory. She pushed back on the SEO-to-AEO air-pocket read of Q2’s soft net adds, tying it instead to deliberately longer 14-30-day agent trials, and said ‘even though AEO is nascent, the volumes are not there’ to replace SEO.

-

**ADSK +2.4%** as CEO Andrew Anagnost and CFO Janesh Moorjani at Goldman yesterday, a week before Autodesk University, dismissed OpenAI’s Astra as ‘a perfect example of misunderstanding where Autodesk value is... The value is not in creating the model. The value is in using the model.’ The next-generation Autodesk Assistant unveiled at AU is ‘very much agent first’; Neural CAD models are million-parameter, deterministic and cost pennies to run; consumption is already 16-17% of the business with no step change expected at $8bn-plus of revenue.

# **ELSEWHERE**

-

**AAPL +3.5%** as investors got excited about Duo launch. *AB *lays out bull case in TMTB Slack:


-

**COIN -1.4%** as CFO Alesia Haas at Citi, a day after Goldman, confirmed 12 product lines each above $100m annually, said Coinbase One passed 1m paid subscribers with benefits still spot-only (extending them to derivatives and prediction markets is an untapped lever), and named Sep 15 as the CLARITY cloture date, ‘cautiously optimistic’ but it ‘does not change our roadmap.’ Prediction-market revenue is all incremental with no cannibalization and no marketing yet; Circle is now ‘frenemies’; derivatives are 70%-plus of the crypto market still to capture.

-

**HPE -6.2%** as IR chief Shannon Cross at Goldman said the FY26 networking-for-AI order target has gone from $1.5bn to $2.5-3bn, with the gigawatt Oracle win ‘a competitive bid’ won on the liquid-cooled Tomahawk 6 switch and talks under way with others. New: AMD Helios goes to market as full HPE racks plus white-label trays for competing server makers; the $3.5bn deal is a hyperscaler buying mostly CPU compute for on-prem inference at traditional server margins. Cloud & AI margin is guided to 13% for FY27 versus 17% in Q3.

-

**INTU -0.4%** as CFO Sandeep Singh Aujla at Goldman explained the FY27 J curve: Intuit lost share among sub-$50K-AGI filers and will take a deliberate ARPC hit to be price-competitive there, monetizing through the consumer platform instead (35% of tax customers took early refund access). New: SEO is only ~20% of the QuickBooks funnel and AI-referred traffic converts ‘multiples higher’; no AI-native tax competitor showed up this season; a $150 price-disruptive assisted offer against $400-500 elsewhere; ~700K QuickBooks entities remain upgradeable to mid-market tiers. Investor day is next week.

-

**S +1.9%** as CEO Tomer Weingarten at Goldman swung at the industry’s AI-attack narrative (’fundamentally, there’s nothing new... Products are failing. Everybody’s failing’) and argued AI alignment will supersede every other cybersecurity problem, solved external to the model and possibly needing a new chip design or stronger enclave. He dismissed red/blue agent loops and fine-tuned security LLMs, called Chinese open-source models roughly 5x better than Nemotron-class bases, and said rivals compete on ‘bigger microphones’ and FUD. CFO Sonalee Parekh said the guide was raised by ‘significantly more than the beat.’

-

**PYPL +2.2%** as CEO Enrique Lores at Goldman yesterday said branded checkout growth this quarter should land ‘in the 1%-2% range’ because the European tariff hit to cross-border volume, especially from China, is bigger and longer than the half-to-one-point originally expected, while affirming EPS and transaction-margin guidance. On strategic alternatives the standalone plan is the benchmark and ‘we will always objectively recommend whatever option we think is better’; the company is now three businesses with segment reporting next year, and M&A enters the picture ‘in the next quarter.’

# **BULLISH AND BEARISH WEEKLY OPTION FLOW**


_Source: RunningoftheBulltards_