A Teraflow perspective on McKinsey’s Quarterly report, May 2026
TL;DR: Nearly 90% of organisations now use AI in at least one business function, and most are deploying the same large language models.
A May 2026 McKinsey Quarterly article by Dago Diedrich, Evan Williams, Tanguy Catlin, and Tim Fountaine argues that competitive advantage comes from building moats competitors cannot copy. The article identifies six strategic moats and three capability moats. The harder question for enterprise leaders is operational: what infrastructure, data architecture, and operating-model decisions actually translate AI investment into durable advantage.
This piece unpacks the framework and examines what it takes to execute.
What is an AI competitive moat?
A competitive moat is a hard-to-replicate set of assets, capabilities, or relationships that allows a business to sustain superior returns over time. In the AI context, a moat is what remains valuable after the underlying model has been commoditised.
If a competitor can license the same model, access the same cloud, and hire from the same talent pool, the moat lives in everything around the model: proprietary data, integrated workflows, scaled infrastructure, embedded customer relationships, regulatory permission, and trusted brand.
The McKinsey argument is direct. When access to models is universal, advantage shifts to whatever competitors cannot replicate.
Why same-model adoption does not create advantage
This is not a new pattern.
McKinsey points to the digital transformation wave of 2018 to 2022, when banks raced to deploy mobile apps. Mobile-app adoption increased across the industry, but the leaders did not extend their advantage over laggards on app deployment alone. What widened the gap was integrating digital and AI across the full customer journey to reduce friction, lift online sales, and drive total shareholder return outperformance.
The lesson holds for AI: apps and tools can be copied. Value comes from advantages that compound and resist replication.
The six strategic moats
Strategic moats are areas where the business can build a structural advantage that is difficult for competitors to match.
1. Economies of scale: infrastructure that drives marginal cost to near zero
AI’s most immediate economic effect is to collapse the marginal cost of cognitive work. McKinsey notes that a customer service interaction that cost $15 in labour can now cost a fraction of that in compute, and is poised to keep dropping.
The moat is not access to LLMs. New entrants have that too. The moat is the infrastructure that turns cognitive work into a repeatable, governed, low-incremental-cost system: data pipelines, fine-tuned models, integrated workflows, governance layers.
McKinsey points to Resolution Life, a US and Australian life insurer, which uses an AI platform to triage incoming claims in 15 seconds rather than weeks, and to introduce additional insurance products at a fraction of historical cost. The same platform absorbs new books of business acquired through M&A through fixed infrastructure.
Teraflow lens: This is the platform layer most enterprises underestimate. Across clients, Teraflow helped scale a foundation supporting 2,900+ data pipelines, 3.6 PB of data, 150+ ML models, and 37M digital users. The unit economics only work when the platform is treated as core infrastructure, not a project deliverable.
2. Privileged data: treating data like an asset class
Privileged data becomes a moat when AI models trained on it deliver products and services competitors cannot match. The most valuable data sets are cumulative, protected, and feed into closed-loop systems where every interaction generates more labelled behavioural and outcome data.
McKinsey cites Amazon’s commerce data flywheel: search behaviour, product views, purchases, fulfilment, advertising response. The economic value shows up in Amazon’s advertising business, which reached $68 billion in revenue in 2025.
The CEO question McKinsey poses is sharp: what unique data could you capture, label, or generate (including unstructured data you couldn’t use before) that would compound model performance faster than rivals can match?
Teraflow lens: Data as an asset class requires three things most enterprises lack: a defined data-product taxonomy, instrumented capture at the workflow level, and a governance model that survives leadership change. Without these, “data strategy” defaults to consolidation projects that never pay back.
3. Embeddedness: making switching expensive
AI solutions move from convenience to necessity when they are embedded in core work. McKinsey identifies three compounding layers of embeddedness:
- Capabilities integrated into core systems (CRM, ERP, productivity suites, industry platforms), making workflow migration expensive
- Systems that learn from proprietary data through feedback loops, becoming more valuable over time
- Employees who rely on the solution and resist change, creating organisational switching costs
McKinsey’s example is Microsoft Dragon Copilot, deployed across more than 150 hospitals on Epic EHR software. It has driven a roughly 50% reduction in documentation time and material reductions in clinician burnout.
The flip side matters for buyers. Every workflow handed to an embedded AI vendor is a bet on that vendor’s roadmap, pricing, and continued existence. Negotiate data rights and portability before deployment, not after.
4. Network effects: AI as the network architect
AI reshapes network dynamics in two ways. First, it reduces the cold-start problem by generating initial supply (listings, content, descriptions) and personalising recommendations from day one. Second, it strengthens network effects by ranking, recommending, verifying, and filtering as the network grows.
McKinsey points to TikTok’s recommendation engine, and forecasts that agentic commerce could orchestrate up to $1 trillion in US retail by 2030. The strategic question shifts: as agents mediate transactions, will your business own the agent layer, or be selected by it?
5. Business model disruption: shifting who owns the customer
AI accelerates business model disruption along two vectors.
The customer relationship. In brokered markets (commercial insurance where third-party distributors account for more than 85% of sales, life insurance at more than 50%, Australian mortgages at about 75%), AI agents can replicate broker functions and sell direct. McKinsey points to NEXT Insurance, where more than 600,000 entrepreneurs buy coverage in about ten minutes without a human broker. About 60% of small business owners now buy insurance fully online.
Outcome-based pricing. AI makes real-time outcome measurement feasible, which in turn makes outcome pricing viable. McKinsey cites Michelin’s EFFIFUEL, which guarantees fuel savings over multi-year commitments tracked via telematics. Time-and-materials businesses face internal resistance to cannibalising the legacy model, and that is precisely the opening AI-first competitors exploit.
Teraflow lens: At Comair (Kulula), applying AI to dynamic pricing and operational optimisation contributed to a 98% aircraft fill rate across more than 4M passengers. Outcome-based pricing only works when the underlying data and decision systems can prove the outcome in production.
6. Constrained assets: where AI meets the physical world
As AI drives down the marginal cost of digital intelligence, advantage shifts toward physical assets competitors cannot quickly replicate: mines, ports, power generation, data centre sites, distribution networks, installed equipment bases.
McKinsey points to John Deere’s See & Spray technology, which uses computer vision on field equipment to identify weeds and apply herbicide selectively. The moat is not the model. It is the installed equipment base, agronomic data, dealer network, and years of operational learning that a software-only competitor cannot replicate.
The three capability moats
Capability moats are organisational strengths that let a company repeatedly turn AI into advantage.
1. Velocity: faster learning and deployment
McKinsey research shows companies in the top quartile of software development velocity achieve four to five times faster revenue growth and 60% higher total shareholder returns than bottom-quartile peers. Agentic AI is shifting development cycles from two weeks to 24 hours.
Velocity requires more than tooling. It requires what McKinsey calls a “rewired” operating model: small cross-functional teams, flexible platforms, data embedded in the business, reusable components. Rewired companies typically improve EBITDA by 10 to 30%, with an average of 20%.
McKinsey cites DBS Bank, which over a decade reduced AI solution development and deployment time from 12 to 18 months down to two to three months.
Teraflow lens: Velocity is where Teraflow’s FloJo framework fits. Reducing learning cycle time from quarters to weeks is the single highest-leverage move most enterprises can make. It changes the experimentation budget, the talent profile, and the board’s appetite for risk.
2. Regulation and compliance: built into the stack
As AI expands into sensitive data and automated decisions, regulatory scrutiny is intensifying. The EU AI Act sets requirements on copyright, security, and transparency for AI models.
The moat develops when compliance is engineered into the solution: audit trails, explainability, data lineage tracking, bias monitoring, human-in-the-loop controls. Attackers can exploit regulatory grey zones temporarily. When enforcement catches up, advantage shifts to firms that built compliance infrastructure ahead of time.
3. Trust: the gatekeeper to adoption
In finance, healthcare, and identity, trust functions as a strategic moat because it gates customer willingness to share data and accept automated decisions. McKinsey notes that more than half of consumers now use generative AI tools and want their banks to offer them, and nearly all say they would switch providers if their bank fell behind.
Global surveys indicate only about 30% of people embrace AI, while 35% reject it. The companies that close that gap embed responsible AI into how systems are designed, deployed, and monitored.
McKinsey highlights JPMorgan Chase, ranked first on the Evident AI Banking Index for four consecutive years, and one of the few banks publicly reporting realised AI returns approaching $2 billion.
The execution gap: why most enterprises stall
The framework is clear. The execution is hard. Across banking, insurance, airlines, telecoms, and energy, four patterns appear consistently:
Moats confused with use cases. “We have an AI moat in claims” usually means there is one model in production. A moat is the platform, data, workflow, and governance that allow the next ten claims models to ship in weeks, not quarters.
Strategy decoupled from infrastructure. Boards approve AI strategies that depend on data products, MLOps platforms, and integration capability the organisation has not built. The strategy implies a moat; the infrastructure prevents it.
Velocity measured at the wrong layer. Time-to-prototype is fast. Time-to-production is slow. Time-to-second-deployment of the same pattern is often slower than the first. Real velocity is reuse.
Compliance bolted on, not designed in. Audit, explainability, and lineage are treated as legal review rather than engineering requirements. The cost is paid in slow deployment and a constrained set of viable use cases.
How to start building moats
McKinsey closes with three actions for boards and CEOs. Practical experience sharpens them.
1. Align on one to three moats and make trade-offs explicit. Few firms build all nine. The leaders pick the moats that match their structural position, then commit budget, talent, and executive attention. The trade-off is real: focus on a data flywheel means de-prioritising other claims on the platform roadmap.
2. Build the enabler system that supports the moat. Define how data is captured, how models improve, how solutions scale, and how quickly the organisation can iterate. This is the platform layer Teraflow refers to as DAPA (Digital AI Platform Accelerator). The moat lives or dies on whether this layer is treated as core infrastructure rather than a project.
3. Govern the moat like a core business, not a portfolio of experiments. Moat building is a multi-year programme. Boards should track a small number of leading indicators tied to the chosen moat: cost per transaction, network liquidity, share of customer interactions mediated by AI, reuse ratio of platform components. This is what disciplined resource allocation looks like.
Frequently asked questions
What is an AI competitive moat? An AI competitive moat is a hard-to-replicate set of assets or capabilities (data, infrastructure, workflows, customer relationships, regulatory permission, or trust) that lets a company sustain superior returns from AI investment, even when competitors have access to the same models.
How many AI moats does McKinsey identify? McKinsey identifies nine: six strategic moats (economies of scale, privileged data, embeddedness, network effects, business model disruption, constrained assets) and three capability moats (velocity, regulation and compliance, trust).
Why don’t apps and tools create competitive advantage in AI? Apps and tools can be copied. McKinsey’s analysis of the 2018 to 2022 mobile banking wave showed that app adoption did not widen the gap between leaders and laggards. Advantage came from integrating digital and AI across the entire customer journey.
What is the most common execution mistake enterprises make on AI moats? Confusing a single model in production with a moat. A moat is the platform, data architecture, and workflow that allow the next ten models to ship at low marginal cost. One model is a use case.
Where should a CEO start? Identify the one to three moats where the business has a structural position, commit resources explicitly, and build the platform (data, MLOps, governance) that lets those moats compound over time.
Conclusion
Same models, same access, same tooling: that is the new baseline. Advantage will not come from the model itself. It will come from the organisations that convert shared technology into proprietary moats faster than their competitors.
The strategic question is not which AI to deploy. It is what infrastructure, data architecture, and operating-model decisions make AI compound for you and not your competitor.
Source: Dago Diedrich, Evan Williams, Tanguy Catlin, and Tim Fountaine, “From AI table stakes to AI advantage: Building competitive moats,” McKinsey Quarterly, May 2026. Read the original.
