AI Labs and Physical Machines Are Where Investor Dollars Are Landing

Software velocity sets the benchmark

Lovable, the vibe-coding platform that lets non-engineers build software through natural language, disclosed an annualized revenue run rate of $600 million at the HumanX summit in Amsterdam, up from roughly $500 million just three months earlier, as TechCrunch reported ^[1]. Co-founder Fabian Hedin added that two-thirds of Fortune 500 companies now use the product, signaling that enterprise adoption is driving the acceleration, not just individual developers.

For founders raising a pre-seed or seed in any AI-native software category, this number matters less as a competitive benchmark and more as a market-signal anchor. It shows that institutional buyers have moved past pilots with AI development tools. The adoption curve is steep and the enterprise budgets are real. Investors evaluating your category will now expect you to articulate why your tool earns a durable place in that stack, not whether the market exists.

Physical AI draws serious institutional capital

Three hardware and deep-science rounds closed within days of each other, and together they sketch the contours of where patient capital is moving. BigHat Biosciences closed a $75 million Series C co-led by DFJ Growth and Premji Invest to advance AI-designed antibody therapeutics and its agentic protein design platform, as BioSpace reported ^[5]. The round supports a clinical-stage pipeline, meaning investors are not backing a research thesis — they are funding a program that has already cleared early human trials.

At the seed stage, New York-based Midcentury emerged from stealth with a $15 million round and launched what it describes as an egocentric dataset of more than 2 million hours of human behavior, alongside a cloud simulation platform called Matrix, according to Ventureburn ^[4]. The framing is deliberate: the company is positioning first-person human action data as the physical-world equivalent of the web scrapes that trained Signal Copilots. Whether that analogy holds commercially is unproven, but the check size and the stealth-to-launch trajectory indicate the founding team had strong conviction from investors before going public.

In China, Lexiang's founder Guo Renjie — who built his operational track record scaling Dreame's robot-vacuum division to over 1,500 people — has shipped a humanoid robot standing 88 centimeters tall and weighing under 13 kilograms for less than $1,300, with Ant Group leading a 500 million yuan pre-Series A round, as reported by Robot Belt ^[3]. The sub-$1,300 price point is the story: it pushes humanoid hardware below the threshold where individual businesses and eventually consumers could absorb the cost, which is a different market gravity than the high-cost industrial robots that have dominated the category.

Autonomous labs compress R&D timelines

Lila Sciences, a Cambridge-based startup, published results showing its AI-directed lab proposed, synthesized, and screened 2,942 catalysts for green hydrogen production in three months, ultimately identifying six high-performing material families, as the company reported ^[2]. That is a pace no human-staffed lab can match through conventional experimental iteration, and it is precisely the kind of result that reframes what deep-tech due diligence should look like.

For founders in biotech, materials science, or energy working on anything with a lab component, this is a double-edged data point. It raises the bar on what investors will call a meaningful experimental result — having run 200 experiments is no longer a differentiator if autonomous systems can run thousands in a quarter. But it also validates the category: if you can show your platform or data infrastructure enables this kind of compression, you are speaking a language investors in physical AI are actively learning.

What this week's rounds have in common

Across these five items, the rounds that closed share a structural feature: each company had converted some form of technical output — revenue, screened catalysts, clinical data, shipped hardware, a curated dataset — into a tangible signal before the check arrived. Lovable had revenue growing month over month ^[1]. BigHat had a clinical-stage pipeline ^[5]. Midcentury had a product ready to launch on the day of the announcement ^[4]. Lila had published experimental results ^[2]. Lexiang had a product on sale ^[3].

At the pre-seed and seed stage, you will not have clinical trials or $600 million in annualized revenue. But the pattern still applies. Investors in AI and physical-world categories are calibrating around evidence of conversion — from idea to data, from data to output, from output to customer or user signal. The gap between a deck and a working prototype has become a more consequential gap than it was two years ago.