Enter password to access
| Factor | Current | Q | Hist. Fwd 6mth % | Wt |
|---|
Where current valuations rank vs. historical range (green = cheap, red = expensive)
Average monthly relative returns (Stock Return โ Index Return) by calendar month
| Ticker | Company | EV/Rev | EV/EBITDA | P/E | EV/FCF | Avg Val | GR% | ||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| CY26 | CY27 | CY26 | CY27 | CY26 | CY27 | CY26 | CY27 | ||||
| Loading sector comparisons... | |||||||||||
Current selection รท comparison โข >1 = more expensive | <1 = cheaper โข Select sector and/or peer to compare
The semiconductor industry is entering 4Q25 with both cyclical recovery and structural AI growth aligned โ a rare and powerful setup.
Investor focus has shifted away from quarter-to-quarter noise toward multi-year earnings durability through 2026โ2028.
The print reset investor confidence that AMAT can grow at/above WFE cycle rather than undergrow it โ which is what its valuation had been implying. Morgan Stanley calls it a "leap toward shifting the narrative."
Q1 Results:
The Big Tell โ Guidance:
Key point: Management went beyond typical "one-quarter-at-a-time" cadence by anchoring CY26 more firmly. Stifel calls this confidence "a testament" to positioning.
What's NOT driving upside: NAND remains "modest" (~10% of WFE), ICAPS flat. These are no longer headwinds but not contributors.
Cleanroom availability is now the "governor" on how fast demand converts into shipments:
Synthesis: Demand > near-term physical readiness = backlog converts over longer window rather than "peaking and rolling."
One of the biggest relief valves versus prior skepticism:
Conclusion: China/ICAPS now less of a near-term earnings trap, more of a multiple suppressor that can unwind.
Multiple desks connect earnings reset to discount closing narrative:
AMAT used this print to reclaim "cycle leadership" status with:
Practical implication: Rerate candidate (narrowing discount) if company converts visibility into shipments without supply chain stumbles.
| Theme from AMAT | Read-Through | Beneficiary |
|---|---|---|
| DRAM/HBM strongest driver | Memory equipment demand confirmed; HBM undersupply structural | LRCX (memory torque), MU (memory producer) |
| Cleanroom availability gating | Demand > capacity = extended upcycle, 2027 visibility | All semicap โ cycle duration extends |
| Leading-edge foundry/logic | 2nm/3nm ramps on track; TSMC capex visibility | ASML (EUV monopoly), TSMC |
| Advanced packaging demand | CoWoS/HBM integration driving new equipment categories | AMAT, LRCX, TSMC |
| China/ICAPS stabilizing | Removes overhang that suppressed semicap multiples | AMAT (most exposed), LRCX |
Current exposure: LRCX +1.2% (borderline due to valuation), ASML +2.0% (conviction)
AMAT consideration: Not currently in portfolio. AMAT's rerate thesis is compelling, but:
Watchlist status: Monitor AMAT for (1) China policy clarity, (2) valuation discount narrowing confirmation, (3) potential swap with LRCX if memory thesis matures.
Expected softness in early 2026 is widely anticipated and viewed as an entry opportunity, not a structural risk.
AI capex has overwhelmed traditional risks (China, cyclical pauses, trade data).
This Is Not a Typical Semi Cycle: AI has transformed WFE and semi-cap from a short-cycle trade into a multi-year earnings durability story extending into 2028.
Memory + AI = Sustained Capital Intensity: HBM undersupply, AI server content growth, and process complexity anchor elevated WFE spend.
Bottom line: Bullish semi-cap stance into 4Q25 and beyond, with AI-driven WFE growth underpinning above-consensus earnings power through 2026โ2028.
Software has materially underperformed other tech subsectors (especially semis, AI infrastructure, memory) and has undergone significant multiple compression.
Software sector trading 21โ34% below trailing 5โyear averages on valuation depending on growth cohort. FCF multiples remain 50โ70% below peaks.
Broad underperformance driven by: AI disintermediation fears, weaker seatโbased demand narratives, slowing enterprise budgets, mixed Q4 software prints from bellwethers (MSFT, NOW, SAP).
Conclusion: Valuations have reset deeply. Some see this as the first credible "buy software" moment in a while โ especially in highโquality names attracting early generalist accumulation.
Conclusion: Enterprise AI deployments in 2026 expected to stabilize growth narratives for highโquality SaaS companies rather than replace them.
Vertical Software
Security Software
Conclusion: Consumptionโbased names may outperform subscription names through 2026.
Conclusion: Capital discipline and efficient AI reinvestment determine winners. Emphasize companies with structurally improving FCF and controlled SBC.
Conclusion: Software may remain under pressure until AI semi cycle decelerates or compute scarcity moderates, enabling rotation back.
Positive Setups:
Neutral to Cautious:
Where analysts see opportunity now:
Where caution is advised:
DDOG, SNOW, MDB, NET, and PLTR are infrastructure-centric growth software names with durable AI-driven demand.
Valuation compression has outpaced any real deterioration in fundamentals. Hyperscaler-led cloud acceleration is the critical swing factor.
Key insight: Observability, data platforms, databases, and edge networks are framed as beneficiaries of AI complexity โ not victims of automation. This is the opposite of application-layer SaaS.
| Stock | Street View | Key Catalyst |
|---|---|---|
| DDOG | "AI-safe" best-in-class | $100M+ TCV wins, AI-native customers |
| CRWD | Security platform leader | Module adoption, post-outage recovery |
| PLTR | Idiosyncratic AI operating layer | Production AI deployments, AIP |
| NET | Edge/security/AI intersection | Workers/R2 monetization, edge AI |
| MDB | Structurally sound, sentiment-challenged | New management, execution consistency |
NVDA is widening its platform beyond GPUs: selling Arm-based 88-core "Vera" CPU stand-alone and deepening "extreme co-design" with hyperscale partners (CoreWeave), expanding TAM into head-end servers. This explicitly pressures x86 share.
Custom accelerators are now a durable pillar: Google's TPU roadmap (Broadcom) accelerates through CY27 with 6-7M units targeted; Microsoft launched Maia 200 inference/training ASIC (TSMC 3nm). Verticalized ASICs are a durable complement to GPUs for specific workloads (inference at scale, cost per token).
NVLink scale-up vs Ethernet scale-out:
MRVL building end-to-end scale-up franchise: XConn deal adds PCIe/CXL switches for memory pooling, larger context, lower TCO โ aligned with UALink roadmaps.
Memory content and value-share keep rising:
Hyperscaler designs dramatically increasing HBM per accelerator, shifting system BOM toward memory. Price strength and multi-year visibility continue.
Anchored by TSMC:
Platform proofs: NVLink 5 switch on TSMC 4NP; Maia 200 on TSMC N3 โ TSMC central to AI accelerators and custom silicon ramp.
Position sizing should assume periodic "supply scare" drawdowns
Maia 200's Ethernet-based scale-up and domain size reinforce that networking silicon is a parallel profit pool to GPUs/XPUs:
Beyond GPUs, critical differentiators for latency, context size, and TCO:
Total hyperscaler capex approaching $716B in 2026E, $885B in 2027E (MS estimates). GOOGL alone at $185B โ 32% larger than total DC spend across six largest players three years ago. This is the demand engine for Growth Semi names.
The 4Q25 earnings cycle confirmed three structural themes:
| Metric | META | GOOGL | NFLX |
|---|---|---|---|
| Revenue | $59.9B (+24%) | $113.8B (+15%) | $12.05B (+18%) |
| vs Consensus | Beat (+3%) | Beat (+1.4%) | Beat |
| Op Margin | 41.4% | 31.6% | 24.5% |
| Key Highlight | $8.02 EPS vs $6.77 | Cloud +48%, $240B backlog | 325M members |
Key Insight: MS frames GOOGL + META as the "second indicator" (after META's guide) that leading scaled companies with data, reach, and investment capacity are seeing flywheel benefits accelerate โ "the gap between them and smaller players is likely to widen faster than expected."
| 2026 Guidance | META | GOOGL | NFLX |
|---|---|---|---|
| CapEx | $115Bโ$135B | $185B (MS est) | โ |
| 2027E CapEx | $144Bโ$151B | $250B (MS est) | โ |
| Y/Y Growth | +60โ85% | ~100% | โ |
| Content/OpEx | $162Bโ$169B (+40%) | Higher D&A | $20B content |
| FCF Impact | $5Bโ$15B (trough) | $26Bโ$18B ('26โ'27) | Buybacks paused |
Combined META + GOOGL CapEx: $320B in 2026, $400B+ in 2027 (MS estimates). Total hyperscaler capex approaching $716B in '26, $885B in '27. GOOGL's '26 capex alone ($185B) is 32% larger than total DC spend across six largest players three years ago.
| Stock | Rel %ile | Signal | 4Q Rev Growth |
|---|---|---|---|
| META | 8th | STRONG | +24% |
| GOOGL | 29th | MODERATE | +15% |
| NFLX | 3rd | MODERATE | +18% |
Paradox: Strongest fundamental quarter in years, yet all three trade at historically cheap valuations. Market is pricing in "peak growth" or investment cycle concerns that may be overdone.
Combined Internet Weight: 33% of portfolio (vs ~24% benchmark) โ significant overweight based on valuation + execution combination.
The $320B combined 2026 capex from GOOGL + META isn't an abstraction โ it flows into specific hardware, foundry, and interconnect layers. Understanding the supply chain clarifies both the ROIC path (what are they getting for the money?) and the execution risk (where are the bottlenecks that could delay returns?).
GOOGL and META are pursuing different compute strategies with different cost structures and vendor dependencies:
Portfolio read: GOOGL's TPU program makes AVGO a direct beneficiary of GOOGL capex. META's GPU dependence makes NVDA the primary beneficiary. Both strategies flow through TSMC foundry (N3/N5 for custom silicon, CoWoS for packaging). The AI Infrastructure theme in the portfolio (+14.5% active) is structurally linked to the Internet theme (+6.8% active) โ they're two sides of the same capex cycle.
HBM (High Bandwidth Memory) is the bottleneck for both hyperscalers' AI ambitions:
Portfolio read: MU (+1.3% active, +96% revisions) directly benefits from this dynamic. Memory supply constraints = GOOGL/META compete on allocation, supporting MU pricing power. The risk is symmetric: if HBM supply loosens faster than expected, MU's premium valuation (100th percentile) gets tested.
Interconnect architecture is a strategic differentiator, not just plumbing โ and it directly impacts inference cost-per-token for both hyperscalers:
Portfolio read: MRVL (+2.0% active) and watchlist candidate ANET are direct beneficiaries of Ethernet-centric inference buildout. GOOGL's networking investment creates revenue (Cloud), META's creates cost. This is another facet of the GOOGL vs META capex ROI visibility gap.
Both GOOGL and META ultimately depend on TSMC for everything:
Portfolio read: TSMC (+2.5% active) and ASML (+2.0% active) are structural enablers of GOOGL/META capex. If either TSMC or ASML can't deliver capacity, hyperscaler capex guidance becomes aspiration rather than commitment โ and Cloud backlog conversion timelines slip. Foundry capacity is the supply-side constraint on demand-side AI monetization.
Data center power is becoming the binding constraint on hyperscaler buildout โ and it directly impacts the capex-to-revenue timeline:
Portfolio read: Power constraints create execution risk for the capex guidance that underpins both Cloud backlog conversion (GOOGL) and AI ad efficiency gains (META). If power/cooling delays push out capacity timelines, the FCF trough extends. This is one reason MS models GOOGL FCF declining through 2027 โ capex front-loaded but revenue recognition lags capacity availability.
Key insight: The portfolio's AI Infrastructure theme (+14.5% active) and AI Monetization theme (+6.8% active) are not independent bets โ they're the supply and demand sides of the same AI buildout cycle. GOOGL/META capex is NVDA/AVGO/MU/TSMC/ASML/MRVL revenue. If internet capex disappoints, semis get hit. If semi supply constrains, internet revenue timelines slip. The portfolio is long both sides of this flywheel by design.
Street consensus vs reality โ what matters for positioning.
What Mattered in the Print:
Street Debate:
Cash generation trough in 2026 (FCF compression + depreciation ramp) makes near-term valuation optically worse
"Meta-last-year dรฉjร vu" โ AI productization โ engagement โ monetization โ operating leverage with a lag
Positioning / Flow:
Blended Ern% of -7.4% is misleading. Component breakdown (Jan 23 โ Feb 20):
FCF collapsed $78.9B โ $19.1B entirely on $175-185B capex guide. Operating business is accelerating โ blended Ern% distorted by capex flow-through into FCF. Revenue, EBITDA, EPS revisions are among the strongest in the portfolio. Supports +1.9% active overweight.
Rating: Overweight, PT $330 (stock at $333 = ~0% upside). MS frames GOOGL + META as the "second indicator" that leading scaled companies are seeing flywheel benefits accelerate โ "the gap between them and smaller players is likely to widen faster than expected."
4Q25 Detailed Results (vs MS Estimates):
โ๏ธ Cloud Backlog โ The Structural Differentiator vs META:
๐ค AI Productization Breadth:
โ ๏ธ FCF Compression โ Most Severe in Portfolio:
MS models $50B short-term debt in 2026, additional $50B in 2027. Could shift to net debt. But operating margin expands 32% โ 36% โ 35.5% and incremental EBITDA margin of 83% in 2026E shows real operating leverage even through the capex ramp.
๐ MS Estimate Changes (Post-4Q25):
๐ฏ Valuation & Risk Scenarios:
Cloud backlog story is excellent and differentiates vs META structurally. But stock at/above PT with zero margin of safety. Research confirms thesis (hold +1.9% OW) but doesn't build a case for adding. The valuation-to-upside gap vs META is the key relative sizing input โ META offers 10-35% upside across three analysts, GOOGL offers ~0% per MS. Supports current positioning: GOOGL hold, META has stronger case for conviction upgrade to +2%.
What Mattered in the Print:
AI + Monetization Narrative:
Street Consensus:
Positioning / Flow:
Blended Ern% of -8.2% masks strong operating revisions. Component breakdown (Jan 23 โ Feb 20):
FCF cut from $21.5B โ $10.2B on capex guide. Revenue revisions strongest of the three hyperscalers (+10.6%). Ad machine funds the AI spend โ stock flat (-0.5%) because market sees through the FCF optics. Supports +1.5% active overweight.
JPM (Doug Anmuth, OW, PT $825), Bernstein (Mark Shmulik, OP, PT $900), BMO (Brian Pitz, MP, PT $730). Two bulls + one skeptic = strongest consensus setup in portfolio. Even BMO's Market Perform implies 9% upside from $669.
4Q25 Results โ Beat Across All Metrics:
๐ฅ 1Q26 Guide โ The Acceleration Signal:
๐ค Seven AI Monetization Paths (Bernstein Framework):
JPM adds: AI ad tools closing the SMB long-tail โ millions of advertisers who previously couldn't afford creative production now can. TAM expansion, not just optimization.
โ ๏ธ Key Risks Flagged Across All Three Analysts:
๐ Three-Analyst Estimates Comparison:
| Metric | JPM | Bernstein | BMO |
|---|---|---|---|
| PT | $825 | $900 | $730 |
| 2026E Rev | $224B | $221B | $218B |
| 2026E EPS | $29.30 | $30.94 | $29.02 |
| 2027E EPS | $31.80 | $37.18 | โ |
| 2026E Capex | $132B | $127B | $132B |
META at +1.5% active vs GOOGL at +1.9% may understate relative conviction. META has: (1) faster revenue growth (+24% vs +15%), (2) better near-term upside (10-35% vs ~0%), (3) strongest revenue revisions of any hyperscaler (+10.6%), (4) seven identified AI monetization paths with $10B already running. GOOGL has: (1) Cloud backlog providing structural capex ROI visibility META lacks, (2) superior operating leverage through capex ramp (margin expanding vs compressing), (3) broader AI surface area across Search/Cloud/YouTube/Workspace. Research supports META for conviction upgrade to โฅ+2%. GOOGL hold at +1.9% โ thesis confirmed but priced for it. Per Lesson 7: monitor, update notes, but no urgency to trade.
Core Operating Reality:
Overhang: WBD / PSKY Situation
AI Narrative (Relative Weakness):
Positioning / Valuation:
The market now differentiates between types of AI spending:
| Stock | Positioning Dynamic | Current State |
|---|---|---|
| META | Heavy short base โ violent unwind | Real money replacing shorts |
| GOOGL | Capitulation selling post-capex shock | Digestion phase โ flows stabilizing |
| NFLX | M&A uncertainty dominating | Sentiment battleground |
These themes recur across software, semis, internet, and telecom research โ they represent the market's core debates.
The market is pricing a binary outcome where reality is gradient. AI changes software economics (pricing power, seat expansion, labor leverage) rather than eliminating them.
What to watch: AI monetization inside incumbents (bundled pricing, usage-based attach) rather than new logo losses.
AI demand is real, durable, and still supply-constrained in memory, advanced logic, networking, and packaging. This is infrastructure build-out with multi-year visibility, not a speculative bubble.
What to watch: Memory contract pricing into mid-year; whether hyperscalers lock in multi-year supply (structural vs cyclical signal).
The argument that AI capex is reckless is repeatedly challenged. Capability jumps require disproportionate infrastructure before monetization shows up.
What to watch: Disclosure on internal vs external compute usage; whether AI services begin to show margin inflection.
The sharp bid into telco, networking, and "defensive tech" is flow-driven de-risking, not a re-rating based on improved fundamentals.
What to watch: Whether software volatility subsides before fundamentals stabilize โ a tell that positioning, not earnings, was the driver.
Moving from search and discovery to delegated action (booking, buying, managing). This reframes the ad vs subscription debate.
Analysis of three potential portfolio additions in the Internet sector with distinct positioning and risk profiles.
| Name | Inbox Tone | Key Driver | Positioning |
|---|---|---|---|
| SPOT | โ Strongly bullish | Margin durability + AI upside | Rebuilding โ owned |
| LYV | โ Constructive | Visibility + venue economics | Steady compounder |
| RBLX | โ ๏ธ Positive but debated | Engagement + monetization optionality | Owned โ selectively trimmed |
AI went from existential risk โ competitive advantage this month:
Internet names increasingly differentiated by whether AI is tool vs. threat
Key contrast: SPOT = rebuilding confidence, underowned โ owned. RBLX = owned, debated, selectively trimmed on rallies.
The 4Q25 cycle was the strongest fundamental quarter for Internet mega-caps in years. The market is focused on investment intensity (rightfully so โ $300B+ combined CapEx), but is underweighting:
Sector Verdict: Overweight Internet vs benchmark. The "investment cycle overhang" is real but already priced. When FCF re-accelerates in 2027+, these stocks will re-rate significantly.
Core sector thesis: In a market rotating from long-duration growth to shorter-duration cash flow visibility, Cable/Telco offers structural stability with no embedded growth assumptions to destroy. TMUS is the only telecom expression in the portfolio โ a structural outperformer, not a defensive rotation trade.
| Metric | TMUS | VZ | T |
|---|---|---|---|
| Rolling 1Y Win Rate vs Index | ~65%+ | 18% | 28% |
| Head-to-Head: TMUS Wins | โ | 94% of periods | 79% of periods |
| Cumulative Relative Return | Outperformer | -215% combined | -215% combined |
| Portfolio Action | +2.0% active | Eliminated | Eliminated |
Lesson 5 Applied: "Defensive โ All Telecoms." Head-to-head peer analysis revealed TMUS as the structural winner. VZ at 5th percentile EV/Rev was still a chronic capital destroyer. Low valuation doesn't fix a chronic loser.
Why +2.0% Active Weight:
Research Debate:
"The sharp bid into telco is flow-driven de-risking, not a re-rating... a temporary hiding place that unwinds quickly." Defensive rotation trade, not structural.
We own TMUS for structural reasons (beats VZ 94%, T 79% head-to-head over 7+ years), not as a defensive rotation trade. Fixed wireless broadband is incremental TAM. AI optionality is unpriced upside.
The current market regime is not risk-on/risk-off โ it's a duration rotation. Markets are rotating from long-duration growth (software, -35% off 24M peak) to shorter-duration cash-flow-visible alternatives.
CHTR and CMCSA are in the benchmark but not the portfolio. The underweight reflects:
Portfolio positioning: Underweight cable is an implicit bet that wireless (TMUS) continues to take broadband share. If fixed wireless penetration plateaus, cable could re-rate โ but data doesn't support that thesis yet.
Core thesis: Hyperscaler capex continues at record levels ($200B+ annually). AI infrastructure is the picks-and-shovels play โ these companies benefit regardless of which AI applications win. We're in the "build-out" phase where hardware/infrastructure spending grows faster than software monetization.
โ ROIC Stress Test (Feb 2026): Markets now ask "Who earns acceptable returns on AI investment?" not "Who benefits from AI?" Our infrastructure names earn returns on others' AI spend rather than consuming balance sheets. NVDA, AVGO, MRVL sell picks-and-shovels; TSMC/ASML have monopoly pricing power. This theme passes the capital allocation test.
Why largest overweight: 80% rolling win rate (best in portfolio), 13th percentile P/E (historically cheap despite the run), +45% earnings revisions.
Blackwell demand exceeds supply. Data center is 80%+ of revenue. CUDA moat deepening. Training AND inference markets expanding.
Why conviction: Foundry monopoly for leading-edge AI chips. +57% earnings revisions. N3/N2 capacity sold out through 2026.
Makes chips for NVDA, AAPL, AMAT, QCOM. CoWoS advanced packaging bottleneck = pricing power. Arizona fab de-risks geopolitics.
Why added: 75% win rate. Custom silicon for Google TPUs (8M+ units by CY27), AI networking leadership, VMware integration.
Inverted signal but NOT a value trap โ quality compounder. Zero weight was costing -23% alpha over 3Y. Core AI enabler.
Why conviction: 52nd percentile (fair value), custom silicon wins with hyperscalers, AI networking (BDOQ/L6T switches).
Amazon, Microsoft custom chip programs. Different risk/reward than NVDA โ more levered to networking than training.
Why conviction: EUV monopoly โ no alternative exists. 2027 earnings inflection (~55% YoY EPS growth). High-NA ramping.
Every leading-edge chip requires ASML tools. DUV pull-through underappreciated. Backlog extends visibility through 2026.
Why sized smaller: 21% win rate (data doesn't confirm thesis yet), but 6th percentile EV/Rev and compelling AI observability narrative.
"AI-safe" infrastructure software. Caution: Only long-duration software exposure โ duration rotation is a headwind. Sized at 2.0% reflects this uncertainty.
Core thesis: Memory and equipment are inflecting after 2023 trough. HBM demand for AI accelerators is structural. NAND inventory normalized. But valuations are stretched โ position with caution.
Why conviction: +96% earnings revisions (highest in portfolio). HBM3e leadership for AI accelerators. Memory pricing power returning.
100th percentile valuation is a concern, but revisions support it. AI servers need 8x more DRAM than traditional. Cycle has legs.
Why borderline: 100th percentile valuation (most expensive ever). Strong thesis (WFE share gains, NAND/DRAM recovery) but valuation stretched.
CY26 EPS estimates rising to $5.25+. Share gains in etch/deposition. But +0.9% active weight reflects uncertainty at these levels.
Core thesis: Defensive tech isn't an oxymoron. These stocks have lower beta, stable business models, and held up better in 2022. They're insurance against a risk-off environment โ and insurance has a cost in bull markets.
โ Duration Rotation (Feb-Mar 2026): This is NOT risk-on/risk-off โ it's a duration rotation. Markets rotating from long-duration growth (software, -35% off 24M peak) to shorter-duration cash-flow-visible alternatives (hardware, telecom, analog). HOLT drawdown study (3/2/26) confirms: at -35%, software outcomes are historically mixed (52% hit rate). No urgency to buy the dip โ recoveries take 12-24 months. TXN benefits from "no embedded growth assumption to destroy." ADI under review โ approaching no man's land.
Why conviction: EDA duopoly with SNPS โ every chip design needs their tools. Only -24% in 2022 (defensive). Record backlog model.
AI chip complexity increasing EDA intensity. System design TAM expanding. 78th percentile is concern but structural position warrants premium.
Why conviction: Analog bellwether with massive fab investment paying off. -24% in 2022. Dividends + buybacks. Industrial/auto exposure diversifies from AI.
300mm fab capacity ramping. Inventory digestion ending. SLAB acquisition adds embedded/wireless. Group trades together โ TXN up = analog up.
Why borderline: Best-in-class analog margins (71%+). Datacenter growing to $2B+ run-rate. But 98th percentile valuation limits conviction.
Cleanest analog story (industrial + test + datacenter). +0.7% active reflects uncertainty at these valuations. Would add on pullback.
97th percentile EV/Rev โ near most expensive ever. Redundant to NVDA exposure. +0.2% active = no conviction. If bullish on AI GPUs, NVDA is the better bet with 80% win rate vs AMD's 63%.
Weak signal: Valuation has near-zero correlation with forward returns. When data says "no edge," don't make an active bet. Azure growth solid but no valuation timing opportunity.
๐ Feb 20 Update โ Best alpha call in portfolio (+65bp in 4 weeks). Stock -15.2% while BM -2.3%. Revisions still positive (Rev +4.0%, EBITDA +5.1%, EPS +4.8%, FCF +2.5%) but weakest of three hyperscalers on revenue growth. Unlike GOOGL/META, almost zero revision change around earnings event โ market punishing lack of top-line acceleration and Copilot monetization narrative, not estimate deterioration. At 0.89x Rel_PE, MSFT now trades at a discount to the Tech Index โ unusual. Either more downside ahead or valuation is getting compelling.
100th percentile: Most expensive in its history relative to sector. STRONG signal but valuation fully stretched.
Challenge: If market favors "FCF machines and capital discipline," AAPL is best-in-class mega-cap on that dimension. Our underweight is valuation-driven, but duration rotation could favor AAPL.
NOW: WEAK signal at "cheapest ever" (0th pctl) โ -37.5% last year. Cheap doesn't mean buy.
Core thesis: 4Q25 proved AI monetization is real, not theoretical. These companies are converting AI investments into actual revenue and margin expansion. The "show me the money" phase has begun.
โ ๏ธ Capex Scrutiny Intensifying (Mar 2026): GOOGL $185B and META $127-135B face growing ROIC questions, but structural difference is clear: GOOGL's $240B cloud backlog (+55% Q/Q) provides contractual revenue validating heavy capex. META's return must come entirely through ad monetization โ at $10B AI ad run-rate, 7 paths) but inherently less visible. MS: GOOGL op margin expands 32%โ36% through ramp; META compresses 41%โ34% before recovery. NFLX weakest on "mission-critical" test โ AI defensive not offensive.
Why conviction: Revenue accelerating to ~30% Y/Y guided (1Q26). Seven AI monetization paths with $10B run-rate. Three analysts: PTs $730-$900 (10-35% upside). Strongest revenue revisions of any hyperscaler (+10.6%).
Capex without cloud revenue (BMO concern) is the key structural risk vs GOOGL. Margin compresses 41%โ34% in '26 before recovering. But revenue growth + AI ad flywheel compounding = strongest near-term risk/reward case among mega-caps. Candidate for +2% conviction upgrade.
Why conviction: Cloud +48% Y/Y with $240B backlog (+55% Q/Q). Search AI driving fastest ex-FX growth in ~4 years. Waymo optionality undervalued.
Op margin expanding 32%โ36% through capex ramp. MS models GCP at 73% growth in 2026E. Structural cloud backlog advantage vs META. FCF trough โ patience required.
Why conviction: Ad tier scaling, password crackdown tailwind. 3rd percentile EV/Rev and compelling AI observability narrative โ cheap.
Caution: Weakest on "mission-critical" test โ content/media, not infrastructure. AI is defensive not offensive. M&A overhang.
Why conviction: Cleanest sentiment inflection this month. Margin durability + AI upside (DJ, discovery, personalization). Pricing power validated.
4Q25 beat + guide surprised positively. Gross margin expanding. "AI kills DSPs" narrative fading. Rebuilding โ owned positioning.
Core thesis: Defensive tech isn't an oxymoron. TMUS has lower beta, stable business model, and held up better in 2022. It's insurance against a risk-off environment โ and insurance has a cost in bull markets.
โ Duration Rotation (Feb-Mar 2026): This is NOT risk-on/risk-off โ it's a duration rotation. Markets rotating from long-duration growth (software, -35% off 24M peak) to shorter-duration cash-flow-visible alternatives. HOLT drawdown study (3/2/26) confirms: at -35%, software outcomes are historically mixed (52% hit rate). No urgency to buy the dip โ recoveries take 12-24 months. TMUS benefits from "no embedded growth assumption to destroy."
Why largest defensive bet: Only telecom worth owning. Beats VZ in 94% of rolling 1Y periods. Beats T in 79%. Was UP +10% in 2022 when NVDA fell -66%.
Wireless share gains continuing. 5G fixed wireless expanding TAM. 11th percentile P/E โ cheap for quality. Eliminated VZ (18% win rate) and T (87th pctl PE).
18% rolling win rate โ loses to index 4 out of 5 times. -215% total relative return over 7 years. 5th percentile valuation didn't help. Chronic loser regardless of how cheap it gets.
87th percentile P/E โ more expensive than VZ! 28% win rate. -213% total relative return. If you want defensive telecom, TMUS beats T in 79% of head-to-head periods. No reason to own.
CMCSA: Inverted signal + structural decline (cord-cutting) = true value trap.
| Sector | Ticker | Company | Port Wt% | Bmk Wt% | Active Wt% | Theme |
|---|
3/6/26 โ Rebalance #2: Benchmark weights recalculated using actual S&P 500 GICS sector weights. MSFT flipped from largest underweight to +2.5% overweight. ADI sold. VZ and RBLX added to Comm Svcs. Combined view uses actual within-sector active weights (not index-scaled). Tech 75.2% / CS 24.8% of S&P 500 two-sector universe.
2/27/26 โ Rebalance #1: NVDA trimmed, MSFT added, GOOGL/META trimmed, CDNS added, TSMC trimmed.
1/23/26 โ Inception: 18-stock portfolio launched. Tech + Comm Svcs sectors.
Complete history of portfolio changes, rebalances, and benchmark adjustments.
| Date | Type | Details |
|---|---|---|
| 2026-01-23 | INCEPTION |
Portfolio launched โ 18 stocks (Tech + Comm Svcs) Tech (13): NVDA 21.2%, AAPL 11.5%, MSFT 7.6%, AVGO 7.4%, TXN 3.5%, MU 2.8%, TSMC 2.5%, CDNS 2.5%, LRCX 2.3%, ADI 2.1%, DDOG 2.0%, ASML 2.0%, MRVL 2.0% CS (5): GOOGL 14.9%, META 7.2%, TMUS 3.5%, NFLX 3.2%, SPOT 1.8% Benchmark: blended Tech/CS index (^SP500-45 / ^SP500-50) |
| 2026-02-27 | REBALANCE #1 |
Rotational rebalance โ 6 positions adjusted BUY: MSFT +2.5% (7.4% โ 9.9%), CDNS +0.5% (2.5% โ 3.0%) SELL: NVDA โ1.0% (20.4% โ 19.4%), GOOGL โ0.5% (15.1% โ 14.6%), META โ0.5% (7.2% โ 6.7%), TSMC โ1.0% (2.6% โ 1.6%) Rationale: Reduce concentration in NVDA, begin reducing MSFT underweight, trim GOOGL/META on revision deterioration, trim TSMC on valuation. Fund CDNS as AI-levered software. |
| 2026-03-06 | REBALANCE #2 |
Major rebalance โ benchmark recalculation + 5 position changes โ BENCHMARK CHANGE: Recalculated all benchmark weights using actual S&P 500 GICS sector weights (sourced from Prices.xlsx). Prior combined/blended weights were incorrect. Tech uses S&P 500 Info Tech sector, CS uses S&P 500 Comm Svcs sector. Tech portfolio (13 โ 12 held): BUY: MSFT +4.5% (large add, flipped from biggest UW to +2.5% OW) SELL: ADI all (was in no-man's land at +0.2% active), NVDA โ1.5% New active weights: CDNS +4.1%, ASML +3.4%, AVGO +2.8%, DDOG +2.8%, TSMC +2.5%, MRVL +2.5%, NVDA +2.5%, MSFT +2.5%, LRCX +1.7%, TXN +1.7%, MU +1.4%, AAPL โ0.8% Comm Svcs portfolio (5 โ 7 held): BUY: VZ 5.0% (new, defensive yield anchor), RBLX 0.5% (new, small conviction bet) Funded by light trims across existing CS positions New active weights: TMUS +4.2%, NFLX +3.2%, SPOT +2.0%, GOOGL +2.0%, META +1.6%, VZ +0.5%, RBLX +0.5% Combined view: Uses actual within-sector active weights scaled proportionally to 100% (not index-cap weighted). This gives equal voice to Tech and CS conviction levels. |
Sorted by alpha contribution. Weights drift from inception as stocks outperform/underperform the portfolio.
| Date | Port | BM | Active | Cum Port | Commentary |
|---|
| Date | Port | BM | Active | Cum Port | Commentary |
|---|
All portfolio rebalances since inception. Each rebalance is chain-linked in performance attribution โ returns before and after use the correct weights for each leg.
| Stock | Before | After | ฮ |
|---|---|---|---|
| MSFT | 7.4% | 9.9% | +2.5% |
| CDNS | 2.5% | 3.0% | +0.5% |
| Stock | Before | After | ฮ |
|---|---|---|---|
| GOOGL | 15.1% | 14.6% | โ0.5% |
| META | 7.2% | 6.7% | โ0.5% |
| TSMC | 2.6% | 1.6% | โ1.0% |
| NVDA | 20.4% | 19.4% | โ1.0% |
Principles derived from building and refining this portfolio. Updated February 24, 2026.
Every position must be a real active bet with conviction. Active weights between -1% and +1% are "no man's land" โ not making a real bet either way. Either size up to express conviction or eliminate and redeploy capital to higher-conviction ideas.
Positions with -1% to +1% active weight are not making a real bet. They dilute conviction and use capital that could be deployed more effectively.
Action: Every position must have โฅ1% or โค-1% active weight. If you can't justify that level of conviction, don't own it.
T (+0.3%), VZ (+0.3%), and AMD (+0.2%) were all in no man's land. Instead of small tweaks, we eliminated all three and redeployed capital to high-conviction ideas.
Result: Freed 5.5% capital, concentrated in NVDA (+2.6%) and other conviction bets.
A stock's rolling 1-year win rate (% of periods it beats the index) reveals whether it's a chronic outperformer or underperformer.
VZ (18% win rate) and T (28% win rate) have been chronic underperformers for 7+ years. Combined -215% relative return vs index.
Lesson: A low valuation percentile doesn't fix a chronic loser. VZ at 5th percentile EV/Rev was still a capital destroyer.
Long-term investing allows for thesis bets where historical data may not reflect future potential. But narrative alone isn't enough.
Rule: If betting on narrative against data, size smaller. Position size reflects uncertainty, not just conviction.
DDOG has a 21% win rate (worse than T!) but a compelling AI infrastructure narrative. Initial recommendation was 2.5%.
Revised: 1.5% โ meaningful bet on the thesis, but sized to reflect that data doesn't yet confirm the story.
A "6th percentile" sounds cheap, but it's relative to the stock's own history. DDOG at 10.2x EV/Rev is "6th percentile" only because it traded at 119x during the bubble.
Reality check: 10.2x EV/Rev is more expensive than TSMC (8.7x) and MRVL (7.4x) in absolute terms.
Always check both percentile (vs own history) AND absolute valuation (vs peers/market).
Within a sector, compare stocks directly against each other, not just vs the index. This reveals which is the best way to express a sector view.
Example: TMUS beats VZ in 94% of rolling 1-year periods, beats T in 79%. If you want telecom, TMUS is the only choice.
T and VZ were held as "defensive anchors" but had worse YTD performance than TMUS most years. Defensive thesis was right, but wrong stocks.
Lesson: Validate thesis by comparing within the peer group, not just assuming all stocks in a sector behave the same.
Some stocks have persistent seasonal weakness. VZ/T have brutal April (-6-7%) and July (-6%) patterns. TMUS is much more even.
Forward seasonal (Feb-Jul): TMUS -0.9%, VZ -15.4%, T -16.7%. Timing matters for entry/exit.
Comprehensive analysis should include: valuation percentile, absolute valuation, earnings revisions, win rate, head-to-head vs peers, and seasonal patterns.
Process: Quick screens on valuation โ Deep dive on candidates โ Seasonal timing for execution.
You will constantly receive information that both challenges AND supports your positions. Reacting to each piece creates churn, destroys alpha through transaction costs, and leads to timing errors.
Rule: Change the portfolio as infrequently as possible, and only when you have strong conviction that the thesis has fundamentally changed.
Cross-sector research suggested NFLX fails the "mission-critical" test. Initial reaction: trim NFLX, add to META. But on reflection:
Before making any portfolio change, ask:
Not "is there new information" but "does this fundamentally change why we own it?"
One research note โ market regime shift. Wait for confirmation across multiple sources.
If building the portfolio from scratch today, would I exclude this position entirely?
If the answer to all three isn't clearly "yes" โ don't trade. Monitor, update notes, but hold the position.
Strong data support + compelling thesis + reasonable valuation. Example: NVDA (80% win rate, 13th pctl PE, +45% revisions)
Mixed data + strong thesis OR strong data + uncertain thesis. Example: DDOG (weak data, strong narrative = 1.5%)
Can't articulate why you'd over/underweight. Either hold benchmark or eliminate entirely.
Portfolio benchmark = 75% S&P 500 Tech Index + 25% S&P 500 Comm Services Index.
Derived from S&P 500 GICS sector weights (as of Feb 2026): Info Tech 31.7%, Comm Services 11.1%. Within the combined Tech+CS universe: 31.7/42.8 = 74.1% Tech, 25.9% CS โ 75/25.
Rule: All alpha attribution uses this blended benchmark. Don't use Tech Index alone โ it ignores the CS exposure (GOOGL, META, NFLX, SPOT, TMUS) which is ~33% of the portfolio.
Initial 4-week alpha was calculated using Tech Index only (-2.2%). Blended benchmark was -2.1% โ modest difference in this period, but the CS component moved -1.7% vs Tech -2.2%.
Why it matters: In periods where Tech and CS diverge significantly (e.g. w3: Tech -1.9% vs CS -5.6%), using wrong benchmark materially distorts alpha attribution for Internet names.
The Ern% column is an average of Revenue, EBITDA, EPS, and FCF revisions. A negative blended number can mask healthy operating trends if one component (typically FCF) is distorted by capex guides.
Rule: Never make portfolio decisions based on blended Ern% alone for capex-heavy names. Always decompose into individual components before acting on apparent revision deterioration.
GOOGL blended Ern% of -7.4% appeared alarming. Decomposition revealed: Rev +6.1%, EBITDA +6.8%, EPS +8.5% (all healthy) โ with FCF -75.7% entirely from $175-185B capex guide. Same pattern for META (Rev +10.6%, FCF -52.6%).
Outcome: Without decomposition, might have trimmed GOOGL/META on "deteriorating revisions." Instead, confirmed operating strength supports overweight. MSFT has weakest revenue revision (+4.0%) โ supports underweight despite positive blended Ern%.
| Stock | Action | Rationale |
|---|---|---|
| NVDA | โ 17.8% โ 21.0% | 80% win rate, 13th pctl PE, +45% revisions โ best data in portfolio |
| AMD | Eliminated | 97th pctl valuation, redundant to NVDA, +0.2% active = no conviction |
| VZ | Eliminated | 18% win rate, -215% total relative return, chronic loser |
| T | Eliminated | 87th pctl PE (expensive!), worse than VZ on valuation, 28% win rate |
| TMUS | Keep 3.5% | Only telecom worth owning โ beats VZ 94%, beats T 79% head-to-head |
| DDOG | Add 1.5% | 21% win rate but compelling AI narrative โ sized for uncertainty |