1. Why This Sector Exists
Information Technology sells productivity in a bottle. Companies pay because one software seat or one GPU replaces hours of labor — the ROI is measurable on a CFO's spreadsheet. Customers keep paying because switching costs (data, training, integrations) compound. Portfolios need IT because it's the only sector where deflationary cost curves (Moore's Law) coexist with pricing power.
2. What's Happening Right Now
What happened: Per Bloomberg's June 18 close, Information Technology gained 2.68% on the day, led by Semiconductors +4.85% while Software & Services lagged at -0.74%. The S&P 500 IT index sat near 6,680 in late May, grinding to fresh highs into June. The tape was supported by an interim US-Iran peace deal that compressed risk premia — though talks hit a snag late week.
Why it happened: Three forces collided. (1) Continued strong hyperscaler AI infrastructure spend, with semis playing the starring role. (2) Rising memory prices are lifting device ASPs but slowing replacement demand to 6.1% growth — bullish for memory names, bearish for PC OEMs. (3) Falling geopolitical risk premium re-rated long-duration tech.
What it sets up: A bifurcated tape into Q2 earnings — semis/memory beat-and-raise, software guidance scrutiny on AI monetization.
3. How the Money Works
Revenue is either per-seat subscription (software) or per-unit shipped (hardware/chips). Stickiness comes from data gravity and workflow embedding. The two cost lines that matter: R&D (you must outspend rivals or die) and COGS — wafer cost for chips, cloud hosting for SaaS. Scale is brutally favorable: Nvidia ships one more H200 at near-zero marginal design cost. Great businesses earn 80%+ gross margins (Microsoft); average ones earn 40% (Dell). Analogy: software is a toll bridge — build once, collect forever.
4. The 4 Macro Drivers
Driver 1: AI Capex Cycle (Hyperscaler Spend)
Mechanism: Hyperscaler capex ($350B+ run-rate in 2026) flows directly into Nvidia, Broadcom, TSMC, and memory. It enters as demand, not multiples.
Now: Strong AI infrastructure spending continues, with semis the primary beneficiary. Nvidia's Vera Rubin superchip should enter production in 2026. 2nd-order effect: Juniors watch Nvidia. Pros watch the power grid — substation transformers, gas turbines, cooling. The bottleneck moved from chips to electrons.
Threshold: If any top-4 hyperscaler cuts FY27 capex guide by >10%, sell the picks-and-shovels basket immediately.
Driver 2: Long-End Rates & Discount Rates
Mechanism: Tech cash flows are back-end loaded. Higher 10Y → higher discount rate → terminal value compression → multiple de-rating, especially in unprofitable software.
Now: Iran de-escalation pulled the term premium lower, helping long-duration software bid. Memory inflation is the new wildcard for headline CPI.
2nd-order effect: When rates fall, juniors buy mega-cap quality. The real trade is profitless SaaS — highest duration, highest beta to rate cuts, most short interest to squeeze.
Threshold: 10Y back above 4.75% kills the software rally; below 4.00% triggers a junk-tech melt-up.
Driver 3: Enterprise IT Budget Growth
Mechanism: Enterprise IT spend funds the second leg — software seats, cybersecurity, consulting. Enters as revenue growth for Microsoft, ServiceNow, Accenture.
Now: Total software spending remains above $1.4 trillion, with GenAI model spending growing 80.8%. Overall Gartner forecasts worldwide IT spending up 10.8% in 2026 to $6.15 trillion. 2nd-order effect: Big IT budgets ≠ big software revenue. CIOs are redirecting dollars from headcount and legacy SaaS into GPUs and tokens. Legacy SaaS gets squeezed.
Threshold: Watch Accenture bookings — first canary if enterprise digital projects pause.
Driver 4: US-China Tech Decoupling & Tariffs
Mechanism: Export controls cap addressable market for advanced chips; tariffs raise hardware COGS. Enters as both revenue ceiling and margin compression.
Now: After the "tariff turmoil" of March-April 2025, the risk premium normalized, but China revenue at Nvidia/AMD remains structurally lower. 2nd-order effect: Decoupling forces duplicate fab build-out — TSMC Arizona, Samsung Texas, Intel Ohio. Capital equipment names (ASML, AMAT, LRCX) get a multi-year tailwind even if end-demand softens.
Threshold: Any new export control extending to HBM3e or sub-7nm equipment is a sell signal for ASML and KLAC.
5. Sector Map
| Sub-Industry | What It Does | Key Driver | Main Risk |
|---|---|---|---|
| Semiconductors | Designs/makes AI chips | Hyperscaler capex | China export curbs |
| Semi Cap Equipment | Sells fab machinery | Fab construction cycle | Capex digestion |
| Systems Software | OS, cloud, security platforms | Enterprise IT budgets | AI disintermediation |
| Application Software | Vertical SaaS workflows | Seat growth, AI attach | Copilot cannibalization |
| IT Services | Consulting, integration | Digital transformation | GenAI labor arbitrage collapse |
6. Company Case Studies
Case Study 1: Nvidia (NVDA) — Still the only end-to-end AI stack at scale
Business: Sells data-center GPUs (Hopper, Blackwell, soon Rubin) bundled with CUDA software and NVLink networking. Revenue is per-unit but ASPs ($30K-40K per GPU) and 75%+ gross margins make it look like software. Key cost: TSMC wafer allocation and HBM memory from SK Hynix/Micron. At scale, incremental GPU = incremental dollar of gross profit.
Moat: Nvidia is far ahead because no other company offers the combination of integrated hardware, software and algorithms, especially since the Blackwell chipset launched in 2024 with significant performance and energy efficiency gains. CUDA lock-in is widening, not eroding.
Macro Linkage: Driver 1 (AI capex) is the entire thesis. Every dollar of hyperscaler capex flows ~25¢ to Nvidia. Driver 4 caps the China TAM — but redirected demand from sovereign AI (Saudi, UAE, EU) more than offsets.
Watch: (1) Data center revenue QoQ growth — currently still expanding double digits; deceleration below 8% QoQ flags peak capex. (2) Vera Rubin production ramp timing in 2026 — any slip pushes earnings right.
Risk: Hyperscaler in-house silicon (Google TPU, Amazon Trainium, Microsoft Maia) reaching parity for inference. Early warning: any hyperscaler disclosing >30% internal-chip mix.
Valuation: Forward P/E ~32x on FY27 EPS. Fair, not cheap. PEG <1.0 still attractive if growth holds.
Case Study 2: Microsoft (MSFT) — The toll booth on enterprise AI
Business: Three engines — Azure (cloud infra), M365 + Copilot (productivity SaaS), and gaming. Revenue per-seat and consumption-based. Key costs: data center capex ($80B+ run-rate) and OpenAI revenue share. Scale advantage: every Azure region built lowers unit cost for the next AI workload.
Moat: Distribution. 400M+ commercial seats globally means Copilot deployment is a price decision, not a sales decision. Switching from Office to Google Workspace requires retraining the entire enterprise — unchanged in 20 years.
Macro Linkage: Driver 3 (enterprise IT budgets) hits hardest. Copilot at $30/seat/month is the largest single line-item add in CIO history. Driver 2 matters at the multiple — Microsoft trades like a long-duration bond.
Watch: (1) Azure AI revenue contribution — currently the swing factor for the whole stock. (2) Copilot attach rate within E5 customers — leading indicator for FY27 ARPU.
Risk: GenAI commoditization. If Anthropic/Google reach Copilot parity at half the price, ARPU resets. Early warning: enterprise win-rate disclosure softening.
Valuation: ~31x forward P/E, ~22x EV/EBITDA. Fair. Premium justified by AI optionality but no longer a bargain.
Case Study 3: ASML (ASML) — The only EUV game in town
Business: Builds the lithography machines (DUV, EUV, High-NA EUV) that print every leading-edge chip. Revenue per-tool runs $200M-$380M for High-NA. Costs: 5,000-supplier ecosystem and 10-year R&D lead times. Gross margin ~52% and rising as service revenue grows.
Moat: Effective monopoly on EUV. Nikon and Canon can't catch up — the physics took ASML 20 years and $20B to crack. Service contracts on installed base create recurring revenue tail lasting 15+ years per tool.
Macro Linkage: Driver 4 (decoupling) is the underrated tailwind — every new fab (TSMC AZ, Samsung TX, Intel OH, Rapidus Japan) buys the same ASML tools. Driver 1 indirect: AI chip demand → more leading-edge wafers → more EUV.
Watch: (1) Bookings ($/quarter) — choppy quarter-to-quarter; 4Q trailing is the signal. (2) China revenue mix — was 45%, now structurally ~25% under export controls.
Risk: Order push-outs from Intel or Samsung if their foundry strategies stumble. Early warning: any customer-specific delivery delay commentary on the call.
Valuation: ~28x forward P/E. Fair-to-cheap given monopoly economics and 2027-28 High-NA ramp visibility.
7. How to Value These Companies
Use EV/Sales × Rule of 40 for SaaS (growth + margin should clear 40%; below that, multiples halve). Use forward P/E adjusted for cycle position for semis — never use trailing, you'll buy the top. Use EV/EBITDA for IT services. The common junior mistake: anchoring to historical multiples without adjusting for where you are in the capex cycle. A "cheap" 18x semi at peak earnings is a 30x semi at trough — that's the trap.
8. KPIs That Actually Matter
| KPI | What It Signals | Why It Beats EPS | Benchmark |
|---|---|---|---|
| Net Revenue Retention | Pricing + expansion power | EPS hides churn | >120% elite |
| Data center revenue QoQ | AI capex pulse | Real-time vs lagging | >10% healthy |
| RPO / Backlog | Future revenue locked | Smooths quarterly noise | Rising YoY |
| Free Cash Flow margin | True earnings quality | EPS gameable via SBC | >25% software |
| Capex intensity | Reinvestment discipline | EPS ignores capital | <15% software |
| Bookings book-to-bill | Forward demand | EPS is rearview | >1.0x semis |
9. Risk Map
Risk 1: AI Capex Digestion Air Pocket
Hyperscalers front-loaded GPU buys in 2024-26. If 2027 capex grows "only" 5% instead of 30%, semis de-rate from 30x to 18x overnight. Transmission: orders pause → channel inventory builds → ASPs crack → margins follow. Precedent: 2001-02 telecom capex bust crushed Cisco from $80 to $8. Early warning: any hyperscaler CFO using the word "optimize" or "efficiency" on capex in earnings prepared remarks.
Risk 2: Memory Price Spike Choking Devices
Rising memory prices are increasing average selling prices and discouraging device replacements. Transmission: DRAM/NAND cost up → PC/phone ASPs up → unit demand down → Dell, HP, Apple Services growth slows. Precedent: 2017-18 memory cycle masked weakness in end-device demand until 2019 collapse. Early warning: PC unit shipment data from IDC turning negative YoY while ASPs still rise.
Risk 3: Software Disintermediation by AI Agents
If GenAI agents can replicate Salesforce/ServiceNow workflows, seat-based SaaS shrinks. Transmission: customer cuts seats → NRR falls below 100% → multiple compresses from 12x to 6x sales. Precedent: 2012-15 on-prem to cloud transition halved Oracle's multiple. Early warning: application software NRR slipping below 110% across the cohort — already happening at the long tail.
Risk 4: Export Control Escalation to Equipment
If US extends curbs to ASML DUV or AMAT etch tools for China, $15-20B of semi-cap revenue vaporizes. Transmission: revenue cut → fixed-cost deleverage → margin shock → multiple compression. Precedent: October 2022 controls knocked KLAC down 18% in two weeks. Early warning: BIS staffing changes, Dutch government press leaks, Commerce Department speeches.
10. Cycle Playbook
| Phase | Sector Behaviour | Why | What to Own |
|---|---|---|---|
| Early Expansion | Outperforms sharply | Rate cuts + EPS revisions up | Semis, profitless SaaS |
| Mid Cycle | Steady gains | Earnings drive returns | Quality compounders (MSFT) |
| Late Cycle | Volatile, narrow leadership | Multiple-driven gains | Mega-cap defensives |
| Recession | Underperforms early, leads out | Capex/IT budgets cut | Cash-rich balance sheets |
| Recovery | Strong outperformance | Operating leverage on rebound | Semis, semi cap equipment |
Now: Late-cycle with AI-driven narrow leadership — the top 7 names doing the heavy lifting while broad software lags. Trim the crowded long, keep quality.
11. Structural Themes
Theme 1: Power as the Binding Constraint
The bottleneck has shifted from GPUs to gigawatts. Success increasingly depends on system-level integration aligning chips, servers, racks and data center infrastructure. Why now: AI training clusters need 100MW+ campuses; grid interconnect queues run 4-7 years. Winners: Vertiv, Eaton, Schneider, gas turbine OEMs, nuclear SMRs. Losers: Hyperscalers without secured power PPAs. Position: Own the power-adjacent names before consensus stops calling them "industrials" and starts calling them "AI infrastructure."
Theme 2: Sovereign AI as a New End Market
Nations (Saudi, UAE, India, France, Japan) are funding domestic AI compute as strategic infrastructure — a new buyer class beyond the US hyperscalers. Why now: China decoupling shows compute access = sovereignty. Winners: Nvidia (G42, Humain deals), Dell/Supermicro for systems, oil-state-funded data center developers. Losers: Pure-play US enterprise SaaS — sovereign budgets bypass them. Position: Track Nvidia's geographic revenue disclosure; sovereign now >15% and rising fast, structurally less cyclical than US hyperscaler spend.
12. Portfolio Reference
| Factor | Value |
|---|---|
| S&P 500 weight | ~32% |
| Typical dividend yield | ~0.6% |
| Beta vs S&P 500 | ~1.20 |
| Overweight when | Rates falling, ISM rising |
| Underweight when | Capex peaks, rates rising |
| ETF | Focus | Expense Ratio |
|---|---|---|
| XLK | S&P 500 IT, cap-weighted | 0.08% |
| SOXX | Semiconductors | 0.35% |
| IGV | Application software | 0.41% |
13. Three Questions You Should Be Able to Answer
Q1: Why do software companies trade at higher multiples than semiconductor companies despite often slower growth?
A: Two reasons most juniors miss. First, revenue durability: a SaaS customer pays every month until they actively cancel; a chip customer must re-order each cycle. Second, incremental margins: each new SaaS seat carries 90%+ gross margin; each new chip carries 60-75% and requires fab capacity. Higher persistence + higher incremental returns on capital = higher justified multiple. Example: Microsoft trades at 31x because its cash flows look like a 15-year bond. Nvidia at 32x is the exception — it's pricing in semi-software hybrid economics.
Q2: How does a 50bp move in the 10-year Treasury yield transmit into IT sector returns, beyond the obvious discount-rate effect?
A: The obvious channel: higher rates → higher WACC → DCF terminal value compresses → multiples fall. The missed second-order channel: higher rates → enterprise CFOs delay digital transformation projects → IT services bookings slow (Accenture, Infosys) → software seat growth decelerates 2-3 quarters later → consensus revenue estimates cut → second multiple compression. Then a third effect: venture funding dries up → startup customers of AWS/Azure shrink → cloud revenue growth slows. The bond market move takes 9 months to fully express in tech earnings.
Q3: Bull vs bear case for IT given today's macro?
A: Bull: AI capex still accelerating, IT spending growing 10.8% to $6.15T, Iran de-escalation compressing risk premium, rate-cut path intact, sovereign AI is a brand-new TAM. Multiples justified by FCF growth. Bear: Sector concentration at record highs, hyperscaler capex mathematically must decelerate, memory inflation hitting devices, software NRR slipping, China decoupling capping semis. What flips it: A single hyperscaler guiding capex flat YoY. That's the domino. Until then, stay overweight semis and quality software, underweight legacy hardware and unprofitable SaaS.
Research via live web search | Saturday, June 20, 2026 | GICS Rotation Series