The 100GW alliance and the rise of internal agent armies

The intelligence race is increasingly a race for physical power and vocal presence.

Today’s issue looks at the new 100GW energy alliance between Google and Nvidia, Meta’s push into outbound voice agents, and how labs are deploying internal agent armies.

Google, Nvidia, and Anthropic have formed the AI Energy Management Alliance to secure 100 gigawatts of power. The coalition aims to solve the energy bottleneck by using software rather than physical expansion.

Traditional data centers rely on diesel generators during grid stress. This new alliance plans to use dynamic demand response, algorithmically shifting non-critical computing tasks to regions with spare capacity.

By tapping surplus headroom during off-peak periods, the industry hopes to bring massive compute clusters online years faster. The move signals a shift toward software-driven energy management for the scaling race.

The partnership with grid provider Emerald AI highlights that power, not just hardware, is the primary limit for frontier labs. Securing this capacity is essential for the next decade of AI growth.

Founders context: The physical limits of the power grid are now a software problem. Expect compute availability to become more localized and time-dependent. If your workflows are energy-intensive, look for providers that offer dynamic pricing tied to grid load.

Conversational assistants are moving from text chat to telephone calls. Meta and startup Instinct have both launched outbound calling capabilities, enabling digital agents to interact with the offline economy.

The new services can dial local businesses that lack digital booking systems. Agents are now handling dental office waitlists, restaurant reservations, and utility customer service phone trees without human intervention.

Meta’s Muse agent and the Instinct Concierge service are pushing AI assistants into a realm previously reserved for humans. This move bypasses the need for digital APIs for millions of small businesses.

As digital agents gain a voice, the barrier between online automation and local services is dissolving. This distribution shift allows AI tools to execute complex tasks in the real world.

Founders context: Your agents no longer need an API to take action. Outbound voice calling opens up the millions of service businesses that still operate via phone. Test these "voice-first" workflows for scheduling and simple procurement tasks.

OpenAI and Anthropic are now running thousands of concurrent autonomous agents internally. These digital workers are being used to write software, track research experiments, and monitor system logs at scale.

This heavy internal adoption provides a blueprint for the high-efficiency company. By "eating their own dog food," the labs are finding where agents excel and where physical bottlenecks still anchor progression.

While the move has reignited debates about recursive self-improvement, researchers note that economic and scaling costs remain a check. The internal efficiency gains, however, are delivering measurable results in developer output.

The labs are effectively becoming the first large-scale automated organizations. This shift demonstrates that the primary value of agents today is in augmenting complex, internal engineering and operations workflows.

Founders context: Do not just build agents for your customers; build them for your team. The frontier labs are proving that the biggest short-term ROI is in automating your own internal research and engineering loops.

📡 Google-Flipkart In-Chat Checkout: Google is testing native "Buy" buttons inside Gemini for Flipkart shoppers in India, turning conversational search into a direct commerce platform.

📡 Anthropic's $11.6B Akamai Bet: Anthropic has committed billions to secure general-purpose CPU capacity, signaling a massive infrastructure shift toward agentic retrieval and code execution.

📡 Nscale $3.36B Raise: neocloud provider Nscale secured billions in convertible debt led by Nvidia, positioning the company for a $35B public debut on the NYSE.

📡 Noam Brown on Sandbox Breaches: OpenAI’s reasoning lead warned that models are increasingly capable of exploiting sandboxes, highlighting rising security stakes for autonomous systems.