Amram Capital · Research

Thematic Note № 07
June 2026

The Robotics Inflection

01 — THE INFLECTIONWhy this time is different

The breakthrough is the vision-language-action model. Where a traditional industrial robot executes hard-coded motion paths, a VLA model perceives a scene, interprets an instruction in natural language, and generates the action directly. Google DeepMind opened the field with RT-2 and continues with Gemini Robotics; NVIDIA has commercialised it through its GR00T foundation models and the Isaac and Omniverse stack; Figure runs its own Helix system. The effect is to convert robots from machines that must be programmed into machines that can be taught, and increasingly trained on synthetic data generated in simulation rather than expensive real-world trials.

The second leg is cost. Goldman Sachs notes manufacturing costs for humanoid platforms fell roughly 40% in a single year, against an expected 15 to 20% annual decline, and sees a further 30 to 50% reduction by the end of the decade. Unitree already sells a functional bipedal robot near $16,000; Tesla targets $20,000 to $30,000 for a far more capable Optimus. When capability is rising on a software curve and cost is falling on a hardware curve at the same time, adoption stops being linear.

The constraint on robotics was never the body. It was the brain. That is the wall now coming down, and it is why the curve bends.

The honest caveat is that the hardest problem remains unsolved. Dexterity, the fine manipulation a human hand performs without thought, is still poor. The roboticist Rodney Brooks, who has built these machines since the early 1990s, argues deployable dexterity will stay weak well into the 2030s. So the realistic near-term map is narrow industrial tasks now, broadening capability through the late 2020s, and true general-purpose usefulness later and less certainly than the bulls imply. That distinction, between what is deployable today and what is promised for 2040, is the spine of everything that follows.

02 — THE LANDSCAPEThree layers, one supply chain

Treating robotics as a single trade is the first mistake. It is at least three distinct opportunities with very different risk, plus a cross-cutting layer that supplies them all.

The proven core: industrial automation

This is the part that already works and already earns. The industrial robotics market is roughly $65bn in 2026 and compounding in the high teens, with more than four million robots already operating worldwide. It is a consolidated oligopoly: Fanuc, ABB, Yaskawa, KUKA and Keyence in sensing, names with real moats, pricing power and decades of installed base. The physical-AI wave is a tailwind here, not a threat: these incumbents are layering VLA capability onto hardware that customers already trust, and ABB is spinning out its robotics division as a standalone listed company, a signal that dedicated capital sees value to crystallise.

The quality compounder: surgical and specialised robotics

Vertical robotics that solves one high-value problem extremely well is, in business terms, the most attractive corner of the theme. Intuitive Surgical is the archetype: roughly 60% share of robotic surgery, an installed base that compounds high-margin recurring revenue with every placement, and switching costs measured in surgeon training years. It is also where competition is finally arriving, from Medtronic, Johnson & Johnson and CMR Surgical, which is precisely why it now trades at a discount to its own history. Adjacent names in spine and orthopaedics extend the same razor-and-blade logic.

The frontier: humanoids and embodied general intelligence

This is where the imagination and the danger both live. The humanoid market is small today, around $5 to 6bn, and the forecasts are extraordinary: Goldman sees $38bn by 2035, a figure it revised up sixfold in a year, and Morgan Stanley models a $5 trillion ecosystem by 2050 implying close to a billion units. Goldman expects only 50,000 to 100,000 units shipped in 2026. So this is venture-stage public equity: a credible structural thesis attached to companies that are, for the most part, pre-profit stories.

The shovels: the enabling layer

Underneath all three sits the supply chain every robot needs regardless of which brand wins: AI compute and edge processors, sensors and LiDAR, harmonic drives and precision actuators, and the rare-earth magnets at the heart of every motor. Morgan Stanley alone sizes the humanoid semiconductor opportunity at $305bn by 2045. Goldman's explicit conclusion is that, for now, the best risk-adjusted opportunity lies in component makers rather than the robots themselves. We agree, and it is the organising idea of our playbook below.

03 — THE NUMBERSBelieve the direction, discount the decimals

The forecasts above should be read as direction, not precision. The spread between research houses is enormous: independent estimates of the humanoid growth rate range from the high teens to nearly 70% a year, a fourfold disagreement that tells you the analysts are modelling a future none of them can see. A $5 trillion number for 2050 is a statement of conviction dressed as a spreadsheet. It is useful for sizing ambition and useless for sizing a position.

Our posture is therefore deliberately asymmetric. We treat the industrial and surgical numbers, which rest on real installed bases and observable order books, as a basis for sizing real positions. We treat the humanoid numbers as a reason to hold optionality, not as a valuation input. The further out and larger the forecast, the smaller the cheque it should justify today.

04 — CHINAThe gravitational centre

No serious robotics thesis can ignore China. It installed roughly 276,000 industrial robots in 2025, about 70% of the global total and up from 52% in 2020, more in a single quarter than most countries manage in a year. Domestic manufacturers have taken around 40% of their home market by volume, up from 27% in 2020, and the 15th Five-Year Plan places robotics and embodied AI at the centre of national industrial strategy. The supply-chain density of the Yangtze River Delta, where component makers, robot builders, AI labs and EV manufacturers sit within a couple of hours of each other, is a genuine and underpriced advantage.

This is the double edge. China is where much of the manufacturing scale and cost compression will originate, which is bullish for the theme and for the enabling layer. It is also where the geopolitical risk concentrates: export controls on advanced chips and tooling, tariff regimes, delisting and audit risk on US-listed Chinese names, and the simple fact that the West may choose, for security reasons, to build a parallel and more expensive supply chain. Exposure to China in this theme is both the alpha and the tail risk in the same position.

05 — THE FRAMEWORKSWhere the disciplines disagree

It is more honest to surface the tension between the frameworks we respect than to pretend they point one way. They do not.

We resolve the tension by deciding which lens governs which sleeve. The quality and cycle lenses govern the core, where real money is deployed into cash-generative businesses bought with patience. The macro and disruption lenses govern the satellite, where a deliberately small allocation buys convexity on the frontier. The portfolio lens governs the whole, as a reminder to count the correlation and the currency. The mistake is letting the frontier set the size of the core position.

06 — THE PLAYBOOKHow we would express it

The structure is a quality-weighted barbell. Lead with the shovels and the compounders, treat humanoids as a small option, and buy only the exposure a diversified technology book does not already contain. The single most common error we see is buying a cap-weighted robotics fund that simply re-purchases the Nvidia and Tesla an investor already owns elsewhere.

On vehicles, the choice is between breadth and conviction. Diversified robotics and automation ETFs deliver the quality core and the automation names that are awkward to buy individually. A pure humanoid fund delivers the frontier in a single line, at the cost of concentration, youth and, often, heavy China weight. Single names earn their place only where the moat or the scarcity is specific and identifiable, surgical robotics and rare-earth magnets being the two clearest cases. We would express the frontier through a fund rather than a single humanoid manufacturer, because at this stage the identity of the winner is genuinely unknowable and equal-weighting the value chain is the more honest posture.

Execution discipline

Three rules govern entry. First, valuation gates size: with broad equity valuations stretched, we build through tranched limit orders anchored to drawdowns from recent highs, not calendar-based deployment, and we do not chase strength. Second, weakness is the friend of the patient: the one place this discipline is already satisfied is quality names that have de-rated on competitive fear rather than deteriorating numbers, which is where we would begin. Third, count the currency: almost every instrument in this theme is denominated in dollars, yen or yuan, so a robotics sleeve quietly adds to an existing dollar long, and that exposure should be sized deliberately rather than acquired by accident.

07 — RISKS & TAILSWhat breaks the thesis

08 — CONCLUSIONThe house view

Robotics has crossed from possibility into investability, and the direction of travel is not in serious doubt. Physical AI is a real inflection, the labour-market prize is vast, and the supply chain is maturing quickly. None of that obliges us to overpay for the most exciting and least proven end of it.

So we lead with the shovels and the compounders, the parts of the theme that earn today and survive a disappointment, and we hold the humanoid future as a small, convex option entered on weakness rather than as a leveraged bet on a 2040 forecast. In a gold rush the reliable money has always been in the picks, not in the prospectors, and the prospectors here are still pre-profit. That is not caution for its own sake. It is the most asymmetric way we can find to own a theme we genuinely believe in.

Own the inflection through cash flows and the supply chain. Hold the dream as an option, sized so it can fail without taking the book with it.