A couple of recent events, highlighted in the media, gave me an opportunity to reflect a bit more on something we all know, that scarcity can be a gift. It forces us to innovate. When there are little or no resource constraints, we brute-force way to the future. We lose the true innovative spark and think only about scale and cost efficiencies that come with scale. We see this in action in all kinds of arenas. A recent military example is of course the Ukrainians' ability to stand up to might of the Russian military. Innovation is not a function of the size of your budget (cf. US Military), abundance builds bloat. As I write in last week’s article about the Voyager missions, downsizing the plan from The Grand Tour and limiting the budget probably pushed NASA and JPL to eke out a lot more from the resources they were given.

Both Moonshot AI and Skyroot Aerospace are examples where the legacy approaches (of OpenAI or SpaceX) are not possible paths for these upstarts. Skyroot cannot afford a 100 million dollar launch complex, and it cannot just build a cheaper rocket - so it stops building rockets the way they have been built before. When a Chinese AI lab is constrained by geopolitics and cannot scale dense models across unlimited advanced compute infrastructure, it stops scaling the models the same way it has been scaled by the leaders. Scarcity is a forcing function that drives them down new paths. It is the true economics of scarcity: not doing more with less (though that is needed too) but a story of doing something structurally different because some choices have been removed. The best way to build the future is to sometimes be denied the budget (even while the DoD asks for $1.5T for 2027!).

Skyroot launched Vikram-1, India’s first commercial space launch. Skyroot is following a radical simplification path (echoes of Bernard Motissier touched on here). Liquid engine rockets require massive ground infrastructure (implies → huge capital outlay), while solid fuels rockets are cheaper and reliable but lack orbital precision. By opting for a hybrid architecture with a liquid-fuel upper stage you get the low manufacturing cost and operational ease of solid boosters combined with the fine-tuned directional control of liquid engines.

Furthermore, precision metal alloy tooling and machining require capital-intensive manufacturing; Vikram-1 features all-carbon composite airframes. by choosing the lighter structural components not only is the mass reduction benefiting them but heavy manufacturing costs are avoided. Don’t expect Skyroot to compete with SpaceX, but for the cost-conscious customer whose specific needs are met by Skyroot’s offering, this will be a boon.

Moonshot Ai announced Kimi-3 which has nearly caught up to frontier models from leaders such as OpenAI and Anthropic who are, orders of magnitude, much better resourced. . Facing hardware access constraints Moonshot couldn’t rely on brute force hardware scaling. Instead they focused on innovations at the algorithmic and architectural layer. Without boring you with details, their approach essentially gets you the deep knowledge capacity and reasoning depth of 3-trillion-parameter brain through an inference time engine that behaves like a much smaller and nimbler model (details here) that makes long context reasoning viable without blowing up your budgets.

The layers (from our pace layering viewpoint) at which these upstarts have chosen to innovate are hardened layers for established players; it is much harder for the established players to innovate at their slow moving layers. Constraints are not an impediment to progress, they can be the catalysts for massive innovation. Are you rethinking your constraints or are you hamstrung by them?