Deepseek, an AI lab from China, doesn't have access to nVidia's latest chips because of export restrictions. Yet they created a 'DeepSeek moment'—delivering a state-of-the-art model with limited hardware and dramatically lower training costs.

Moonshot, another Chinese lab, just released their Kimi K2 thinking model, which is beating OpenAI and Claude on key benchmarks. They use INT4 instead of FP8 quantization to run inference faster on cheaper hardware. And it is fast.

For the technically inclined, the Kimi K2 technical paper is a great read.

Constraints Are Reality

Every executing team is resource-constrained. More people, more budget, more time—it's the universal ask. The trickiest part of any business model is delivering value while being constrained across all three dimensions.

So what are the positive aspects of constraints?

"Necessity Is the Mother of Invention"

Cliché as it sounds, constraints necessitate innovation. Just as Chinese AI labs developed new algorithms to overcome hardware limitations, we must innovate our way through constraints.

"Fail fast, fail often" remains the most reliable approach. As much as HBR, MIT Technology Review, and other management gurus claim there's a method to innovation, the loop of experimenting, failing, and repeating until you succeed is still the best way forward. Frameworks like Jobs to Be Done and Design Thinking are useful but limited—they're being rapidly transformed by AI's prototyping capabilities anyway.

Artificial Constraints

Sometimes, you need to introduce constraints.

When we started Consumer Mesh, our backgrounds were in building portals and dashboards. Creating a Figma-to-portal tool would have been an obvious fit. But we saw how low-code/no-code platforms like PowerApps and Bubble suffered from performance issues that killed user experience and doomed applications.

To avoid that trap, we focused on public web experiences where performance is critical. Google evaluates Core Web Vitals to measure true end-user experience, so we architected for static sites powered by a robust CMS (Drupal). This gave clients the best of both worlds: lightning-fast websites and a superior authoring experience.

That discipline positioned us perfectly. Now we're taking on portal applications with an unwavering focus on performance, user experience, and content management—ready to outmaneuver established vendors.

The DeepSeek moment wasn't about having less—it was about achieving more with less.

That's the fundamental shift constraint-based innovation demands: stop seeing limitations as obstacles to overcome and start seeing them as design principles to embrace.

Moonshot's quantization choice and our static-first approach share a common DNA. Both refused to accept that "good enough" performance was good enough. Both imposed discipline that vendors with bigger budgets could afford to ignore—until they couldn't. While others patch performance issues later, constrained teams bake excellence in from the start.

This isn't about being scrappy for scrappiness' sake. It's about intentional limitation as a strategy. Real constraints are coming whether you invite them or not: market shifts, resource cuts, competitive pressure. By practicing with artificial ones, you build the innovation muscle memory that turns future crises into opportunities.

The method remains stubbornly simple: ship, fail, learn, repeat—faster than everyone else. Frameworks might evolve and AI might accelerate prototyping, but the core loop of constraint-based innovation is timeless.

Your greatest product advantage isn't what you can build with unlimited resources. It's what you're forced to discover when you have none.