Why AI’s Next Phase Belongs To Infrastructure

By Laura Connell and Andreas Cleve

The artificial intelligence wave is entering its most valuable phase. Even in conservative scenarios where AI capabilities plateau at current-generation models, analysts project tens of trillions in value creation as companies integrate AI into their operations. In more ambitious outlooks, the impact could rival the industrial revolution itself.

The question isn’t whether this transformation will happen, but where value will accrue as the market matures from frontier research to broad deployment — and over what timeframe. Investment dispersion reflects genuine uncertainty about AI’s trajectory, but infrastructure value compounds regardless of which scenario unfolds.

Why this time is different

Every tech cycle attracts skeptics who compare it to past bubbles.

The data tells a different story. During the dot-com boom, 97% of fiber capacity sat unused. In 2025, the opposite is true: Every unit of compute is active, utilization rates remain high, and returns on AI infrastructure are already positive.

Global investment in generative AI reached $49 billion in the first half of the year, driven by hyperscalers reinvesting profits, rather than speculation.

The first wave of AI value went to the foundation model builders — OpenAI, Anthropic and others — whose breakthroughs triggered an explosion of experimentation at the application layer. That wave proved what was possible.

Now, as investment scales and adoption spreads into regulated sectors, the challenge has shifted downstream. The frontier is no longer building larger models, but getting AI live — safely, reliably and within real-world constraints. Deployment at this stage depends on more than compute access or APIs; it requires embedded teams who understand the domain, the workflows and the regulations that shape how AI performs. That blend of infrastructure and expertise is becoming the new differentiator — the layer that turns potential into production.

Physical capacity is tightening. Data centers now consume up to 10 gigawatts per site.

But the greater bottleneck is operational: compliance frameworks growing more complex, orchestration challenges in global deployment, and the gap between proof-of-concept and production. When AI systems stall in pilots, even the most advanced infrastructure struggles to deliver returns.

The deployment gap

Across sectors, between 80% and 95% of AI projects fail, not only because of inaccuracies but because compliance and validation are treated as afterthoughts. In healthcare, U.S. hospitals spend an estimated $39 billion annually on compliance and administrative oversight. Similar dynamics exist across financial services, energy and any domain where AI must operate within regulatory boundaries.

Developers are asking new questions: How can models remain auditable as they evolve? How can performance stay consistent across jurisdictions with different data rules? How can costs be contained as usage grows unpredictably? In healthcare, this means API platforms that handle medical-grade data securely, automate audit trails for regulators, and enable deployment in weeks instead of months. Building that capability from scratch delays product launches and drains engineering resources most teams don’t have.

The next decade of value lies in the infrastructure making AI both compliant and scalable — the layer that allows innovation to move from impressive demos to mainstream deployment.

The rise of vertical infrastructure

The next evolution of infrastructure will be vertical. General-purpose compute makes AI possible, but domain-specific infrastructure makes it usable. The highest stakes industries — healthcare, energy, finance, precision manufacturing — depend on systems that understand their regulations, workflows and risk thresholds. That’s where the next generation of durable value will form.

The demand signal is clear. To secure long-term success, developers need to build on AI infrastructure that makes their solutions fully deployable. Accuracy is necessary but not sufficient. Deployment is the bottleneck.

Corti’s 1 experience illustrates how this is playing out. Health systems need AI they can deploy and trust, not just test. By embedding validation, compliance and audit directly into its APIs, Corti enables developers to integrate clinical-grade AI in weeks rather than months. What began as a healthcare challenge is becoming a broader design pattern — an infrastructure model that abstracts away friction between innovation and safe adoption at scale.

Europe’s structural advantage

Europe’s early emphasis on interoperability, privacy and safety once seemed like a constraint. As the market shifts from experimentation to widespread deployment, those principles have become a competitive advantage.

This is playing out in real procurement decisions. When a top-three global healthtech provider evaluated infrastructure for their clinical AI deployments, they chose Corti over Microsoft, OpenAI and Anthropic. What began as months of technical due diligence became a landmark agreement — a signal that global buyers now prioritize compliance architecture and deployment readiness alongside model capability.

Companies that embedded regulatory principles from day one are structurally better positioned for this phase. European builders have been designing for this complexity from the start, treating compliance as a core product requirement rather than a go-to-market barrier.

Every transformative technology becomes more efficient over time, and AI is no exception. Model efficiency improves by an order of magnitude annually, while infrastructure-led automation removes friction across regulated sectors. This is not a bubble deflating; it is a market maturing from frontier research to scaled production.

The next era belongs to the builders who recognized early that deployment, not just capability, would define who wins. The hype will fade, as it always does. What remains is infrastructure purpose-built for the hardest problems, allowing thousands of companies to turn AI’s transformational potential into measurable reality.

Laura Connell is a senior partner at Atomico, the founder-built European venture capital firm. Connell leads growth-stage investing at Atomico, where she focuses on AI infrastructure and applications.

Andreas Cleve is the co-founder and CEO of Corti, a pioneering AI company building infrastructure and foundation models for healthcare developers. After nearly a decade as a multientrepreneur in artificial intelligence, Cleve founded Corti with Lars Maaløe to help healthcare developers make clinical workflows faster, smarter and more efficient. Today, Corti’s trusted AI powers real-time consultations across the U.S. and Europe — eliminating administrative burdens and bringing expert-level reasoning to every corner of healthcare.

Illustration: Dom Guzman