A robot can cost $2 or $35 an hour, depending on who you ask. The question that matters is how much useful work it can reliably complete.
Utilization decides the economics. In @Fortune, @Mark Minevich and I examine what it will take for Physical AI to turn technical progress into dependable labor.
My advice to founders building in Physical AI:
1. Sell labor, not robots. Get paid for the work itself. In industrial automation, that might be pallets moved, trailers unloaded, and parts sorted.
2. Treat utilization as the business model. An idle robot is an expensive robot.
3. Start with a wedge. Pick one high-volume task, like moving pallets, and get it working on a factory floor.
4. Build for real-world conditions. Robots need to operate reliably in factories, warehouses, and other customer environments.
5. Let deployments compound. Every hour in the real world improves the model, expands its capabilities, and lowers its cost.
Sell labor, not robots. Get paid for completed work: pallets moved, trailers unloaded, and parts sorted.
I believe Physical AI will usher in the era of Labor as a Service and change how work gets done across a global labor market worth more than $30 trillion.
The winners will be judged by how much work their robots complete.
Google announced a $17 billion investment to build its first data center in Wyoming. The project includes a planned 702-megawatt fuel cell development with carbon capture and storage, alongside commitments to cover its electricity and infrastructure costs. Click here
A consortium of major financial institutions announced a $63 billion financing arrangement to support Anthropic’s AI infrastructure through Broadcom. Bank of America, Blackstone, Citi, and Morgan Stanley are backing a financing structure for AI accelerators and networking equipment to be leased to Anthropic, with deliveries expected to begin in 2027. Click here
NVIDIA and Oracle announced plans to expand Vera Rubin AI infrastructure to multi-gigawatt scale. Oracle also made NVIDIA Vera Rubin NVL72 systems and Vera CPUs generally available on Oracle Cloud Infrastructure to support frontier model training and agentic workloads. Click here
OpenAI launched its Grid Protectors Program to help AI data centers support electricity grid reliability. The initiative will work with utilities and regulators to reduce compute campus electricity demand during periods of grid stress. Click here
Enterprise
OpenAI introduced GPT-6 Cyber, a specialized model for advanced cybersecurity research. The model improves vulnerability discovery and patch development, with access initially limited to approved security teams. Click here
OpenAI and ServiceNow expanded their partnership to bring AI agents into enterprise workflows. Planned integrations will connect ChatGPT and Codex with ServiceNow’s IT, security, and application development workflows, while expanding OpenAI adoption across ServiceNow. Click here
OpenAI and Oracle expanded their partnership to bring enterprise applications directly into ChatGPT. The companies are developing integrations that will allow employees and AI agents to access Oracle business data and execute approved workflows through conversational interfaces. Click here
Box partnered with NVIDIA to introduce an AI-powered enterprise content retrieval system. Built on NVIDIA Nemotron embeddings, the platform is designed to help agents retrieve information from corporate documents more efficiently. Click here
Perplexity launched Teammates, a new system of always-on AI agents designed to collaborate on workplace tasks. The agents can manage multiple assignments, access shared project context, and coordinate work through Perplexity and Slack. Click here
Capital Flows
Anthropic filed its S-1 registration statement with the SEC, formally advancing its plans for an initial public offering. The filing provides investors with detailed disclosures on the frontier AI company’s financial performance, infrastructure commitments, and operating risks. Click here
Marvell announced a $115 billion agreement to acquire optical networking company Lumentum. The all-stock transaction would combine electrical and optical technologies to address growing AI data center connectivity requirements, subject to regulatory and shareholder approvals. Click here
Dropbox announced a $350 million acquisition of Tigris Data to expand into AI infrastructure. The acquisition will support DBX Agent Cloud, an object storage platform designed for developers building AI models and agents. Click here
Shield announced a $5.3 billion agreement to acquire radar technology company Echodyne while raising $2.5 billion in financing. The transaction would expand its autonomous defense systems business, with the accompanying funding including a $1 billion Series H at a $28.8 billion valuation. Click here
AI inference startup fal raised $1.3 billion at a $16 billion valuation, doubling its valuation since April. The financing was led by Bessemer Venture Partners, with participation from NVIDIA, Sound Ventures, and T. Rowe Price. Click here
Research
Independent researchers found evidence that AI may have doubled the pace of capability development at Anthropic. METR’s assessment estimates significant acceleration in Anthropic’s AI research and development, while noting uncertainty and limitations from redacted evidence. Click here
OpenAI demonstrated an autonomous cybersecurity research pipeline that reproduced 1,724 Linux kernel vulnerabilities and developed working exploits for 948 of them. The project also generated candidate fixes and coordinated disclosure with software maintainers. Click here
Anthropic published research showing how Claude is accelerating drug development across 6 life sciences partnerships. In one collaboration, BigHat Biosciences used Claude to help design and test approximately 1,010 antibody variants in 10 weeks, demonstrating AI’s growing role in laboratory research. Click here
OpenAI demonstrated how GPT-6 and Codex can accelerate hardware engineering, simulation, and product design. Its research included AI-assisted circuit board and mechanical design, as well as collaborations with Mazda and Chip Ganassi Racing. Click here
Policy
The 2026 midterm elections delivered a major setback for AI data center expansion, with candidates supporting tighter restrictions winning key races. An Associated Press analysis found that 32 of 34 winning gubernatorial candidates supported stronger oversight, while voters approved restrictions in 44 of 45 completed local referendums. Click here
OpenAI published a National Community Compact and policy blueprint for responsible AI infrastructure development. The commitments address electricity costs, water usage, public transparency, employment, and community participation in compute-campus construction. Click here
The U.S. Justice Department is examining the competitive implications of NVIDIA’s proposed $12.93 billion acquisition of Hugging Face. Officials have questioned whether the deal could give NVIDIA significant influence over AI model distribution and competing technology providers. Click here
The White House plans to clarify the federal regulatory framework governing U.S. AI developers later this month. Pentagon CTO Emil Michael said the administration’s Super Intelligence Task Force will outline the existing laws and regulations that apply to AI labs as policymakers seek a more coordinated approach to oversight. Click here
Global AI Strategy
China is reportedly withholding approval for large-scale imports of NVIDIA H200 AI chips, leaving approximately 600,000 chips in Hong Kong. Beijing is balancing demand for advanced compute against efforts to reduce dependence on U.S. semiconductor technology. Click here
France’s Mistral released the model weights for its 1 trillion-parameter Mistral Large 4 model, expanding Europe’s sovereign AI capabilities. The release allows enterprises and governments to deploy the model within their own infrastructure, strengthening alternatives to proprietary U.S. frontier models. Click here
South Korea secured a record $2 billion in foreign investment pledges across AI, semiconductor, and energy industries. The commitments from 7 global companies were announced at the Invest Korea Summit, with most investments planned outside the Seoul metropolitan region. Click here
Japan announced a 10 billion yen ($63 million) emergency cybersecurity package that includes AI-powered defensive capabilities. The government plans specialist teams to identify vulnerabilities, coordinate incident responses, and strengthen collaboration with international AI developers. Click here
Russia ordered privately operated data centers to reserve 10% of their capacity for potential government use. The decree requires operators to make reserved computing resources available within 24 hours when directed by authorities, strengthening state control over domestic digital infrastructure. Click here
Talent Signals
Each week, we spotlight key roles tied to the themes shaping this week’s AI headlines, connecting talent to the companies driving the news.
PlusAI is building Physical AI technology to power autonomous commercial trucking, bringing intelligent systems into real-world freight and logistics operations. The Mayfield-backed company is advancing autonomous driving through AI-powered perception, planning, and continuous learning, and is hiring across AI research, robotics, machine learning, hardware, software, and fleet operations. Click here
Gecko Robotics is combining AI and specialized robotics to inspect, maintain, and modernize critical infrastructure across the energy, manufacturing, and defense sectors. The company is hiring across robotics software, hardware engineering, field deployment, AI, and operations. Click here
Destro AI is building an intelligence layer for industrial robotics, coordinating machines, workers, and warehouse systems to improve throughput and operational efficiency. The company is hiring across robotics engineering, Physical AI orchestration, perception, software, and field deployment. Click here
You can see all the opportunities at Mayfield-backed AI companies here.
Social Signals
The most important conversations in AI are unfolding across social media, where top voices are shaping the next wave of signals and strategy. Here are some of the top social signals and their takes from the past week.
Gavin Baker (Click here) — “The governor-elect of Texas, Gina Hinojosa, has explicitly said that all currently permitted data centers in Texas will be allowed to finish construction. This essentially covers all of 2027 and likely well into 2028.” In a post viewed 15M+ times, Baker argues that the election result does not mean an immediate halt to the AI infrastructure buildout, but could reshape the economics of new projects by introducing requirements related to power costs, water, taxes, and labor. The broader signal is that data center expansion is becoming a political and community issue alongside a compute and capital one.
Rikiya Takehi (Click here) — “We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B’s reasoning loops and improving math up to 6x and even instruction following 3x, without gradients or training.” In a post viewed 3.3M+ times, the MIT researcher introduced “Weight-Space Programming,” demonstrating a frontier model that directly reads and modifies the weights of another neural network. If the approach generalizes, AI systems may increasingly help improve other models directly, opening a new path beyond conventional training and fine-tuning.
Aaron Levie (Click here) — “If you could have done what you’re doing before AI, you’re likely not pushing it enough.” Levie argues that the real benchmark for AI adoption is not making existing work incrementally faster, but enabling work that was previously impractical or impossible. For founders and enterprises, that means redesigning products, research, engineering, and workflows around entirely new capabilities rather than simply layering AI onto existing processes.
