The moat has moved: why hardware companies are winning the AI era

What happens when the thing that made your business defensible can be generated for free?

That’s the question confronting every software founder right now. In the first six weeks of 2026, nearly $2 trillion in market capitalisation has been wiped from the global software sector. Atlassian down 35%. Salesforce down 28%. Thomson Reuters down 16% in a single session. The catalyst was not a recession, a rate hike, or a regulatory crackdown. It was a set of plugins.

On 30 January, Anthropic released a set of open-source plugins for its Claude Cowork platform - simple files that automate tasks across legal, sales, finance and data analysis. By the following Monday, roughly $285 billion in market value had evaporated. London was not spared. RELX fell 14%, the London Stock Exchange Group dropped 13%, and Sage, Pearson and Experian all took significant hits. As Deutsche Bank's Jim Reid put it, there has been "a clear shift in markets from AI euphoria towards growing concern about disruption to existing business models."

And the drama continues this week as IBM has just seen its shares tank by over 13% in a single day. The decline came after Anthropic announced that its Claude Code tool could be used to modernize legacy systems running COBOL, a programming language widely associated with IBM's mainframe systems.

The market has given this moment a name: the SaaSpocalypse. But what does it actually mean for founders? And which businesses will still have defensible moats when the dust settles?


The seat is dead

The core of the crisis is structural, not cyclical. For two decades, SaaS companies charged per user, per month. The more humans using the software, the more revenue the vendor captured. But when AI agents can perform the same tasks without a human in the loop, the entire pricing model collapses.

PitchBook's February 2026 analyst note "SaaS Is Dead, Long Live SaS" frames the shift bluntly. Traditional SaaS captures around $1,200 per seat per year. Outcome-based "service as software" pricing could capture $10,000 per automated workflow - an 8x revenue uplift per unit. Software budgets will merge with payroll budgets as the distinction between tool and worker dissolves. IDC predicts that by 2028, 70% of software vendors will have refactored their pricing around consumption, outcomes or organisational capability. The per-seat model is not just under pressure. It is becoming obsolete.


The super-UI trap

In the agentic AI land grab, not every SaaS company will make the transition. Many will simply be absorbed.

Anthropic and OpenAI are no longer content to sit beneath the application layer selling API access. They are moving up the stack, positioning themselves as what one analyst described as "the new operating systems of the enterprise." They capture the high-value intent layer - the user says what they want done - while legacy platforms risk being relegated to dumb databases that agents query in the background. PitchBook's advice to VCs is unsparing: if a startup's core value is a dashboard, it is going to zero. Startups developing "workflow wrappers" that facilitate processes without owning proprietary data are high-risk investments requiring immediate consolidation.

The risk is a world in which a handful of master agentic integrators create a new value layer that simply sucks in the data and functionality of thousands of mid-tier software companies. Think of it as the super-UI problem: your product becomes a feature inside someone else's agent, and you lose control of the customer relationship entirely. Gartner strikes a more measured tone, noting that agentic tools expose how much knowledge work remains manual but are not yet replacing SaaS for critical operations. Still, the direction of travel is unmistakable.


Investors are betting on atoms

While software burns, investors are rotating decisively towards businesses rooted in the physical world. The most visible sign is AI infrastructure spending: the hyperscalers plan to invest over $660 billion in 2026, nearly doubling 2025 levels, and Deloitte's outlook puts the global AI chip market at nearly $500 billion. Memory makers Micron, Samsung and SK Hynix are surging. Analysts are calling it a rotation towards the 'Physical Backbone' of AI.

But the shift is broader than chips and data centres. Enterprises are redirecting IT budgets away from traditional software subscriptions. And investors are reassessing which businesses have moats that AI cannot erode. You cannot download a data centre, but you also cannot download a jet engine, a warehouse robot, or a pharmaceutical manufacturing process. Hardware companies - whether they build infrastructure, industrial equipment, or consumer devices - possess barriers to entry that no amount of code can replicate. That is why atoms are winning.


The compounding data moat

The real advantage of hardware companies is not just that they make things that are hard to build. It is the combination of physical products with decades of proprietary operational data that AI now makes extraordinarily valuable.

Consider Rolls-Royce. The company has been collecting engine telemetry data for over 30 years. That dataset now feeds AI-powered digital twins that have extended the time between engine maintenance by up to 50%. Every new contract adds to a compounding data flywheel - more engines, more data, better models, better service, more customers. No competitor can retroactively capture three decades of linked operational history. The knowledge gap is widening, not closing.

Siemens is building an Industrial Foundation Model trained on 150 petabytes of product data and patents - a model designed to speak the language of engineering and manufacturing. With Nvidia, the company is planning the world's first fully AI-driven adaptive manufacturing site at its Erlangen factory. And in a separate collaboration with Microsoft and Rolls-Royce, Siemens has demonstrated how AI-powered design tools can produce aerospace components that are 25% lighter and 200% stiffer than conventional equivalents.

These companies are combining deep engineering knowledge, regulatory certification, and proprietary data moats in ways that no pure software business can replicate.

This is not AI as a feature bolted onto a product. This is AI as the engine of a fundamentally more defensible business model.


What this means for founders

So, is software finished? No. But the moat has moved.

PitchBook identifies 10 structural moats that still protect software incumbents - distribution, data context, switching costs, deep business logic among them. The companies that own the system of record, that hold proprietary data AI agents need to function, remain powerful. Deloitte counsels against overgeneralising, noting that cybersecurity, dev tools and industry-specific platforms will not all be disrupted at the same pace.

But the founders best positioned to thrive may be those building at the intersection of hardware, software and services. As we highlighted last week, Europe's DeepTech founders face unique commercial challenges in scaling complex solution architectures. Yet it is precisely that complexity - the combination of physical products, proprietary datasets, regulatory barriers and deep domain expertise - that creates the most defensible businesses in the AI era.

The UK is already producing companies built for this world. London-based Dexory combines autonomous warehouse robots with an AI-powered digital twin platform that has now captured over half a billion location scans. That physical-world dataset - growing with every client deployment - creates exactly the kind of compounding data moat that no pure software competitor can replicate. The company raised $165 million in its Series C last October. Lead investor Eurazeo described it as bridging the constraints of the physical supply chain with the power of AI-optimised intelligence.

McKinsey estimates DeepTech now accounts for roughly a third of all European venture capital. The opportunity is enormous.


The takeaway

The big lesson from the SaaSpocalypse? The moat is no longer code - AI has made code free. It is not even the user interface - agents are making that irrelevant. What remains defensible is proprietary data, deep domain expertise, and measurable outcomes.
 
For software founders, that means the old playbook of building a dashboard and charging per seat is over. But software businesses that own the system of record, that hold data AI agents cannot function without, and that price on outcomes rather than access still have powerful moats ahead of them.
 
And for hardware companies, many may now find they have been quietly creating the deepest moats of all. When you combine real-world barriers to entry with decades of compounding operational data and AI that turns that data into better products and services, you get a flywheel that no competitor can shortcut.
 
For founders building at the intersection of hardware, software and services, this is the opportunity of a generation.

 
Let's talk.