# What Makes a Moat in IoT

> What is a moat in business, and what is one in IoT? Installed fleets, datagraphs, domain physics, certifications, and why adding AI does not create a moat.

![What Makes a Moat in IoT](https://tago.io/og/blog/what-makes-a-moat-in-iot.png)

I talk with hardware executives who can name every competitor in their category and none of their own moat. In IoT that gap decides who is still selling in ten years.

## What is a moat in business?

Warren Buffett gave business its most useful metaphor. A company is a castle, and the castle will be attacked. What protects it is the moat: the structural advantage that keeps competitors from taking your customers even when they copy your product. In his 1999 Fortune essay "Mr. Buffett on the Stock Market" he put it plainly: the key to investing is not judging how much an industry will change the world, but "determining the competitive advantage of any given company and, above all, the durability of that advantage." In his [2007 letter to shareholders](https://www.berkshirehathaway.com/letters/2007ltr.pdf) he sharpened it: "A truly great business must have an enduring 'moat' that protects excellent returns on invested capital." Morningstar later turned the metaphor into a rating system, classifying companies as wide moat, narrow moat, or no moat based on how long they can defend returns above their cost of capital.

The classic moat sources are stable across a century of business history: network effects, switching costs, intangible assets like patents and brand, cost advantages from scale, and dominance of a niche too small to invite a second entrant. Hamilton Helmer's _7 Powers_ is the modern refinement of the same list. For technology companies specifically, [Vaultinum's analysis of moats in the tech industry](https://vaultinum.com/blog/moat-in-tech-industry) translates those sources into software terms: proprietary code and IP, architecture that took years to build, data advantages, and deep integration into customer systems. It is a good map. It is also a generic one, written for software companies, and IoT is not a software business. It is a physical business with software inside, and its moats behave differently.

## What is a moat in IoT?

The best academic treatment of this question came from Harvard. In 2014, Michael Porter and James Heppelmann published [How Smart, Connected Products Are Transforming Competition](https://hbr.org/2014/11/how-smart-connected-products-are-transforming-competition) in Harvard Business Review, and it remains the sharpest analysis of IoT competition written. Their argument: once products carry sensors, software, and connectivity, the five forces themselves shift. Smart, connected products can raise barriers to entry, deepen switching costs, and hand the incumbent a data advantage that compounds with every unit shipped. A decade later, the companies that read that article carefully are collecting on it.

Working with hardware manufacturers in more than 130 countries, I see the IoT versions of the classic moats up close. They are physical, slow to build, and slow to breach.

**The installed fleet.** A SaaS competitor can steal a customer with a better demo and a migration script. Nobody migrates 40,000 water meters with a script. Every deployed device is both a revenue anchor and a sensor feeding you data your competitor cannot buy. Replacing a fleet is a capital project with truck rolls, and buyers know it. Switching costs in IoT are measured in ladders and lift trucks, not export files.

**The datagraph.** Fleet data collected across thousands of customers, linked to how machines actually behave in the field. No competitor can synthesize it, because generating it requires owning the fleet. This is the data advantage Porter and Heppelmann predicted, and the asset at the center of _Fusion Strategy_, the Govindarajan and Venkatraman book I wrote about last week in [Climbing the Fusion Ladder](https://tago.io/blog/climbing-the-fusion-ladder).

**Domain physics.** A software startup can rent cloud servers in an afternoon. It cannot rent thirty years of knowing how a compressor drifts before it fails or how a sensor behaves through a Brazilian summer. That knowledge is baked into your firmware, your calibration curves, and your support team, and it does not transfer through a job posting.

**Certifications and approvals.** UL, FCC, CE, ATEX, FDA. Every regulatory gate you have passed is a wall a new entrant has to climb at full price and full calendar time. Tech investors underrate this moat because pure software rarely has it.

**Integration into operations.** When your devices feed a customer's maintenance schedules, billing, and compliance reports, you are no longer a vendor. You are infrastructure. Ripping you out means redesigning how the customer runs.

![The moat in IoT: the product at the center, defended by the installed fleet, the datagraph, domain physics, certifications, and operational integration; features, connectivity, dashboards, and AI models sit outside the moat because competitors match them in months](https://tago.io/images/blog/what-makes-a-moat-in-iot/moat-in-iot.svg)

Notice what is missing from that list: hardware features. Precision, build quality, and spec-sheet advantages are real, and they erode. A feature is a head start. A moat is a structure. I made this argument at length last week: the fusion ladder, from fusion products up through fusion solutions, is a moat-widening program. Each rung deepens switching costs and grows the datagraph. Buffett tells his managers their daily job is to widen the moat. For a hardware company, climbing the ladder is what widening the moat means in practice.

## Is AI a moat?

No, and the reasoning matters, because in most IoT pitch decks I see this year the moat slide says AI.

A moat is, by definition, something your competitor cannot easily replicate. Foundation models are the most replicable capability in the history of software: every competitor you have can call the same API, at the same price, by this afternoon. A capability that anyone can rent by the token cannot be a moat, whatever else it is. Even the companies building the models know this. The leaked 2023 Google memo that made the rounds was titled ["We Have No Moat, and Neither Does OpenAI"](https://www.semianalysis.com/p/google-we-have-no-moat-and-neither), written by a Google engineer watching open-source models replicate frontier features in weeks. If the model builders doubt models are a moat, the companies merely calling those models should doubt it more.

The refuge answer is "our AI is trained on our data," and it deserves scrutiny too. Martin Casado and Peter Lauten of Andreessen Horowitz took it apart in [The Empty Promise of Data Moats](https://a16z.com/the-empty-promise-of-data-moats/). Most claimed data network effects are really scale effects, and they run backwards from true economies of scale: the cost of collecting data grows while the marginal value of each new record shrinks, because new data increasingly overlaps what you already have. Data alone, they warn, can misdirect founders from the work that actually wins markets.

So AI creates no moat. What it does, on the right foundation, is widen one. The distinction is the whole game.

> AI cannot dig a moat. It can widen one that already exists.

Run the test on your own company: if your closest competitor calls the same model API tomorrow, what do they still lack? If the answer is nothing, AI gave you a feature, and features get matched. If the answer is your fleet, your datagraph, and your domain physics, then AI is the multiplier on an asset only you hold. Failure prediction trained on your fleet's failure signatures is defensible because the signatures are, and the model is the cheap part. The same model pointed at public data is a demo.

This is why the a16z caveat matters more in IoT than anywhere else. Casado and Lauten allow that data becomes defensible when the source is proprietary and hard to reach. An installed fleet is exactly that: the data does not exist until your machines generate it, and your machines are behind your customer's fence. IoT companies hold the rare version of the data moat that survives scrutiny. Most of them have not built anything on it yet.

## Draw your moat

I closed last week's post by asking executives to draw their datagraph. The moat version of the exercise takes ten minutes. Write your product in the middle of a page. Around it, write only the things a well-funded competitor could not match within two years: the fleet, the data, the certifications, the domain knowledge, the operational integrations. Cross out everything they could match: features, connectivity, dashboards, and yes, the AI model. What survives is your moat. For most hardware companies the surviving list is stronger than they expect and less developed than it should be.

The plumbing under all of this, collecting fleet data reliably and turning it into products, is solved; platforms like ours exist so you never mistake infrastructure for strategy. The moat was never going to come from us or from any vendor. It comes from what only you own: the machines in the field and the decades of physics behind them. AI just raised the payoff for finally putting them to work.

## Sources

- Buffett, W., "Mr. Buffett on the Stock Market", Fortune, November 22, 1999
- Buffett, W., [Letter to Berkshire Hathaway shareholders, 2007](https://www.berkshirehathaway.com/letters/2007ltr.pdf)
- Porter, M.E. and Heppelmann, J.E., [How Smart, Connected Products Are Transforming Competition](https://hbr.org/2014/11/how-smart-connected-products-are-transforming-competition), Harvard Business Review, November 2014
- Govindarajan, V. and Venkatraman, V., [Fusion Strategy: How Real-Time Data and AI Will Power the Industrial Future](https://store.hbr.org/product/fusion-strategy-how-real-time-data-and-ai-will-power-the-industrial-future/10689), Harvard Business Review Press, 2024
- Casado, M. and Lauten, P., [The Empty Promise of Data Moats](https://a16z.com/the-empty-promise-of-data-moats/), Andreessen Horowitz, 2019
- ["We Have No Moat, and Neither Does OpenAI"](https://www.semianalysis.com/p/google-we-have-no-moat-and-neither), leaked Google memo, SemiAnalysis, 2023
- Helmer, H., _7 Powers: The Foundations of Business Strategy_, 2016
- Vaultinum, [MOAT in the Tech Industry](https://vaultinum.com/blog/moat-in-tech-industry)

[llms.txt](https://tago.io/llms.txt)