Economies of Scale, Product Cycles, and Supply Chains

Why these are the two forces that matter most right now.

Industries do not evolve uniformly. Some consolidate around a handful of dominant firms, while others fragment as new competitors enter and established advantages weaken. The difference often comes down to two forces: the structure of the supply chain and the speed of the product cycle.

---

Where scale compounds

Consolidation is most common when production requires enormous upfront investment, specialized inputs, technical expertise, and sustained utilization. In industries such as semiconductor manufacturing, automobiles, aerospace, and energy infrastructure, scale is not merely helpful — it is embedded in the economics. Larger companies can spread fixed costs across greater volumes, negotiate more effectively with suppliers, invest more heavily in process improvement, and operate complex assets at higher utilization. As scale increases, unit costs fall and margins improve, making it progressively harder for smaller competitors to survive.

The Economics of Scale

Yet technological change can abruptly overturn even mature industry structures. Joseph Schumpeter described this process as "creative destruction": innovation does not simply improve an existing market; it can dismantle the competitive advantages on which established businesses were built.

The consolidation of the early automotive industry offers a useful example. Assembly-line manufacturing did more than make cars faster. It changed the economics of the entire industry. Companies that adopted standardized, high-volume production reduced costs dramatically, while manufacturers dependent on slower, craft-based methods became uncompetitive. The technology strengthened the advantages of scale and ultimately concentrated the market around firms capable of financing and operating increasingly sophisticated production systems.

---

AI pulls in two directions

Artificial intelligence is producing a similarly consequential transition, but its effects are moving in two opposing directions.

Two Opposing Directions

At the infrastructure layer, AI reinforces consolidation. Advanced semiconductor fabrication, high-bandwidth memory, networking, data centers, as well as critical mineral and material security and power generation all require immense capital investment. The leading participants benefit from scale, technical specialization, scarce intellectual property, and deeply integrated supply chains that are unmatched. As AI models become more computationally demanding, these advantages may become even stronger. The cost of competing at the frontier rises, concentrating economic power around the companies that control critical bottlenecks.

At the software layer, however, AI can have the opposite effect.

Software has always been inexpensive to reproduce once built; its marginal distribution cost is effectively zero. What AI changes is the cost, speed, and expertise required to create it. Small teams can now prototype products, write code, design interfaces, automate support, analyze customer behavior, and iterate far more quickly than before, allowing for quicker end-to-end development. A company no longer needs a large engineering organization or extensive internal infrastructure to test whether a product has demand. It can begin with rented computing resources, scale incrementally with usage, and replace fixed investment with variable cost.

That shift weakens some of the organizational advantages historically enjoyed by large software companies. Incumbents may carry extensive legacy code, costly engineering structures, overlapping products, and infrastructure designed for an earlier technological era. Smaller competitors can build directly around AI-native workflows without supporting those inherited systems. They do not need to recreate the incumbent's entire product suite; they need only isolate one valuable workflow and serve it better, faster, or more cheaply.

This does not push software all the way to perfect competition — enterprise software has none of the interchangeability, negligible switching costs, or pricing transparency that would require. The more plausible outcome is monopolistic competition: many vendors offering differentiated products, while a small number of platforms retain substantial power because they control data, distribution, standards, or workflows. AI can therefore produce more application-level competition while simultaneously strengthening the platforms beneath those applications.

---

The software layer, examined more closely

The fragmentation case is directionally right but too categorical. Cheaper creation is not the same as lower barriers to entry, and writing code was never the only meaningful cost of building an enterprise-software company. A new entrant must still acquire customers, earn access to sensitive enterprise data, pass security, privacy, and compliance review, integrate with existing systems, provide reliability and contractual accountability, build a brand and a distribution organization, and convince a buyer it will still exist in several years. AI may make the first eighty percent of a product dramatically cheaper to construct; the remaining twenty — enterprise readiness, distribution, integration — can hold nearly all of its commercial value. And the same technology cuts costs for incumbents too: an established vendor can shorten its own product cycles and push improvements instantly across an enormous installed base.

The case also understates switching costs. A customer does not replace an entrenched system by finding a startup with a nicer interface; it has to reconstruct years of records, permissions, workflows, integrations, reports, automations, and compliance controls, along with the institutional knowledge encoded in them. The relevant test is not product superiority but migration superiority — whether the new product is enough better to justify the cost and risk of replacing the old one. Incumbents can survive for a long time without the best standalone product because the incremental benefit rarely clears the switching cost. Salesforce's fiscal-2026 attrition, measured as lost or reduced annual contract value, was roughly 8%, essentially unchanged year over year — not proof of a permanent moat, but a sign that displacement has not yet become acute.

And "software" is not one industry. AI will not weigh on every category equally.

Narrow point solutions, thin interfaces over third-party data, basic content generators, and features easily absorbed into a larger platform are the most exposed. Systems of record, security and identity infrastructure, regulated-industry platforms, deeply embedded operational systems, and products that own proprietary data or transaction networks are far more defensible. Salesforce is not simply a user interface for entering leads; it is frequently the underlying record, permission, workflow, and automation layer for customer-facing operations. AI may replace portions of the interface while leaving that control plane intact.

---

When proliferation strengthens the platform

The strongest objection to the fragmentation case is that it can invert. AI may spawn thousands of applications, but that very abundance creates a new problem for the enterprise: too many agents, tools, and disconnected data sources. Abundance upstream raises the value of whatever coordinates it — identity and permissions, data governance, auditability, workflow orchestration, integration, centralized monitoring, and a trusted system of record. In that environment a platform can become less valuable as a collection of individual applications and more valuable as the governed layer through which AI agents reach enterprise data and execute actions.

Salesforce is the useful case study. What keeps it in place is not its interface but the layer beneath it. It owns the customer-data schema — how an enterprise defines accounts, contacts, opportunities, cases, and territories — so that its operating language becomes institutional infrastructure rather than a discretionary application. It owns embedded workflows: approvals, forecasting rules, automations, and custom objects built over years, capturing organizational knowledge that is hard to transfer. It sits inside a self-reinforcing ecosystem of partners, integrators, developers, and trained administrators that a challenger would have to displace alongside the product itself. It has senior-level enterprise distribution that technology alone cannot replicate quickly. And it has the financial capacity — roughly $41.5 billion in fiscal-2026 revenue, about 95% subscription, against $72.4 billion of remaining performance obligations — to build, acquire, bundle, or subsidize a response. Structured, permissioned, contextual data is exactly what agents need, which can make that historical data moat more useful in an agentic world, not less.

---

Reading the two forces together

This is why product cycles and supply chains must be analyzed together.

Product cycles reveal how quickly an offering can be created, improved, copied, and displaced. When development cycles compress, incumbents have less time to earn returns on existing products before the market moves again. Competitive advantage shifts away from possessing a static product and toward maintaining a system capable of continuous experimentation, distribution, and adaptation.

Supply chains reveal where the difficult-to-replicate value remains. Even when competition expands at the application layer, it may concentrate underneath it. Thousands of AI software companies can emerge while relying on a comparatively small group of semiconductor manufacturers, memory suppliers, networking vendors, cloud platforms, and power providers. Competition at one layer can therefore strengthen the bargaining power of bottleneck suppliers at another.

The defining feature of the current AI cycle is a divergence: the economics of creation are becoming more distributed while the economics of infrastructure — and of the platforms that govern enterprise data — are becoming more concentrated.

---

Economic demotion, not obsolescence

Understanding where a company sits between those two forces is increasingly essential. The strongest businesses will either control a scarce supply-chain bottleneck or operate product cycles quickly enough to remain ahead of falling barriers to entry. The most vulnerable will be caught in the middle: capital-intensive without genuine scale advantages, or asset-light without durable differentiation.

For an entrenched platform, the danger is rarely immediate obsolescence. It is economic demotion — remaining deeply embedded while the interface, the intelligence, and the pricing power migrate elsewhere. The bearish mechanism is concrete: if agents reduce the number of humans who need seat licenses, seat-based revenue falls. The incumbent's answer is to change the meter, charging through a mix of licenses, conversations, and consumption-based credits so that agent usage can offset declining seats. Salesforce's early Agentforce results — roughly $1.2 billion in ARR, with more than half of Agentforce and Data 360 bookings coming from existing customers, at a 21% GAAP operating margin — suggest the installed base is attaching AI rather than leaving. But that is monetization of captive customers, not proof of competitive wins, and some of it is existing product rebadged around a new name.

The sharper thesis, then, is not that AI displaces incumbents but that it relocates value. AI lowers the cost of creating applications, yet value need not accrue to whoever produces the most of them. It accrues to whoever controls enterprise data, distribution, governance, and the execution of workflows. The decisive question for any embedded platform is whether AI becomes a growing consumption layer on top of what it already owns — or whether outside agents turn it into a commoditized back-end database.

Today's evidence favors resilience over imminent displacement. The bearish case strengthens materially if core organic growth stalls in the single digits, attrition or down-selling rises, human-seat contraction outruns agent consumption, or customers begin routing their important workflows through outside agents without paying the platform materially more. The bullish case strengthens if agent-driven consumption revenue grows faster than seat revenue shrinks, if attrition stays pinned near current levels through the agent transition, and if net-new customers — not just the installed base — start choosing the platform as their agent control plane. Both lists are observable in quarterly disclosures, which is what makes this a trackable thesis rather than a stance. The central investment question is therefore no longer merely whether a company benefits from AI. It is whether AI strengthens or weakens the economic structure surrounding that company — and whether value ultimately accrues to the product, the distribution channel, the workflow control plane, or the supply chain beneath it.

---

Sources: Salesforce FY2026 Form 10-K; Salesforce Q1 FY2027 results; Salesforce Agentforce pricing.