A $15M bet on data center energy

AI data centers are running out of power

AI data centers are wasting up to 20% of their electricity before it even reaches the GPUs.

Peak XV just led a $15 million round into Bengaluru based C2i, which is building a grid to GPU power system that claims it can cut losses by about 10%. BloombergNEF says data center electricity use could nearly triple by 2035, and Goldman Sachs estimates a 175% jump by 2030.

This is where the AI story gets physical. You can walk into a server room, hear the fans, feel the heat, and realize the constraint is no longer model size but how efficiently you step down voltage from 800 volts to something a chip can use.

If C2i is right, saving 100 kilowatts per megawatt adds up fast at hyperscale. Energy becomes the margin lever after the GPUs are bought and installed.

The risk is execution. Power delivery is entrenched, qualification cycles are slow, and incumbents do not move aside easily.

We will know in six months when their first silicon comes back from fabrication.

If power is the choke point, AI’s next winners may look more like electrical engineers than model builders.

Anthropic draws a line with the Pentagon

The Pentagon wants access to Claude for all lawful military uses, and Anthropic is pushing back even with a $200 million contract on the line.

According to Axios, the Defense Department is asking OpenAI, Google, and xAI for similar commitments, and at least one has agreed while others signaled flexibility. Anthropic is holding to limits around fully autonomous weapons and mass domestic surveillance.

This is the first clear test of whether AI companies’ public safety policies hold up when real money and federal pressure show up. It is easy to publish principles on a website; it is harder when the contract is large and the customer is the U.S. military.

The Wall Street Journal reported that Claude was used in an operation targeting Nicolás Maduro, which means these tools are already inside sensitive workflows. That makes the policy debate less theoretical and more operational.

You can picture this discussion happening in a windowless conference room at the Pentagon, fluorescent lights humming overhead.

If Anthropic gives ground, the rest of the industry will follow quickly.

If it does not, Washington may start picking winners.

AI just reset Axios’ org chart

One Axios engineer redid a three week project in 37 minutes using AI agent teams.

Axios CTO Dan Cox says tools like Claude Code and OpenAI Codex now ship features in days, and the company cut its product and tech team from 63 to 43 people to adjust. Over two years, that group went from 90 to roughly half that size while output more than doubled.

That is not a lab demo. That is payroll. When a 12 month backlog can disappear in months, headcount math changes fast.

Cox says the new bottleneck is human adaptation, and I believe him, because most teams struggle to absorb even one major product shift per quarter without confusion and fatigue setting in. You can almost hear the Slack notifications stacking up as another update rolls out.

The part that stands out is contract strategy. Axios is avoiding software deals longer than a year because they expect to build tools themselves at near zero cost.

If this holds, mid level technical roles across media and beyond will shrink quietly while expectations rise for the people who remain.

Are other executives being this blunt internally?

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.