LLMs Will Become the Airlines of Tech

We are heading toward a world where every company needs LLMs - but most LLM providers will not make money.

Not because the tech fails. Because it succeeds too well.

• LLMs are becoming infrastructure, like email, search, or cloud. You do not buy email anymore; you just expect it to work.
• Margins collapse when differentiation vanishes. Once models are good enough, buyers shop on price, SLAs, and integration - not brand.
• Capital intensity stays high. Training, inference, safety, and compliance costs do not disappear; they become the new fuel and fleet maintenance.
• Upstream winners capture value. Chips, fabs, lithography, and data-center power will keep extracting profits - just as aircraft and engine makers do for airlines.
• Downstream, it is a race to the bottom. Open-weight models, distillation, and good-enough APIs turn foundation models into commodities.

This is not doom for AI. It is a signal: the money moves from who has the model to who owns the workflow, data, and distribution.

Why the Airline Analogy Fits

Airlines are indispensable to the global economy - and notoriously low-margin. The parallels with LLM providers are striking:

• High fixed costs + relentless reinvestment. Planes depreciate; models stale. You must keep buying the newest fleet to stay competitive.
• Intense competition, low switching costs. Customers jump from Swiss to EasyJet; enterprises jump from Model A to Model B when price or latency improves.
• Hardware dependency. Airlines depend on Boeing/Airbus; LLMs depend on Nvidia/ASML/TSMC. The upstream monopolies/duopolies skim the profits.
• Commoditization of the core service. A seat is a seat; a token is a token. Differentiation shifts to bundles, loyalty, and operational excellence - not the core product.

As Gary Marcus and others have argued: LLM companies are likely to be like airline companies: small margins, intense competition, high expenses.

The Adoption Curve: Peak Hype, Then Normalization

Every breakthrough tech follows a similar arc:

• Internet, fax, PCs, email, ecommerce, smartphones, EVs - each went from magic to mandatory to meh, it is just there.
• LLMs are now in the magic to mandatory phase. Adoption is exploding; expectations are sky-high.
• Next comes the meh phase. Models are embedded everywhere, but treated as utilities. Buyers optimize cost, latency, and reliability - not marvel at capabilities.
• Growth continues; glamour fades. Usage soars, but pricing power shrinks. The industry becomes essential and unexciting - like electricity or broadband.

This cooling-off is not failure. It is maturity.

So Where is the Money in an Airline-like LLM World?

If foundation models become commodities, value accrues elsewhere. For wealthtech, RIAs, and family offices, that is actually good news.

1) Own the Workflow, Not the Model

• Vertical AI beats horizontal AI. A generic chatbot is a commodity; a compliance-aware, document-grounded, audit-ready workflow is not.
• Bundle models with process. Onboarding, KYC, IPS drafting, meeting prep, client reporting - these are where defensibility lives.

2) Data Moats > Model Moats

• Proprietary data beats public weights. Your client files, meeting notes, investment memos, and decision logs create context no public model can replicate.
• Grounding and retrieval matter more than raw IQ. In wealth management, accuracy, traceability, and governance trump smartest model on the bench.

3) Distribution and Trust Are the New Moats

• Embedded in tools people already use. SharePoint, CRM, portfolio systems, email, calendar - integration is the product.
• Brand and reliability win. In a world of interchangeable models, clients stick with vendors who are safe, stable, and accountable.

4) Cost Engineering Becomes a Core Competency

• Model routing, caching, and distillation will separate profitable AI products from burn machines.
• Small, specialized models (SLMs) will handle 80% of tasks; big models become first class only when needed.

What This Means for Wealth Managers and RIAs

You do not need to bet on which LLM wins. You need to bet on how AI is wired into your business.

• Treat LLMs as infrastructure. Evaluate them like cloud or email: uptime, cost, security, compliance, vendor risk.
• Prioritize governance. Audit trails, data residency, access controls, and human-in-the-loop matter more than model benchmarks.
• Measure ROI in workflows, not tokens. Time saved on onboarding, fewer errors in reporting, faster investment committee prep - these are the real KPIs.
• Avoid vendor lock-in at the model layer. Use abstraction, routing, and fallbacks so you can swap models without rewriting your business.

A Contrarian Take (That is Actually Optimistic)

Yes, many LLM providers will struggle to make money.

But that is exactly what makes AI safe to adopt in regulated, high-stakes environments like wealth management.

• When a technology becomes a utility, it becomes boring - and boring is bankable.
• When no single model has a lasting moat, you gain negotiating power and optionality.
• When the hype cools, the real work begins: embedding AI into processes that compound over years, not quarters.

The winners will not be the companies with the best model.

They will be the firms that turn commoditized intelligence into differentiated outcomes - for clients, teams, and investment processes.

If You Take One Thing Away

• LLMs will be everywhere.
• Most model providers will not be very profitable.
• Your edge is not the model - it is your data, workflows, governance, and distribution.

In wealthtech, the airline future of LLMs is not a warning.

It is an invitation to stop chasing model hype and start engineering durable, client-value-creating workflows.

I work with financial institutions on technology integration and data aggregation (including API/SDK solutions at Collation.AI). Happy to connect and discuss your firm's technology strategy.