Do moats exist for AI-native companies? And if they do, are they different from the traditional SaaS world, and do they actually matter? Everyone has been ruminating on this topic, and the general consensus appears to have a high level of skepticism.
The basic premise of the new prevailing wisdom is that the barriers to building new software have been dismantled with the widespread availability of tools like Lovable and Cursor. Lack of coding ability and lack of time are no longer blockers – to create new software products, you’re only limited by your imagination. This has caused concern for large software incumbents whose functional technology can be particularly readily replicated in its simplest form, for example, Salesforce, Adobe and Docusign.
My hypothesis is that if you dig a bit deeper there are still effective moats to be built within AI-native businesses (even within agentic AI), but these are materially different from the traditional SaaS world. And despite record fast growth from the new set of AI-native businesses (agentic, infrastructure or generative); the difference between the enduring future category leaders and the ‘one hit wonders’ will be related to their ability to create effective versions of these new moats to drive strong and durable retention on top of their explosive new customer acquisition.
But how do you identify the category leaders?
To put some structure to this thinking, I’m using an established framework by Helmer called the 7 Powers. In his book “The Foundations of Business Strategy”, he suggested there exists a set of 7 business characteristics that can provide enduring competitive advantage and defensibility/moats vs new entrants. These are:
1. Economies of scale: (lower unit costs as scale increases)
2. Switching costs: (costs to change provider exceed value from switching)
3. Network effects: (value to a customer increases as userbase increases)
4. Cornered resource: (preferential access to proprietary value driving resource)
5. Branding: (higher perceived value to objectively similar offering)
6. Process power: (team, processes, culture & behaviour that drive operating performance)
7. Counter positioning: (incumbents can’t mimic new business model due to cannibalisation)
I’ve listed these in descending order of my view on their importance in the traditional SaaS world. I think this ranking is (relatively) uncontroversial so won’t dive into the reasons here. However in the new AI paradigm, these rankings have been shuffled and re-dealt so the new ranking is dramatically different.
Looking through a longer-term lens on company competitiveness; data network effects have become the singularly most important characteristic in the new world; whilst two of the previous top three defensible characteristics (scale economies and switching costs) have significantly decreased in importance.
The below visual gives a summary of the movements:
To provide a little more colour on the four powers/categories at the top of the table:
Network effects (data-driven)
- The most powerful networks typically exist in the consumer/marketplace world
- But in enterprise software, there are often community (particularly developer centric tools) or transactional components (fintech / insurtech) where having usage depth improves the quality of the proposition
- In AI, the compelling differentiator is the ability for software to learn from its users and improve the specificity and raw performance over time. Emergence Capital have published some great content on this called the Emergence Rate
- Some of the best examples here are vertical – including Harvey (legal) and Hebbia (finance), but also businesses like Grammarly and GitHub.
- Undoubtedly some fast-emerging companies are well able to succeed without the presence of these effects, but my view is in the medium to long term (3-5 years +), they will be paramount to retaining customers and establishing a robust market position
Process power (operational velocity)
- ‘Process power’ in the framework is a touch ambiguous, but refers to a broader set of team capabilities, culture, processes and use of technology to drive operational efficiency
- Lots of VC’s are focussing on this power in their thinking, claiming “speed is the moat” and I think there’s certainly some truth to this
- One of the best examples here is Perplexity’s speed to ship conversational search vs Google, despite Google’s enormous scale and depth/quality of their team. Their iterative culture, strong leadership and relentless user behaviour analysis and feedback incorporation has allowed them to outflank a global powerhouse in some core capabilities
- But it’s not a singularly effective moat, as speed of execution alone, if not combined with other elements, simply takes advantage of the current arbitrage between what’s technically possible in AI at the vanguard, and the typical enterprise/individual customer’s view of what’s actually possible. Over time, this gap will atrophy and execution alone won’t drive superior outcomes
- Finally another important element of processing power in the AI world is the ability for companies to stay at the vanguard with respect to the foundation models. Capabilities are moving at a breathtaking speed and application layer businesses built on top need to be responsive to this and ensure their platforms are consistently powered by the latest and greatest
Cornered resource (proprietary assets)
- The most prevalent proprietary element of value here is data – and truly proprietary data, for some use cases, can enable fine tuning and product customisation that delivers dramatically superior outcomes to what is generally available
- It can also refer to having unique access to narrow talent, or proprietary underlying technology or process know-how
- With the rise of synthetic data and general trend towards open-source, I believe these benefits are rarer than they used to be; but where they do exist, they can be hugely compelling
- Bloomberg is the traditional example here, but Scale AI’s labelling, tooling and dataset curation capabilities is another powerful one
Branding
- Sits in the middle of the table with equivalent overall importance to the SaaS world, but is undergoing huge change!
- As internet search volumes are dramatically dropping, the storefront for enterprise software needs to change to cater to human and agent shoppers
- Much larger focus is needed on technical comparative capabilities, discoverability through different channels, and on driving an engaged user/customer community
To take this a step further, I think this new version of the framework is interesting due to the potential for the new top categories to interact with each other to reinforce strength: if a business demonstrates an ability to persistently ramp up their processing power, iterate their platform in near real-time based on data signals, and augment their proprietary data/team/IP assets over time, they will become very difficult to displace.
It’s in the combination of the powers that the true moat exists
I believe this theory is a good explanation for the Meta/OpenAI poaching in the last few weeks. They’re treating these top researchers as another ‘cornered resource’, on top of their unique data asset, to drive their ability to execute more rapidly and more tightly loop this data and intellectual horsepower into the platform. On a much smaller scale, this feedback loop is happening in many of the top early-stage AI startups we are seeing emerge.
The reason this matters for earlier stage companies at the moment is there’s a good case that in the application layer or within vertical software, startups have a generational opportunity to win market share from large incumbents, because in many cases, the incumbents aren’t fundamentally well positioned to benefit from high processing power nor effectively implementing data network effects in the new world. This is because there’s been a step change in the capability set of what enterprises can do with AI, and so the old data assets that incumbents collected over many years often proves irrelevant for AI-driven features.
This data point is worth double clicking on. I think there’s two specific impact vectors of this data narrative:
- Lack of specificity – in many cases, historic data (product telemetry, usage, user engagement etc) might be in the wrong format/frequency/fidelity to be able to deliver real value.
- Lack of relevance – it’s possible that the change in user behaviour is so stark that historic backward-looking information loses the entirety of its usefulness and predictive power because the interface and usage paradigm is unrecognisably different.
That’s why we’re seeing such explosive growth of new entrants: enterprises (and individuals) are urgently shopping for new capabilities, incumbents are structurally unable to adequately deliver and there’s less lock-in to old solutions. This has caused the gold rush and rapid digitisation/re-platforming that’s occurring right now.
But the long term winners will not necessarily be those that simply emerge fastest out of the blocks (speed by itself isn’t sufficient as a first-mover advantage); but rather those that thoughtfully combine relevant proprietary IP with relentless execution improvements and can manifest responsive and effective data network effects in their platform. Those companies will become the durable enterprise category leaders of tomorrow.
