Physical AI is the next massive opportunity I see. We are in the golden era of systems, and over the past few years, hardware has eaten software.
Last week, I sat down with Vijay Chattha of VSC at the Dirty Jobs Summit to talk about why the messy, complicated, dirty corners of the economy are where the next iconic companies will be built.
Physical labor represents a roughly $30 trillion market. If 10% of labor moves to physical AI, that creates a $3 trillion opportunity – five times the size of the software industry.
Three forces are converging:
1. A massive market of jobs people don’t want to do, businesses can’t fill, or humans aren’t well suited to perform
2. Technology that finally lets machines perceive, reason, and act
Industrial automation will win first. Warehousing, manufacturing, agriculture, construction, logistics, and infrastructure inspection offer structured environments, acute labor shortages, repetitive workflows, and measurable ROI.
The market structure will likely be hybrid. Foundation models will provide transferable capabilities, while vertical companies will own specialized workflows, safety, reliability, and customer integration.
Over the next decade, I believe the number of physical AI startups that get funded, acquired, and go public will be orders of magnitude larger than anything we have seen in this area.
In this week’s news, AMD agreed to acquire World Labs for $8.2 billion, Anthropic found that robots can already perform roughly three-quarters of physical job tasks in some settings, CoreWeave put NVIDIA’s Vera Rubin systems into production, Japan launched a $140 billion push into AI data centers and power infrastructure, and Meta launched its new Enterprise Platform to bring Muse and its agent stack directly to businesses and developers.
Full Weekend Edition below. 👇
Signals Shaping the Future of AI:
Infrastructure
NVIDIA launched its Open Agent Safety Platform for governing autonomous AI agents. OpenShell creates secure runtime boundaries while NVIDIA Sentry adds hardware-level monitoring and control. Click here
CoreWeave put NVIDIA Vera Rubin NVL72 into production, with Cognition as its first customer. Cognition reported up to 4.8x higher token throughput for Devin workloads versus GB200 NVL72. Click here
Amazon signed a 20-year nuclear power agreement with Constellation covering 690 MW. The agreement supports a 190 MW expansion at Maryland’s Calvert Cliffs plant and more than $3 billion in infrastructure investment. Click here
Samsung committed $1 billion to KKR-backed Helix Digital Infrastructure. Helix is building AI data centers, power, transmission, and connectivity infrastructure, backed by NVIDIA, KKR, and the Kuwait Investment Authority. Click here
Oracle reportedly signed a $7 billion cloud agreement with Tencent for AI compute outside China. The five-year deal would give Tencent access to roughly 100,000 AI chips through Oracle-operated data centers in Southeast Asia. Click here
Enterprise
Meta launched Meta Enterprise Platform and hired former MongoDB CEO CJ Desai to lead it. The new business brings Muse, Meta Business Agent, Muse API, and Muse Code directly to enterprises and developers. Click here
OpenAI launched Dots, a new class of always-on AI agents. The agents are designed to proactively handle multi-step work and extend OpenAI further from chat into persistent autonomous assistance. Click here
Anthropic launched Claude Sonnet 5.5, running more than 30% faster while costing up to 30% less for most work. The model targets everyday enterprise, coding, and document workflows with performance approaching Opus 5.5 on professional tasks. Click here
OpenAI launched GPT-6.1 Sol for complex coding, computer use, and professional work at a lower cost than GPT-6 Astra. The model also adds multi-agent support in the API. Click here
Barclays is expanding Claude across software development, legacy modernization, and internal knowledge workflows. The bank expects Claude Code to reach 50% of its developer population by year-end. Click here
IBM made its Bob agentic development platform available in self-hosted, private-cloud, and air-gapped environments. The release is aimed at regulated enterprises that need to keep sensitive code and data inside controlled infrastructure. Click here
Capital Flows
AMD agreed to acquire Fei-Fei Li’s World Labs for $8.2 billion. The deal gives AMD a major position in spatial and physical AI research while bringing Li in as EVP and chief scientist. Click here
OpenAI is reportedly in talks to raise at least $30 billion at a valuation of roughly $1.4 trillion. Bloomberg reports the prospective financing follows the company’s decision to push back its expected IPO. Click here
AI-assistant startup Instinct raised $1 billion at a $10 billion valuation. The Series C arrived only about a month after a round that valued the company at $2.5 billion. Click here
ElevenLabs completed a $300 million employee tender at a $22 billion valuation. The transaction doubled its valuation from February as enterprise demand for conversational agents continues to grow. Click here
Armadin raised $255.5 million at a valuation above $2.5 billion. The Kevin Mandia-founded cybersecurity startup has now raised roughly $445 million to build autonomous offensive-security agents. Click here
Research
Google introduced Gemini 4 Argon for long-horizon coding, enterprise, and cybersecurity work. Google is initially restricting the model to trusted cyber defenders while it undergoes additional safety testing ahead of broader release. Click here
Anthropic found that robots can perform roughly three-quarters of physical job tasks in at least some settings today. Robots and LLMs together expose roughly 80% of U.S. work by time spent, although robot economics remain a major constraint. Click here
Anthropic found that China’s GLM-5.3 can autonomously build sophisticated end-to-end cyber exploits. Its testing also found that the model’s safeguards can be bypassed or removed relatively easily. Click here
Anthropic published “Claude-shaped science,” exploring how AI can contribute to quantitative scientific discovery. The work examines Claude’s ability to identify and extend reusable mathematical structures across scientific problems. Click here
Google Research introduced Diffusion Controller for steering image-generation models without retraining them. The framework adds lightweight control over diffusion models, including systems where developers have limited access to the underlying model. Click here
Policy
Six major AI companies signed a voluntary White House agreement covering frontier AI safety and oversight. Anthropic, OpenAI, Google, Meta, xAI, and NVIDIA committed to internal controls, independent audits and board-level review, although the agreement is not legally binding. Click here
The U.S. launched America.gov as an AI-powered interface for federal information and services. The system is intended to become a conversational front end for accessing government information and eventually transactional services. Click here
A bipartisan group of U.S. governors is organizing around state-level AI governance. Maryland Governor Wes Moore and Indiana Governor Mike Braun are working through the National Governors Association as states increase their role in AI and data-center oversight. Click here
Congressional scrutiny is growing over secret agreements used in AI data center development. House Judiciary ranking member Jamie Raskin requested information from Amazon, Google, Meta, and Oracle regarding NDAs with local governments that cover energy, water, incentives, and other project details. Click here
Global AI Strategy
JERA, Dell, and RHAELM are planning a national-scale AI infrastructure buildout in Japan that could ultimately reach roughly $140 billion. The first project in Chiba is expected to exceed $15 billion and provide up to 400 MW, with the partners aiming to replicate the model at a multi-gigawatt scale. Click here
The U.S. and China agreed to establish a bilateral AI safety dialogue and incident channel. The agreement brings AI risk management directly into the strategic relationship between the world’s two largest AI powers. Click here
Mubadala and Together AI partnered to expand AI infrastructure and sovereign capabilities in the UAE. Together, AI will establish a presence in Abu Dhabi as the two organizations explore infrastructure and ecosystem investments across the country. Click here
The UK government is exploring domestic alternatives to Palantir as it reassesses dependence on foreign technology providers. The government is examining alternatives for major NHS and defense technology contracts as part of a broader push toward domestic technology capability. 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.
Rhoda AIis building foundation models for Physical AI, training robots to perceive, adapt, and operate reliably across real-world manufacturing and logistics environments. The Mayfield-backed company is hiring broadly across robotics, AI, hardware, deployment, safety, and software engineering roles. Click here
Gritt is building Physical AI for construction, combining robotics and AI with equipment already used on job sites to automate labor-intensive infrastructure work. Fresh off a $32.4 million financing backed by Obvious Ventures, Union Square Ventures, First Round Capital, and others, the company is hiring across robotics, AI, mechanical engineering, hardware, software, and field operations. Click here
Array Labs is building advanced radar satellites and sensing systems designed to continuously understand changes across the physical world. Following $41 million in financing announced this year, the Redwood City company is hiring across RF, electrical, and mechanical engineering, embedded software, computer vision, radar algorithms, and hardware. 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.
Jensen HuangClick here — “This is bigger than a single product. It’s the beginning of an open ecosystem to build the trust layer for safe agent systems.” Huang argues that increasingly autonomous agents will require enforceable security boundaries outside the model itself. NVIDIA’s Open Agent Safety Platform combines runtime controls, monitoring, and hardware-backed containment, pointing toward agent security becoming a new infrastructure layer as enterprises give AI more authority.
Andrej Karpathy (Click here) — “As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding. Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, disposable software artifacts that would have never made sense to create before.” Karpathy argues that the interface for AI output is moving beyond text toward diagrams, interactive web pages, and eventually bespoke explainer videos generated for a specific question or task. His broader thesis is that as intelligence and code become abundant, humans will spend more time interpreting and supervising AI output while software itself becomes increasingly disposable and generated on demand.
Ethan Mollick (Click here) — “As a heads-up: every firm’s customer service agents are about to be overwhelmed with Dots and Muses and other Clawlikes negotiating for better deals using voice/chat channels made for humans. Stories of people delegating this sort of work to their agents & saving money as a result are popping up and are going to only go more viral.” Mollick argues that always-on personal agents could overwhelm customer-service systems built around human friction, because agents can wait, retry, navigate phone trees, and negotiate indefinitely. His broader point is that businesses may soon need to redesign customer-facing workflows for agents as well as people.
