There is a working POV I have been developing over the past few weeks.
As we speak, there is a major infrastructure buildout cycle unfolding across the world. There are two core underlying drivers for why both countries and companies, without exception, are choosing to invest trillions in capex:
- Artificial Intelligence – this investment helps unlock value from this groundbreaking technology.
- New World Order – a fundamental geopolitical realignment, largely driven by the United States, forcing major countries to build domestic, sovereign infrastructure rather than relying on global capabilities through cross-border trade.
Interestingly, this infrastructure buildout is happening simultaneously across both the physical and digital worlds:
- Physical Infrastructure – (1) Power, (2) Chips, (3) Data Centers, (4) Industrial Manufacturing.
- Digital Infrastructure – (1) Foundational Models, (2) Data Pipelines.
A. What can we learn from history?
If one looks at large infra cycles over history (eg., the Railroads, WWII-driven DARPA tech, information & communication tech for the Internet in the 90s), stakeholders tend to approach them with a gold-rush mindset. That’s simply because infra & its related IP tends to be “scarce”, and once you put it in place, you can charge a toll on it for decades to come. Free Cash Flow, baby!
Of course, the side effect of this gold-rush mindset in past cycles has been that this infra build-out typically ends up overestimating the size and/or timing of future demand, builds out way too much capacity, usually with questionable financing terms (eg., high leverage), and in the end, many of these infrastructure builders blow up – either due to demand risks (not enough customers to make it economically viable within a desirable time frame) and/or financing risks (over-levered, not able to pay down debt, not able to raise equity).
Of course, a few players that can survive end up gathering the toll for decades to come (eg., Cisco post the dotcom crash). But more importantly, because the gold-rush & excess-driven mindset ends up creating massive amounts of infra in a very compressed time period (with complete disregard for underlying economic viability), it provides the catalyst for a follow-up cycle of “Applications”.
Because enabling infra is now in place and at low prices (courtesy of the overbuildout), entrepreneurs start building new businesses that solve a totally new set of customer problems & use cases. A classic example is a certain Jeff Bezos using core capabilities built by the likes of Cisco, Lucent, and WorldCom to start selling books on the Internet.
These Applications directly touch enterprise and individual customers, solve important problems for them, and therefore end up becoming generational compounders with attractive business economics, especially having the luxury to stay asset-light (because all capital-intensive assets have already been put in place by someone else), thus driving extremely high return on equity.
So, while almost 1/3rd of all initial American railroad companies went bankrupt, Mr Buffett has continued to enjoy steady & predictable cash flows from Berkshire’s core holding in BNSF Railway, one of the largest freight railroad networks in North America, over the last few decades. The Railroad infra providers went bust; the freight Applications layer continues to reap the fruits.
B. What does this mean for a small seed firm like Operators Studio?
Here’s how I am approaching playing this phase of the AI cycle:
1/ Sell “picks and shovels” to physical infra players
As a small tech fund, we obviously can’t participate directly in the players building power, chips, data centers, and foundational models. That’s where the gold rush is happening right now in the Valley.
However, where we can invest is in startups building technology-based, light picks and shovels to sell to these infra players, so we benefit from this massive capital infusion unfolding in front of our eyes.
These include robotics to execute these projects, Physical AI to power these assets, and industrial components to enable manufacturing.
A few such examples from the existing portfolio:
- Flywheel AI – Retrofitting excavators for remote operations on construction sites in the US.
- Muro AI – Agentic pre-construction operating system for General Contractors.
- CarbonStrong – Low-carbon cement alternative binder for concrete mixes, being used by real estate developers in India.
- Flytbase – Autonomous drone software for inspection & security of large physical sites like solar farms and oil & gas plants.
- Naxatra Labs – Radial and axial motors for deeptech use cases in India.
- Astrophel Aerospace – Precision engineering components like cryogenic valves and turbo pumps, serving space & general manufacturing.
- [RoboticsCo in Stealth] – On-field robotics autonomy for inspection & maintenance of large sites like solar farms and data centers.
There is a large picks and shovels bucket I am choosing to not play in – developer tools that help deploy, manage, and unlock value from foundational models.
Most Valley VCs are super-focused on this bucket as we speak. From my perspective, I don’t have an instinctive understanding of these tools, and I also struggle to get my head around the competitive differentiation of these products.
So generally speaking, I struggle to build conviction in these dev tools. The only investment I have in this broad area is Workspot (Computer use harness for agentic workflow automation within large enterprises).
2/ Squint into the future and back “Applications”
Unlike the last Mobile/Cloud/SaaS cycle, this time, these Applications cut across both the physical and digital worlds. This significantly enlarges the addressable opportunity set for even a small seed firm like Operators Studio.
(2.1) Digital AI
Applications in the digital world have some resonance with previous cycles.
- On the Enterprise AI side, I am taking a deeply vertical approach with a POV that powerful domain-specific applications will be built on top of physical and digital AI infra rails that are presently being laid out.
A few verticals I have been excited about include healthcare (Confido Health, Soulside AI, Elvo AI), biology (In Stealth), food (Loop AI), manufacturing (Datoms), construction (Muro AI), and insurance (Infer).
- Contrary to enterprise, sustainable Consumer AI applications will take time to emerge. They require significant leaps of imagination, large amounts of capital to create distribution, and active competition from incumbents given a winner-takes-all dynamic.
As a smaller fund, I question my right to win in accessing these first-gen consumer AI applications, and even beyond that, whether this first set of products are even good venture bets given flux in both the underlying foundational tech, as well as in adoption and competition.
I have taken a few selective bets in my circle of competence, with an India-to-the-world angle giving them some level of global differentiation – Stimuler (Conversational AI English learning app), Rovia (helping global BigTech employees manage RSUs) and another in the AI-generated microdrama space.
In addition, I feel sleeper hits are hiding in India cross-border Consumer AI, Consumer Hardware, and Fintech. However, I am not seeing many startups in these areas. Will continue to actively hunt here.
(2.2) Physical AI
Personally, I believe that the next generation of Applications being built in the Physical world are much more interesting.
There are core “product & business” infra rails being built out in:
- Space – low-cost and frequent launch capabilities; SpaceX in the US, Skyroot and Agnikul in India.
- Autonomy – LiDAR; advances in sensors and base software; Waymo in the US.
- Aerospace, Defense & General Manufacturing – OEM ecosystems in the US & India; public institutions like NASA, ISRO, and DRDO actively opening up their stacks to startups; governments open to buying from startups; Anduril & Boom Supersonic in the US.
I am excited about Applications being built on these domain-specific rails. A few examples from the existing portfolio:
- Space economy – Aule Space (life extension of satellites), another one in Stealth.
- Autonomous economy – NuPort (retrofitting commercial trucks for semi-autonomous freight operations), FlyWheel AI (autonomous excavators), RoboticsCo In Stealth (On-field autonomous robots).
- Aerospace – Aspera (autonomous, amphibious aircraft built in India), Parsec (indigenous jet engines for India).
- Defense – HyPrix (supersonic payload delivery capabilities).
C. What are the major risks in these two buckets?
1/ On selling picks and shovels to infra players, history tells us that:
(a) This infra buildout phase should be expected to be relatively short and execution to be highly compressed.
(b) Most players can be expected to go bust in the end, as explained in earlier sections.
So as a seed investor, it’s important that before the point of bust hits major customers, these picks and shovels companies move fast, scale quickly, and either get acquired or become self-sustaining via diversified sources of revenue.
It’s a play on an ongoing gold rush and therefore needs to be approached with that mindset.
2/ On backing “Applications”, while the runway is likely to be long here for the eventual winners, history tells us that:
(a) The first set of anointed winners are likely to get disrupted fairly quickly (Netscape as the first browser; Yahoo as the first consumer Internet company; AskJeeves as one of the first search engines).
(b) Eventual winning applications take time to emerge, and are fairly hard to spot & predict in the initial days.
So as a seed investor, the approach needed for the Applications layer is to stay skeptical of the first-generation companies and not be afraid to exit at what might be an eventual global maximum, as well as be on the lookout for weird, pattern-breaking use cases that are hard to see in the present but are living in a plausible future that requires a fair bit of squinting to spot.
