Phaidra, a startup building AI agents that autonomously optimize industrial infrastructure, has raised $50 million in Series B funding, led by Collaborative Fund with participation from Helena, Index Ventures, NVIDIA, Sony Innovation Fund, and others.

The new round will accelerate Phaidra’s mission to transform data centers and other energy-intensive facilities into intelligent, self-optimizing “AI factories.” Its software autonomously manages cooling, power, and workload distribution -  unlocking both energy efficiency and higher performance.

---

Solving the Hidden Bottleneck of AI Infrastructure

The AI boom has triggered an equally massive strain on the world’s data infrastructure. Each new model release drives exponential compute demand -  and behind that demand is an invisible cost: electricity.

Global data centers already consume roughly 2.5% of total electricity, and that figure is expected to reach up to 8% by 2030 if unchecked. (IEA, 2024) Cooling alone can account for 25–40% of a facility’s power draw, often wasted due to poor coordination between systems.

Phaidra’s flagship agent, known internally as “Alfred,” uses reinforcement learning to act as a virtual operator for these systems -  constantly adjusting controls to minimize energy use while maintaining peak performance. Instead of running fixed rules, it learns dynamically from sensor data, temperature shifts, and historical patterns.

The impact? 10–30% energy savings on average, and in some cases, even more dramatic results when integrated into complex, multi-zone facilities.

---

Why This Moment Matters

The AI data infrastructure market is projected to expand at a 20–25% CAGR through 2035, with industrial AI software alone expected to surpass $45 billion by 2030. (Allied Market Research, 2025)

Yet the real story isn’t just about efficiency -  it’s about system control. The more interdependent AI infrastructure becomes, the more valuable orchestration layers like Phaidra’s will be. As compute, cooling, and power are increasingly managed in feedback loops, control software becomes the brain that defines how energy translates into intelligence.

That’s why NVIDIA’s participation in this round matters: the hardware leader sees intelligence not just in GPUs, but in the physical systems that feed them.

---

The Founder Lesson Hidden in Phaidra’s Playbook

Here’s where Phaidra’s story becomes pure strategy gold for founders: in complex industries, real leverage doesn’t come from building the fastest product -  it comes from owning the coordination layer everyone else depends on.

Phaidra didn’t start by building new hardware or reinventing cooling tech. It started by asking a higher-order question: What if the real inefficiency isn’t in the machines themselves, but in how those machines talk to each other?

That shift in thinking is profound. Most startups attack one component -  a better sensor, a better algorithm, a better interface. But system-level problems require orchestrators, not components.

By focusing on orchestration, Phaidra positioned itself as the command center of the entire industrial stack. Once your software becomes the medium through which multiple critical systems coordinate, you’re not just a tool -  you’re infrastructure. And infrastructure compounds value every time a new subsystem plugs into you.

That’s the same architectural principle behind Snowflake’s data cloud and Databricks’ lakehouse. They didn’t compete on a single feature -  they became the environment that made every feature elsewhere interoperable.

For founders, the takeaway is sharp: if you want enduring defensibility, don’t chase motion -  design leverage. Instead of optimizing within one silo, find the intersection where silos collide and chaos reigns. Solve there, and you’ll build something others can’t easily dislodge.

Phaidra’s moat isn’t its algorithm -  it’s the dependency graph it’s building across the world’s most complex physical systems.

---

From DeepMind Roots to Industrial Scale

Phaidra’s CEO Jim Gao, a DeepMind alumnus, previously led the team that helped Google cut data-center cooling energy by 40%. That project proved the power of reinforcement learning for continuous control -  but at the time, it was limited to Google’s internal infrastructure.

Phaidra takes that proof public. Alongside co-founders Veda Panneershelvam (AI researcher) and Katie Hoffman (industrial systems expert), Gao is deploying a scalable platform that any data-intensive company can use -  from hyperscalers to energy utilities and pharmaceutical manufacturers.

The company’s system integrates directly into existing building management or SCADA systems, using APIs and digital twins rather than physical retrofits -  a key advantage for adoption.

---

Where the Market Is Moving

The broader data center infrastructure market is forecast to reach $395 billion by 2032, growing at 11.1% CAGR, as hyperscalers and enterprises rush to expand AI capacity. (Fortune Business Insights, 2025)

Meanwhile, industrial automation and energy optimization -  Phaidra’s adjacent verticals -  are projected to top $350 billion combined by the early 2030s. The convergence of those markets under sustainability mandates gives Phaidra a unique dual tailwind: climate policy and AI adoption.

And in a world where every watt counts, reinforcement-learning-driven control could become standard, not optional.

---

The Road Ahead

With fresh capital, Phaidra plans to expand its agent suite beyond cooling and power, into workload orchestration, renewable integration, and grid-responsive demand optimization.

The company will deepen partnerships with NVIDIA and Index Ventures, expand across Europe and Asia, and grow its data-science and control-engineering teams. The ultimate goal: make autonomous infrastructure a default expectation, not an experiment.

---