AI Services Economics Model

The Delivery Stack — Who Actually Does the Work

I group public AI-implementation providers into six tiers. Their economics are different. Their customer access is different. Their vulnerability to AI-driven disruption is different. Treating them as a monolith ("consulting stocks") is how investors miss DELL-style mispricings.

Who Wins the Delivery War

If you take McKinsey at face value — 88% of enterprises using AI in some function, but only 6% capturing material EBIT — the gap between adoption and value represents a $500B-$1T implementation services opportunity over five years. But that opportunity is not evenly distributed.

The counter-argument: AI eats its own implementation layer. Cline, Codex-type agents, Claude Code, Microsoft 365 Copilot — these compress consultant billable hours. The data says this is happening (9.2 hours/week saved at one Toronto strategy firm, 12.2% task completion lift, 25.1% speed gains), but firms are redirecting reclaimed hours to utilization and business development. Q1 2026 data shows that firm's utilization jumped from 79% to 94%. Billable hours didn't fall. Margin expanded.

This is the same pattern that played out in software a decade ago: productivity tools were supposed to kill agencies, and instead agencies used them to staff bigger engagements with the same headcount. The economics are stubbornly similar. Don't bet against a fee-per-outcome model replacing billable hours until you see pricing power actually transferring to clients. The data says it isn't — 79% of law firms use AI for delivery; only 6% charge less.

Under-the-Radar Picks

In the infrastructure report, DELL was the hidden gem — #1 AI server ODM at 10x forward P/E despite 19-23% revenue growth. The market was looking at Nvidia and ignoring the OEM that bundled 50%+ margin services on top of every AI server. Here are the services-layer analogs. All are public, all have meaningful AI implementation exposure, all have mispricings I think matter.

Hidden Gems — Summary Table

The Thesis in One Paragraph

The Fortune 500 overwhelmingly wants AI. 88% of enterprises already use it in at least one function, per McKinsey's 2025 State of AI. 92% of executives plan to increase AI spending over the next three years. 78% use GenAI somewhere in the stack. And yet — only 5-6% extract material EBIT impact. Two-thirds of organizations never scale beyond pilots. BCG says 60% of GenAI investments generate no material value. This is not an adoption problem. It is an implementation problem. And the companies that solve implementation — data engineering, systems integration, change management, governance, training — are about to absorb a multi-year wave of spend that dwarfs the model-layer economics everyone's watching.

Why "Pilot Purgatory" Is the Real Business Model

The phrase "pilot purgatory" appears in 2026 McKinsey, BCG, and Gartner reports with striking consistency. It describes the state of an enterprise that has funded dozens of AI pilots, produced PowerPoint decks showing ROI in narrow sandboxes, and then never moved to production. The gap between pilot and production is where the implementation services opportunity lives.

McKinsey's data: 88% of organizations use AI somewhere, but only 1% describe their GenAI strategies as "mature." BCG: less than 30% of CEOs are satisfied with AI returns. Gartner: $1.5T in prior-year global AI spend, with most organizations admitting minimal production-scale impact. These are not numbers that describe a mature market. They describe a market that is pre-integration.

What does "integration" cost? Smartbridge's 2026 enterprise AI implementation guide puts production-ready systems at $500K-$1.5M per project, with organization-wide transformations at $5M+. Roughly 70% of that spend goes to people and process, not technology — which is exactly the services layer. Hidden costs (data prep, integration, skills gaps, governance, ongoing ops) account for 28-42% of first-year spend beyond the sticker price.

Multiply across the Fortune 500. Assume 250 companies are actively implementing AI in 2026 (matching the McKinsey 50% figure for active implementation). Assume average annual implementation spend of $40M per company, weighted across many projects per company. That's $10B in 2026 F500 implementation demand — and that's before the ramp. At 25% growth (below IDC's 38% APAC AI services CAGR and below the $14.1B → $116.8B market CAGR of 26.5%), 2030 demand is $24B. Cumulative five-year demand: ~$80B just from the F500 implementation layer alone. Add mid-market enterprise, federal, and international, and the number is 2-3x that.

The Six-Tier Delivery Stack

Not all services providers are the same. Their economics, their F500 access, their vulnerability to AI-driven disruption — all different. Here's how I group them:

Tier 1 — Big 4 Consulting (premium strategy + delivery)

Accenture, Deloitte, McKinsey, BCG, KPMG, EY, PwC. Highest margins (25-30% operating). Deepest C-suite access. Fastest GenAI bookings ramp. Accenture's GenAI bookings have grown consistently quarter over quarter — the firm discloses the metric precisely because it's the one investors anchor on. Only Accenture is directly investable on US exchanges.

Tier 2 — Legacy IT Services (large-scale outsourcing + AI bolt-on)

IBM Consulting, Capgemini, DXC, Kyndryl, CGI. The value prop is scale and multi-year managed services contracts. IBM Consulting grew only 1% in Q4 2025 — pricing pressure from AI coding tools is real. Kyndryl is down 55% with an SEC probe on free-cash-flow claims. DXC is forecasting ~39% earnings decline over three years. CGI is the outlier — 7.7% revenue growth, 110% book-to-bill, OpenAI Codex partnership. That's why CGI is on the hidden-gems list and the other four are not.

Tier 3 — Indian IT Majors (scaled delivery + offshore wage arbitrage)

TCS, Infosys, Wipro, HCLTech, Cognizant. Where the bulk of F500 implementation hours land. Cognizant guided 4-6.5% constant-currency growth for 2026 with $28.4B in trailing-12-month bookings. Forward P/E ~10.5x. These firms face the biggest AI-productivity headwind — offshore wage arbitrage commoditizes when AI compresses billable hours. But "commoditizes" ≠ "disappears." The volume of work funnels through these firms because no one else has the seat count.

Tier 4 — Premium Engineering (mid-sized specialist delivery)

EPAM, Globant, Endava, Thoughtworks, Perficient (acquired by EQT in October 2024 — no longer public). Premium billing rates, deep engineering moat, but not Big 4 scale. EPAM is the clearest mispricing in the tier — down 35% YTD to $131, forward P/E ~12x, consensus PT $187-246. Endava is in structural decline — revenue down, net loss, stock $4.86 from $35 — avoid. Globant is middle-of-pack with wide PT dispersion.

Tier 5 — Federal / Defense AI (mission-critical government contracts)

Leidos, Booz Allen, CACI, SAIC, KBR, Parsons, ICF. The federal AI procurement cycle is lumpy but well-funded. Leidos trades at 13x forward P/E on $12.25 guided EPS — cheapest defense-AI exposure. Booz Allen trades below $83 vs fair value estimates of $106.90. KBR's $23B backlog is underappreciated as an AI vehicle. This tier's competitive moat is clearance and established contract vehicles — harder to disrupt than commercial consulting.

Tier 6 — Data Ops + BPO Hybrid (the picks-and-shovels underneath)

Innodata, Appen, TaskUs, Concentrix, Mastech Digital, CNXC, Teleperformance. The forgotten layer. If 70% of AI project spend is data + process (per Smartbridge), this tier captures a disproportionate share of the non-advisory budget. Innodata is the standout with its 4-of-top-5 tech-giant customer base and consensus Strong Buy. TaskUs is more speculative — Wedbush $14 PT vs $7 market price, but 38% decline over 6 months and special-dividend refinancing noise create real risk.

Capacity: Who Can Actually Deliver at F500 Scale

Here's the number nobody publishes cleanly: how many F500-qualified AI-implementation FTE-equivalents exist globally? My estimate, triangulating disclosed headcount and AI-specific allocation:

• Accenture: ~750K employees, ~15% on AI work = ~113K
• IBM Consulting: ~160K, ~20% on AI = ~32K
• Deloitte/McKinsey/BCG/Big 4 private: ~1.2M total, ~10% AI = ~120K
• Indian IT majors (TCS, INFY, WIT, HCL, CTSH): ~1.8M combined, ~20% AI = ~360K
• Tier 4 premium eng (EPAM, Globant, Thoughtworks, Endava): ~125K combined, ~30% AI = ~37K
• Tier 5 federal: ~200K combined, ~15% AI = ~30K
• Tier 6 data/BPO: ~450K combined, ~25% AI = ~112K

Rough total: 800K AI-focused services FTEs globally. At average fully-loaded billing rate of ~$150K/year, that's ~$120B in theoretical annual services capacity. But — and this is where the model bites — F500-qualified implementation capacity is only a fraction of that. Big 4 access, clearances, regulated-industry experience, fluency in client enterprise architecture — these filter available capacity down to perhaps 40-50B$ in 2026 F500-serviceable capacity.

If 2026 F500 demand is $10-25B (depending on where you set the parameters), the market is not in acute shortage today. But demand is growing at 25%+ per year; capacity at best 12-15% (offshore wage inflation + productivity uplift net). That math bites in 2027-2028. By 2029 the shortfall under default assumptions exceeds $15-30B annually — which is why the Tier 4/5/6 firms become absorbers of overflow demand, and why the valuation dispersion is about to compress.

Counter-Theses (Addressed, Not Dismissed)

"AI kills consulting by making consultants 10x more productive."

Real gains are happening — 9.2 hours/week reclaimed in one Toronto firm study, 12.2% task completion uplift, 25.1% speed gain. But reclaimed hours are redirecting to utilization, not dropping billable totals. Q1 2026 utilization at that firm rose 79% → 94%. Billings went up. Margin expanded. Only 6% of law firms charge less despite 79% using AI for delivery. Pricing power is still with the supplier. This counter-thesis will eventually be right; the question is when, and the data says not yet.

"Hyperscaler pro-services arms (AWS PS, Azure, GCP) absorb the spend."

Hyperscaler pro-services exist but are not broken out as independent reporting segments. AWS/Azure/GCP are primarily IaaS plays; their pro-services revenue is 5-15% of total cloud revenue. They're a real competitor for greenfield implementations, but they cannot absorb F500-scale change management or regulated-industry compliance work. They are a secular share-taker, not a market replacer.

"F500 builds internal AI Centers of Excellence and cuts external spend."

Evidence says the winning F500 pattern is hybrid hub-and-spoke, not pure build. Internal CoE for standards and IP; external partners for spoke execution. ServiceNow's data shows 85% of F500 use Now Assist via third-party partner implementation. The CoE model creates more services demand, not less — someone has to stand up the CoE, integrate it with spokes, and train the talent pipeline. 77% of enterprises still look externally for AI consulting per Versich.

"Open-source + low-code eat the low end."

Real at the SMB tier. Not real at the F500 tier, where regulatory overhead, audit trail, data lineage, and integration complexity mean the low-code layer can't carry the weight. The more interesting version of this thesis: low-code commoditizes the middle, which pushes premium firms toward strategy/governance/audit and low-cost offshore toward the remaining scale work. It reshapes the tiers, doesn't flatten them.

What Could Break This Thesis

Stock Picks — Ratings + Price Targets

All data as of April 21, 2026. "My PT" reflects my model output against consensus; "rationale" shows what moves the number. Ratings: STRONG BUY, BUY, HOLD, SELL, AVOID.

The Case for Each Rating

INOD — Strong Buy

Only pure-play AI data-preparation publicly traded. Customer base (4 of top 5 tech) is a moat. Consensus Strong Buy with PTs at 95-113% upside. Small-cap, so size position accordingly. Earnings May 7 is the next catalyst.

ACN — Buy

The quality default in the space. 3.4% dividend. Moderate Buy consensus. $274.50 average PT is 40% upside. The trade is simple: if the F500 implementation thesis plays out at all, ACN is the largest beneficiary. Downside risk is contained by dividend + book-to-bill.

GIB / CGI — Buy

Under-covered, cheaper than all US peers at ~14x trailing P/E. OpenAI Codex partnership is a category-disrupting tailwind the market hasn't priced. DCF fair value CAD $173 vs CAD $105 market = 65% undervalued. April 29 earnings catalyst.

EPAM — Buy

The market priced this like EPAM is structurally broken. It isn't. Premium engineering moat intact. Investor Day trajectory to $6.5B revenue by 2028. Forward P/E ~12x vs normalized peers at 18-22x. Multiple expansion alone gets 50%+ upside before estimate revisions.

CTSH — Buy

Cheapest Tier 3 by forward P/E (~10.5x). Record $28.4B TTM bookings. OpenAI + Palantir + DAMAC partnerships. Visa risk is real but priced in. Consensus $88.41 PT = 47% upside. The bulk-of-F500-hours trade.

LDOS, BAH, ASGN, KBR — Buy (Federal Basket)

Federal IT spend is the most predictable IT spend. AI-specific procurement is ramping via vehicles like SHIELD. LDOS cheapest by forward P/E. BAH trades below fair value. ASGN has a ticker-change dislocation catalyst. KBR is the pivot story. Buy as a basket if you want federal AI exposure without single-name risk.

IBM — Hold

GenAI bookings of $12.5B cumulative is real. But consulting segment grew only 1% in Q4, and the stock is at 22x trailing with a $301 PT — only 19% upside. The software/mainframe thesis is stronger than the consulting thesis here. Hold if you already own it for the dividend + software. Don't add new money.

GLOB — Hold

Wide PT dispersion ($61 to $255) tells you analysts don't agree what this company is. Mkt cap down 26% YoY. Until Q1 clarifies growth direction, the risk-reward is undifferentiated. Wedbush Outperform with $61 PT would be actionable if the stock approaches $40.

PLTR — Hold / Trim

Genuinely great business. 61% YoY revenue growth guide, $8.6B RDV backlog, 115%+ US commercial growth. But forward P/E of 117x bakes in near-perfection for 3+ years. Morgan Stanley's $205 PT already assumes the best-case. If you own it, trim into strength. If you don't own it, wait for a 30%+ drawdown before entering.

DAVA — Sell

Revenue down, £15M H1 loss, stock $4.86 from $35 in 52 weeks. This is not an overshoot, it's a structural break. Avoid catching the falling knife.

DXC, KD — Avoid

DXC forecasting ~39% earnings decline. Kyndryl hit with SEC probe on cash-flow claims, -55% to $10.59. Both are managed-services legacy plays losing share to AI-native specialists. No thesis justifies the risk.

C3.ai — Sell

36% negative guidance revision. Insider selling ($4.5M+ in early April). Down 32-36% YTD. Product-market fit for enterprise AI software has clearly broken. This is the cautionary tale of AI-branded companies that aren't services and aren't infrastructure.

Portfolio Construction

How I Built This Model

The interactive model on the first tab is a Fermi-style estimation of F500 AI implementation services demand versus delivery capacity. The goal is not precision — the goal is orders of magnitude and the direction of the mismatch.

Demand Side

I anchor on three signals:

• McKinsey 2026 State of AI: 88% of organizations using AI, 78% using GenAI, but only 1% describe strategies as "mature." The gap between usage and maturity is the implementation demand driver.
• Enterprise AI project costs: Smartbridge/industry consensus puts production-ready AI projects at $500K-$1.5M, organization-wide transformations at $5M+. I use $40M/year as a weighted F500 average (multiple projects per company).
• AI consulting TAM: $14.1B in 2026 → $116.8B by 2035 at 26.5% CAGR (Smartbridge 2026). IDC projects 38% CAGR for APAC AI/GenAI spending through 2029. I use 25% as a conservative baseline growth rate.

Model formula: F500 demand(year) = 500 × active% × spend/co × (1+growth)^year × regulated_premium × pilot_to_prod_rate_adj

Supply Side

I anchor on three signals:

• Disclosed headcount × AI allocation: Triangulating headcount from 10-Ks and AI allocation from earnings commentary gives ~800K global AI-services FTEs, ~$120B theoretical annual capacity at $150K/FTE fully-loaded.
• F500-qualified capacity haircut: Not all 800K FTEs can deliver F500-scale work. Apply 35-50% haircut for Big 4 access / regulated industry fluency / enterprise architecture experience.
• Capacity growth rate: Historical IT services headcount growth of 8-12% plus AI productivity uplift of 10-20% gives a net 14% capacity-effective growth rate baseline.

Model formula: capacity(year) = base_capacity × (1+cap_growth)^year × (1+productivity_uplift)

What the Model Does Not Attempt

• Individual stock forecasts — those are in the Stock Picks tab and use consensus estimates plus my adjustments, not the model.
• Mid-market enterprise demand — model is F500-only. Add 2-3x for mid-market.
• International demand outside US — assumes US/global-MNC scope. Rest-of-world demand is probably another 1.5x multiplier.
• Subcontracted / reseller economics — where a Tier 1 firm subcontracts 40% of a project to Tier 3, this model counts the revenue at the Tier 1 layer and ignores the Tier 3 share. This matters for individual company analysis but not for aggregate demand/capacity framing.

Key Sensitivities

Sources

Assembled from 50 Perplexity queries run April 21, 2026. Stock prices, market caps, and analyst price targets reflect that date. Only verified-from-research URLs included — fabricated links removed.

Market Sizing & Adoption

• Smartbridge — AI consulting $14.1B (2026) → $116.8B (2035)
• Fullstack Labs — 80% of companies see no GenAI results
• Plaster Group — AI business transformation survey
• Future of Marketing — The AI Innovator's Dilemma
• McKinsey — Superagency in the Workplace
• CMARIX — AI agent statistics, Gartner 40% by 2026
• SSNTPL — Enterprise AI Implementation 2026 Guide
• CloudZero — AI implementation cost breakdown
• Pharos — State of AI Development Costs 2026
• ISG Index 1Q26 Recap
• ISG AI Index — AI impact on services
• IDC — European AI spending to $290B by 2029
• IDC — APAC ICT $647B in 2026
• IDC Directions 2026 — Agentic Shift

Accenture / Big 4

• MarketBeat — Accenture stock metrics April 2026
• LongForecast — ACN April 2026
• Kraken — ACN stock data

IBM

• TIKR — IBM Confluent deal analysis
• Investing.com — IBM AI customer experience with Adobe
• GuruFocus — Jefferies IBM PT
• IBM Newsroom — Consulting's AI Moment

Cognizant / Indian IT

• Tickeron — CTSH metrics
• TIKR — Cognizant 2026 thesis
• SimplyWallSt — CTSH analyst revisions
• GuruFocus — CTSH OpenAI partnership
• StockInvest — CTSH recent prices
• Welthwest — Indian IT AI playbook
• Business Today — Indian IT stocks analysis
• AI with Jason — The Great AI Disruption

Premium Engineering (EPAM / Globant / Endava)

• Quartr — EPAM Investor Day 2026
• TIKR — EPAM 2026 thesis
• SimplyWallSt — EPAM AI security pivot
• MarketScreener — Jefferies EPAM PT
• MarketBeat — Globant consensus
• CompaniesMarketCap — Globant mkt cap
• StockAnalysis — Endava metrics

Federal / Defense AI

• StockStory — Leidos forward P/E
• SimplyWallSt — BAH valuation
• MarketBeat — BAH Wells Fargo upgrade
• BusinessWire — BAH earnings May 22
• StockTitan — BAH Vellox AI
• MatrixBCG — KBR business model
• StockTitan — KBR Applied Computing investment
• GuruFocus — KBR Tagup partnership
• SimplyWallSt — ASGN federal IT recovery
• MarketChameleon — ASGN overview

Data Ops / BPO Hybrids

• StockAnalysis — Innodata metrics
• GuruFocus — INOD valuation
• CB Insights — Innodata profile
• Zacks — INOD Q&A
• MarketBeat — MHH insider buying
• InsiderTrades — MHH alerts
• Morningstar — TASK quote
• GuruFocus — Wedbush TASK PT
• Investing.com — TASK insider sales

CGI Inc (GIB)

• Investing.com — CGI OpenAI Codex partnership
• CGI — DigiOps AI platform award
• SimplyWallSt — CGI analyst PTs
• MarketBeat — GIB chart
• MarketBeat — CGI earnings April 29

Legacy IT Services (DXC, Kyndryl, Capgemini)

• Sahm Capital — Kyndryl data sovereignty
• Intellectia — Kyndryl SEC probe
• OneIO — Top GSIs ranking

Palantir / C3.ai

• TipRanks — PLTR drawdown analysis
• Investing.com — PLTR premium valuation
• TIKR — PLTR 2026 outlook
• Techi — PLTR forward P/E
• Wedbush — Why C3.ai shares sliding
• InsiderMonkey — C3.ai insider sales

Counter-Theses / AI Productivity

• Fusion Computing — AI productivity in professional services
• Santiago Company — New Economics of Professional Services
• HBR — What's Stopping the 4-Day Workweek
• McKinsey — State of AI Trust 2026
• BCG — AI will reshape more jobs than it replaces
• Lines n Circles — AI Center of Excellence 2026
• Symphonize — Build vs Buy vs Managed
• Grant Thornton — 2026 AI Impact Survey
• Versich — AI consulting for US enterprise

ServiceNow / Partner Ecosystems

• SteadyCompounding — ServiceNow deep dive
• Jade Global — ServiceNow Knowledge 2026
• EY — Snowflake/Databricks partner awards