By 2026, 47% of European startups that raised seed funding in the AI boom will have folded not because the technology failed, but because they built businesses on rented ground.
That number hasn't happened yet. But the pattern has. We've seen it with dot-com in 2001, with fintech consolidation in 2018, with crypto in 2022. Every bubble inflates the same way: capital floods in, valuations detach from revenue, and the survivors aren't the ones who rode the hype highest they're the ones who owned something real underneath it.
The question isn't whether the AI bubble pops. It's whether your business is positioned to survive when it does.
Why AI-Dependent Businesses Are More Fragile Than They Look
Most founders building on AI right now aren't building businesses they're building integrations. There's a critical difference.
An integration is a wrapper around someone else's infrastructure. You're selling access to OpenAI or Mistral with a prettier interface. Your margins are whatever they let you keep. Your moat is whatever they don't copy next quarter. When Anthropic, Google, or the EU regulatory environment shifts and it will your "product" is the first thing that gets commoditised.
This isn't pessimism. It's mechanism.
The AI stack is consolidating upward. In 20232024, foundation model providers absorbed over $100 billion in global investment, according to PitchBook data. The companies at the bottom of the stack compute, data centres, model weights are getting more powerful and more capital-intensive at the same time. Small players building application-layer products are being squeezed from both directions: rising API costs below and larger competitors above.
The EU's AI Act, fully enforceable from August 2026, adds a third pressure. Compliance costs for high-risk AI applications are estimated at 300,000400,000 per product for mid-size companies, according to the European Parliament's own impact assessments. That's a moat for incumbents. For founders without deep pockets, it's a wall.
Why "Build on AI" Advice Is Structurally Incomplete [Risk]
The standard advice is to "build AI into your product." Every accelerator says it. Every VC deck demands it. But this advice skips the mechanism entirely.
When you build on AI, you inherit AI's risk profile. That includes model deprecation (GPT-3.5 is already being phased out), data dependency (your fine-tuned model is only as good as the data you licensed), and regulatory exposure (the EU's liability framework for AI-generated outputs is still being written). None of these risks are hypothetical they're already baked into the timeline.
The companies that get this right aren't avoiding AI. They're using it as a cost tool rather than a value proposition. There's a massive difference between "we automate internal operations with AI" and "our product is AI." One gives you margin. The other gives you a feature that your competitors can clone in six weeks.
What Survives a Bubble: The Physical-Core Model
Here's the actual question: what do human beings need regardless of what's happening in venture capital?
Food. Housing. Health. Physical security. Energy. Relationships. Status. These needs don't get disrupted by an LLM. They may get served differently, but the underlying demand is structural, not speculative.
The businesses that weather tech corrections historically share one trait: they own an asset or relationship that technology serves, rather than one that technology replaces.
A restaurant that uses AI for inventory management still sells food. A physiotherapy clinic that uses AI for appointment scheduling still sells physical recovery. A real estate business that uses AI for lead generation still sells property. The AI is a cost lever. The business is the thing.
This is the Legacy Plan framework: identify the human-constant need, build a physical or relational asset around it, and use technology to operate that asset more efficiently not to replace it.
Building Around Physical Assets: What EU Data Actually Shows [Leverage]
European markets have specific structural characteristics that make physical assets unusually defensible right now.
Housing. Across the EU, housing supply has been structurally undersupplied since the post-2008 construction collapse. Germany has a shortfall of 700,000 homes, according to the German Property Federation (ZIA). The Netherlands is short by 400,000 units. Spain's rental market saw 17% average rent increases in 2023 in major cities (Idealista data). None of this gets fixed by AI. It gets fixed by physical construction, zoning reform, and capital allocation to real assets.
Businesses that position in this sector property management, renovation services, short-let operations, construction materials supply are operating with structural tailwind that has nothing to do with tech cycles.
Energy infrastructure. The EU's REPowerEU plan commits 300 billion to energy transition through 2030. Heat pump installation, solar panel fitting, building retrofitting under the Energy Performance of Buildings Directive these are physical trades with high demand, low automation risk (fitting and surveying are hard to roboticise), and EU subsidy support. A solar installation business in Poland or a heat pump service company in France is operating in a market with legislatively mandated demand growth.
Health and physical services. Europe is aging. By 2050, 29% of the EU population will be over 65, according to Eurostat projections. Physiotherapy, elder care, nutrition services, fitness training, manual therapy the demand trajectory here is determined by demographics, not tech cycles. These are also markets where the core service delivery (physical, relational, tactile) cannot be automated in the time horizon that matters to a business built today.
Why Relationships Are the Most Underrated Asset [Quality]
Beyond physical infrastructure, the most durable business asset is a trust relationship that competitors can't replicate with an API call.
Consider the difference between a financial advisor who has managed a family's portfolio through a divorce, an inheritance, and a property purchase versus a robo-advisor with lower fees. The robo-advisor wins on cost. The human advisor wins on everything else that determines whether the client stays. That relationship is worth 712 times annual revenue in private wealth contexts, according to McKinsey European wealth management data.
The same principle holds in B2B. European SMEs spend an average of 4,200 per year on accountancy services, according to EU SME Observatory data not because accounting software doesn't exist and is cheaper, but because the relationship with a trusted accountant who knows the business carries risk-reduction value that software can't replicate.
If you're building a service business, the question is: am I building toward a relationship asset, or am I building toward a transaction that can be cut when something cheaper arrives?
Relationship assets compound. Transaction businesses commoditise.
Using AI as an Operating Layer, Not an Identity [Speed]
Here's the actual architecture of a legacy-proof business: physical or relational asset at the core, AI as the operating system underneath.
This is already playing out in European mid-market businesses that most people aren't writing about because they're not exciting enough for TechCrunch.
A mid-size German Handwerk (trades) firm using AI-powered scheduling, predictive maintenance alerts, and automated invoicing runs 3040% leaner on back-office costs than a comparable firm running manual operations while the core value delivery (skilled labour, physical installation, warranty-backed service) remains entirely human. The AI gave them margin. It didn't create their business.
A Spanish healthcare network using AI triage for appointment routing reduced no-show rates by 22% and increased specialist utilisation to 87% capacity (reported in a 2023 case study by the Spanish Association of Health Management, SEDISA). The doctors still see patients. The AI just made sure the right patients got there.
The formula isn't complex:
The higher your technology dependency ratio the more your core value proposition requires specific platforms, models, or infrastructure to function the more fragile your business is when the bubble adjusts.
Stress-Testing Your Business Against the Bubble Scenario [Cost]
Run this exercise with whatever you're building or operating.
Assume: OpenAI prices rise 3x. EU AI Act compliance costs hit. One major model provider exits the European market due to regulatory friction (Meta already threatened this in 2023). What happens to your revenue?
If the answer is "we lose the product entirely" you're an integration, not a business.
If the answer is "our costs go up but the core service still runs" you have a business.
If the answer is "we actually benefit because our AI-inefficient competitors get hurt worse" you have a moat.
The third category is where the money goes after bubbles correct. During the dot-com crash, Amazon lost 93% of its share price but kept its fulfilment infrastructure and customer data. The physical warehouse network and customer relationships were the business. The stock price was the speculation. By 2005, the distinction was obvious. By 2010, it was worth hundreds of billions.
| Business Type | AI Dependency | Bubble Resilience | Post-Bubble Position |
|---|---|---|---|
| Pure AI wrapper/SaaS | Very High | Low | Acquisition target or closure |
| AI-enhanced service | Medium | High | Competitive advantage |
| Physical asset + AI ops | Low | Very High | Market consolidation winner |
| Relationship-driven + AI tools | Low | Very High | Premium pricing maintained |
| Traditional, no AI | None | Medium | Cost-squeezed but surviving |
The table is blunt on purpose. "No AI at all" isn't the answer either those businesses get out-competed on margin by anyone who does adopt operational AI. The target position is the middle rows: physical or relational at the core, AI used to operate cheaper and faster.
The Build Strategy: Where to Put Capital Right Now
If you're allocating capital or founder time in the next 1224 months, the legacy play looks like this.
Acquire physical assets while AI euphoria keeps attention elsewhere. European commercial real estate is at a 15-year low in several segments (office and retail, specifically) because capital has chased tech. Logistics, light industrial, and residential-adjacent assets are priced more rationally. This is a window.
Build trade or service businesses in EU-subsidised sectors. Energy retrofit, elder care, sustainable construction these aren't glamorous, but they have structural demand, available subsidies, and labour shortages that create pricing power. A well-run solar installation business in Central Europe with good technicians and decent operational software will have higher margins and more predictable revenue than 80% of AI-native startups in five years.
Use AI to hire less, not to sell more. The businesses that win post-bubble will be the ones that achieved lean operations before the correction, not the ones that pivoted to AI as a last resort. Automate your back office now. Keep your front-line value delivery human and relationship-driven.
Own your customer data and relationships directly. No platform intermediaries. No dependency on Meta's ad algorithm or Google's SEO. Email lists, direct contracts, community ownership these are the distribution assets that survived every previous correction.
Start Here
Pick one physical or relational asset that exists in your market independent of what any model provider decides next quarter. Build your next 90 days of founder time around making that asset more valuable. Use AI to do the operational work cheaper. Everything else is speculation.
The companies that people will write about in 2030 aren't being funded right now they're being built by people who understood that the technology was the shovel, not the gold.
Checking account status...
Loading comments...