Unlock Capital Investment Manifesto
March 2026
The Problem With $1T of Investment
In 2024, VC funding into AI cleared $100B in a single year. Stargate landed at $500B. Europe committed €200B to InvestAI. Andreessen Horowitz raised $15B across AI funds. Every fund in the world claimed an AI thesis.
Two years on, the results are in. The early-stage landscape is flooded with copilots, wrappers, and agents that are really just prompts. The majority of AI 1.0 wrapper companies are underperforming their entry rounds. The wave that was supposed to find the next Atlassian funded the 400th AI-enabled CRM instead. A significant share of that capital flowed straight back to Nvidia and the hyperscalers as compute costs.
The problem: a trillion dollars went into building the models, the chips that run them, and the software layer on top. Almost none of it went into the physical systems the world actually runs on.
How will we be any different?
The Smil Test
Vaclav Smil spent his career asking a question many economists skip: what does the world actually run on? Not metaphorically. Physically.
His answer: energy, food, materials, transport. These are the load-bearing columns. Everything else rests on them. Remove synthetic nitrogen and half the world's food supply collapses. Remove steel and cement and civilisation has no physical structure.
Then Smil turns to the internet, and his argument is deliberately uncomfortable. The internet is the most transformative communication technology in history. You could also remove it tomorrow and the physical world keeps running. Painfully, at pre-1990 speeds, but the mines still operate, the crops still grow, the hospitals still function. The internet is extraordinary. Is it essential?
The distinction matters: powerful things change what is possible. Essential things become what everything else depends on. The test is one question: could you remove it, or would the system cease to exist?
Everything built in the AI 1.0 wave sits on the same side of that line as the internet. Close every ChatGPT tab, delete every copilot, kill every SaaS wrapper: the world keeps running. AI 1.0 sits on top of physical systems. It does not replace the substrate beneath them.
We think AI is now crossing that line. We back the companies making it happen. AI woven into the operating layer of physical systems so deeply that removing it means the system stops. The fusion reactor whose plasma dynamics depend on an AI control layer: remove it and the reactor doesn't run worse, it doesn't run. The autonomous port whose decision architecture is AI-native end to end: remove it and operations halt. The biosensor network replacing a clinical care pathway: remove it and the pathway no longer exists.
That is Essential AI.
We Invest in Essential AI
Every investment starts with one question: does it pass the Smil Test? Most seed-stage companies have not yet made their system essential. What we look for is the trajectory: founders building toward essentiality, or building the enabling layer without which essential systems cannot function. The further along that gradient a company sits, the deeper its moat and the harder it is to commoditise.
Smil identified the substrates civilisation runs on: energy, food, materials, transport. Our three pillars map directly onto them.
Sovereign Infrastructure (Energy + Materials). Intelligence is capped by the energy that runs it, a physical limit no software engineers around. We invest in compute architectures, edge inference, simulation software, and the technology that unlocks the critical minerals the supply chain depends on. Energy, materials, and intelligence are linked. We treat one as the driver of the others.
Physical AI (Machines + Transport). The robot economy is arriving. Growing, mining, manufacturing, and moving physical goods is being divorced from human labour entirely. We back the full stack: the platforms specialised robots run on, the simulation that trains them, the operating systems that coordinate fleets, and the data infrastructure connecting physical systems end to end. The moat is real-world operational data that cannot be generated synthetically.
Bio AI (Food + Biology). AI is designing biological function that could not exist without it: proteins with no natural equivalents, enzymes with capabilities no organism evolved, drug molecules at a pace human chemistry cannot match, crop traits designed computationally that conventional breeding could never produce. The WHO projects a shortfall of 10 million health workers by 2030: AI is building the diagnostic and monitoring systems the care model will depend on. The moat is real-time inference data from living systems, the continuous signal from a specific patient or a specific field that synthetic data can augment but cannot replace.
The market confirms the direction. Robotics startups raised $13.8B globally in 2025, up from $7.8B in 2024. Figure AI's Series C raised $1B+ at a $39B valuation. Skild AI raised $1.4B at $14B. Physical Intelligence closed over $1B across three rounds in 20 months.
The People Building Essential AI
The founders at this layer are domain experts: researchers with decades in computational memory systems, physicists who authored foundational plasma models, engineers who have spent careers building for environments where failure is not a UX problem. They were working on these problems before generative AI. What changed is that AI capabilities multiplied their possibilities 10x, 100x, 1000x. These founders care about reach, not headcount or TED talks. An acquisition by a platform that puts their work in the hands of millions is the best possible outcome for this archetype.
Kalanick spent eight years building Atoms in total stealth, thousands of employees across food, mining, and industrial transport, before saying a word publicly. That is the depth of conviction this era rewards.
The wave has been building for years. It is now cresting, and the founders riding it have been in the water long before the crowd noticed the swell.
We offer them the funding and the operating backbone to stay there. We syndicate the first cheque and build the pathway to US capital. Then we hand them the same AI-native operating harness we built for ourselves: infrastructure, skills, automations, and improvement loops that lets a two-person team run like a tier-one operation. We want them to focus on the build.
Why Australia
Australia is the most structurally advantaged launchpad for Essential AI, for reasons that compound rather than merely reduce costs.
Capital efficiency is real: seed rounds price 30-60% below US peers, and Australia produces 1.22 unicorns per USD $1B of VC deployed, the highest ratio in the world and nearly double the US rate.
The research depth argument matters more. Australia ranks among the world's top performers for AI research quality per capita, particularly in robotics, embedded systems, and diagnostics. The founders we back are internationally cited researchers with PCT-filed patents, priced as if they're in Canberra rather than Cambridge.
The deployment argument is the one most funds miss. Australia has live testbeds that US founders cannot access: the world's most automated mining operations, precision agriculture systems, sovereign defence infrastructure, and a healthcare system willing to pilot at genuine speed. The same land, political stability, and renewable energy abundance that powered Australia's export economy now underpins sovereign AI infrastructure.
The 43.5% R&D cash rebate for companies under $20M turnover, the $1.6B Australia Economic Accelerator, ARENA funding, and NVCR turn capital efficiency from good to exceptional. Runway extends. Ownership doesn't dilute to fund salaries and AWS bills.
The Bet
We anticipate superintelligence. We are investing in what it needs to operate.
The AI 1.0 wave compressed enormous value into software companies that will be commoditised. The Essential AI era will create enormous value in companies acquired by Nvidia, Apple, Lockheed, Anduril, and the sovereign AI programmes of nation-states who have decided they cannot outsource the operating system of their critical infrastructure.
Acquisition is a primary exit, and it is structural. The platforms are accelerating consolidation, acquiring teams with irreplaceable domain knowledge before competitors do.
Google acquired the companies that built Android-era mobile infrastructure. Tesla has systematically bought sensor and autonomy capability. Nvidia has done the same across the AI stack. The pattern is consistent: buy the breakthrough, integrate, repeat. Essential AI companies are the breakthrough being bought. For the breakout companies, sovereign procurement mandates and critical infrastructure spend create a public-market pathway on their own terms.
The window is open. World-class founders, priced at a fraction of their US peers. The gap between what is being built here and what it costs to back is the opportunity.
The essential is being built as we type. We exist to unlock it.