The AI Cost Crunch and the Quest for Efficiency
As the industry grapples with runaway compute costs, giants like Meta get creative with tents, while startups face skeptical scrutiny and new agents aim to automate everything from commerce to maintenance.
The Infrastructure Squeeze
The astronomical cost of running large AI models is forcing a major industry reckoning. A deep dive from TechCrunch reveals a shift from pure performance chasing to a frantic search for efficiency and “guardrails” on spending. This scramble is playing out in physical infrastructure: Meta is reportedly using tents to slash data center construction costs, while AirTrunk is committing a staggering $30 billion to build 5GW of AI data centers in India. Not all projects are scaling up, however; investor Kevin O’Leary has agreed to halve the size of a massive planned Utah data center amid local opposition.
Agents on the March
AI agents are moving beyond chatbots into core business operations. Meta launched its Business Agent to automate sales and support directly inside Instagram and Messenger. On Apple’s turf, startup Poke has become the first AI agent approved for Apple’s Messages for Business platform. In the industrial sector, Shell is expanding its use of C3 AI agents to move from anomaly detection to fully-automated predictive maintenance across tens of thousands of pieces of equipment.
Startup Spotlight & Skepticism
The startup scene is a mix of bold claims and measured moves. Anthropic, ahead of its IPO, brushed off doubts about AI’s returns as it reported annualized revenue soaring to $47 billion. OpenAI’s Mira Murati is stepping back into the spotlight to “remind the market” the company exists. Elsewhere, skepticism abounds: a startup called Quilty claims it can predict a film’s success from a script, but early testers are deeply unconvinced by its results.
In Other News
- Apple’s WWDC next week is expected to feature a major Siri revamp and Apple Intelligence updates. The company also touted $1.4 trillion in App Store billings.
- Meta also rolled out an AI creator assistant on Facebook to help influencers parse their performance data.
- Airbnb’s Brian Chesky plans to launch a new AI lab, having previously held off on LLM partnerships.
- In robotics, Hello Robot released the fourth generation of its Stretch home assistant, betting on a domestic robot market.
- On the research front, the NSF renewed support for the MIT-led AI and physics institute, IAIFI, and MIT RAISE partnered with Georgia State on PATH, an initiative to expand industry-aligned AI training.
- A security incident at Dashlane showed how attackers downloaded encrypted password vaults by targeting large numbers of users.
Editorial Take: Today’s news paints a picture of an industry hitting a practical wall. The initial “tokenmaxxing” frenzy is colliding with the physics of energy, cooling, and capital. The response is a fascinating bifurcation: mega-investments in centralized compute hubs paired with extreme cost-cutting tricks (tents!), all while software pushes to make every dollar of compute work harder through autonomous agents. The era of easy scaling is over; the era of creative efficiency has begun.