Ignore the Chatbot Wars, AI is getting more physical than most think

Hello friends, and welcome to The Unsophisticated Investor! Brought to you by Scott & Rob, the founders of Shuttle!

If you want to invest alongside the VC funds who've backed breakout companies like Revolut, Asana, JustEat, Bolt, Lets Get Checked, Loom, Runna, Charlotte Tilbury, Deel, Aircall, AngelList, Carta, TransferWise and many more, regardless of your knowledge, network or net worth, you can sign up to our platform by clicking the link below.

Now, let’s get into it 👇

Last week, the Irish-founded ocean robotics company, Ulysses, closed a $38 million Series A led by Andreessen Horowitz, taking its total funding to $46 million. The company was founded in Dublin in 2023 by four engineers, Akhil Voorakkara, Jamie Wedderburn, and brothers Colm and Will O’Brien. The team started out in a small office on St Stephen’s Green before relocating to San Francisco. The origin story has some charm to it: Wedderburn got the idea on a surf trip to the west coast of Scotland, after a friend described a grim day of hand-planting seagrass shoots that were promptly wiped out by a storm. Two years on, Ulysses builds autonomous underwater and surface vehicles used by the US Navy, the Australian government, the Great Barrier Reef Foundation and a handful of American marine research agencies.

Last week's AI news cycle was busy with other things. Another round of the OpenAI-Anthropic rivalry, fresh model releases from both, and a new instalment of the argument the industry has been having with itself for the past year: what AI is going to do to white-collar work. Anthropic's Dario Amodei has been the loudest voice on the displacement side, telling Axios last year that AI could eliminate up to half of all entry-level white-collar jobs within one to five years, and doubling down at Davos in January. Sam Altman and OpenAI have taken the opposite posture, pitching their products as productivity tools that augment the existing workforce rather than replace it. 

Both framings are about AI inside an office. What is missing from much of the coverage is that the most heavily funded AI companies of the last 18 months are not, for the most part, office software. They are companies where the model is not the product. The product is what the model is strapped to.

The scale of the mismatch

Global venture funding hit roughly $425 billion in 2025, of which about $211 billion went into AI, per Crunchbase. Most coverage tracked a handful of frontier labs and their consumer applications. Robotics alone pulled in over $6 billion in the first half of 2025, on track to exceed 2024’s $7.5 billion total, and physical AI, embodied AI and industrial automation have become among the most heavily financed categories of the year.

Large language models are software talking to software. The input and the output both live inside a screen, which is part of why they can scale globally within hours of release. The other category of AI company is harder to see. It lives in drug discovery labs, on factory floors, inside electrical substations, and, as Ulysses is now demonstrating, on the seabed. Deployment cycles are measured in years rather than weeks. It rarely produces a viral demo. It is also where AI meets the parts of the economy that actually break when you ignore them.

Where the capital is actually going

The interesting thing about physical AI is that it has forced categories that previously lived in different venture worlds to share cap tables. Humanoid robotics is the most visible example. Figure AI has gone from a $2.6 billion valuation in early 2024 to $39 billion in September 2025, after a Series C that drew in Nvidia, Intel, Brookfield, Salesforce and Qualcomm. Physical Intelligence, founded only in 2024, raised $600 million in November at a $5.6 billion valuation and is reportedly already in talks for a round that would push it past $11 billion. Apptronik, Skild AI, 1X and Agility Robotics have all closed substantial rounds on similar logic: once foundation models can reason about the physical world, the value sits in the actuators, the sensors, the supply chain and the training data, not exclusively in the model itself.

This is not a story exclusive to Silicon Valley. Chinese manufacturers including Unitree, UBTECH and Fourier have pushed pricing on basic humanoid platforms below $10,000, with the Chinese Government backing both robotics manufacturing and the AI stack that runs on top. Galbot, a Beijing-based humanoid developer, raised $300 million in 2025 at a reported $3 billion valuation. Horizon Robotics, another Chinese company, has quietly raised billions across multiple rounds building AI chips for autonomous systems. The strategic view from Beijing, made increasingly explicit in industrial policy documents, is that the next industrial revolution will embody AI, and that manufacturing scale rather than model sophistication is the thing worth controlling.

Europe is further behind on humanoids but better positioned in adjacent areas. Germany’s Neura Robotics has raised over €1 billion to mass-produce its systems. Advanced Machine Intelligence, the Paris company co-founded by former Meta AI chief Yann LeCun to build so-called world models, raised $1.03 billion in what has been reported as the largest seed round ever for a European startup. Biology is quietly becoming a European strength too, with AI-native drug discovery businesses building on Alphabet’s AlphaFold platform; Isomorphic Labs, the DeepMind spin-out, signed partnerships with Eli Lilly and Novartis in 2024 worth close to $3 billion combined, and has now moved multiple candidates into preclinical development.

The energy layer underneath all of this is itself turning into a venture category. Commonwealth Fusion Systems raised $863 million last August at an investor list that included Nvidia and Google, bringing its total to roughly $3 billion. Google has separately signed on as a first power customer for its planned ARC plant in Virginia. The International Energy Agency reports that data centre electricity demand rose 17% in 2025, and that the pipeline of small modular reactor offtake agreements between nuclear developers and data centre operators jumped from 25 gigawatts at the end of 2024 to 45 gigawatts by late 2025. AI is, in effect, dragging new energy infrastructure into existence behind it.

The common thread across all of this is that the AI component is the commoditised element of the product development and deployment. The hard part is the physical deployment, the supply chain, the regulatory approvals and the long customer relationships. That is also where the moat lives.

Why hardware plus AI is structurally different

Software models commoditise faster than hardware does. The release of open-weight models from Meta, DeepSeek, Alibaba and others has repeatedly narrowed the gap between frontier closed models and freely available alternatives, and switching costs for a user to move from one chatbot to another are low. Hardware-plus-AI companies operate under different economics. Once a humanoid robot is deployed into an automotive plant, or an autonomous underwater vehicle is contracted to a navy, or an AlphaFold-derived platform is embedded in a pharmaceutical pipeline, the switching cost is high. Contracts run multi-year. Certifications are specific to the deployment.

This changes who the customer is. Defence ministries, utilities, pharmaceutical companies, automakers and infrastructure operators all sign long contracts and care about reliability more than novelty and are increasingly paying attention to the strategic implications of who builds the infrastructure underneath their operations.

How VCs are framing it

What is quietly happening inside the venture industry is a re-alignment of the thesis. The dominant frame of the previous decade, capital-light software scaling without touching atoms, worked well for a specific set of conditions that no longer hold. Andreessen Horowitz has put its version of the shift on paper through its American Dynamism fund, which led the Ulysses round, and which explicitly targets companies serving national interest: defence, aerospace, manufacturing, energy, industrial technology. Other firms have landed in similar territory under different labels: the techno-industrial thesis, re-industrialisation, deep tech, hard tech. The common thread is a recognition that the next decade of returns is likely to come from companies whose product touches the physical economy, and from a customer base where governments and industrial buyers sit alongside enterprises.

The structural consequence is that cap tables are now pulling in investors who would once have kept to separate lanes. A single round can include a chip maker, a telecoms carrier, a defence integrator, a climate fund and a sovereign wealth fund. That combination used to be incoherent. Now it is the signal.

What this means for private market investors

Two things follow. First, the public discourse on AI is a partial guide at best to where AI capital is flowing. The most heavily funded rounds are increasingly going to companies whose end users are not large consumer markets and whose progress does not surface in the news cycle. Second, the re-rating happens early. Figure went from $2.6 billion to $39 billion in roughly 18 months. Physical Intelligence went from founding to an $11 billion valuation inside 24. By the time a physical AI company is generating obvious public metrics, most of the repricing has already happened behind closed doors.

What we’ve been working on at Shuttle

• More commitments coming in as great momentum continues in our funding round 💰

• Reviewing our next set of deals for the community 🚀

• Built and deployed our own internal tools for our compliance operations 🧑‍💻

The Unsophisticated Investor is brought to you by Scott & Rob, the founders of Shuttle. We’re both sick of private markets being a playground exclusive to the ultra-wealthy so we started a company to challenge the status-quo. Shuttle’s singular focus is to unlock private markets for Millennial and Gen Z tech professionals and help them build wealth through the highest performing private market opportunities.

Scott & Rob
Shuttle Co-Founders