Minimum Viable Infrastructure

Innovation often emerges not solely from brilliant ideas but from the infrastructure readiness that makes those ideas viable.

Consider the progression of mobile consumer internet applications over the past two decades: Twitter, Instagram, Snapchat, TikTok. Their launch order wasn't accidental—it precisely followed the trajectory of mobile bandwidth rollout.

• Twitter (2006/text): Came first because text has the smallest data footprint (was also SMS compatible).
• Instagram (2010/compressed images): Emerged next as bandwidth became sufficient for easy sharing of compressed images.
• Snapchat (2011/compressed images + short-form video): Appeared when mobile networks could reliably handle short bursts of video.
• Musically/TikTok (2014/long-form video): Gained momentum when infrastructure supported extended streaming and uploading of longer videos.

Most importantly, the order of founding could not have occurred in any other order. Not only that, but this is also the reason why any further attempts to found mobile text/photo/video platforms have failed. The window of opportunity has long since passed.

Today, we stand at the precipice of another fundamental infrastructure shift: on-device inference for Large Language Models (LLMs). Currently, cloud-hosted models vastly outperform local models, echoing the early disparity between cellular text messaging and IMAX theaters existing at the same time—while similar in medium, completely different pipes and ultimate capability.

The critical bottleneck for local AI is “useful^[1] tokens-per-second”—the rate at which a device can process and generate text or media content. As chipsets and memory improve exponentially, so will local inference capabilities. Crossing key thresholds of tokens/sec will unleash entirely new classes of products and services.

This metric—tokens/sec on-device—will be the "Why Now?" moment for the next wave of startups just as kbps was for a previous wave.

Just as Twitter, Instagram, Snapchat, and TikTok sequentially defined the landscape of mobile-first apps based on bandwidth thresholds, the next era's defining companies will ride the wave of rising local inference capabilities. Startups tuned into this progression—anticipating infrastructure improvements and understanding precisely when the tipping points occur—will create the next set of generationally defining businesses.