We appear to be living on a curve of improving LLMs. On the podcast MOONSHOTS with Peter Diamandis, in a recent episode, one of the hosts commented that they plotted the timeline of Frontier LLM models, or near frontier, and that by January 2027 there will be new releases daily. And that feels about right. In 2025 the releases came at most quarterly. In early 2026 it was monthly. And now it is weekly.
So that seems like a real exponential curve in LLMs at least.
The market needs an exponential curve in LLMs because they believe that means AGI or ASI, at minimum due to reinforcing and recursive learning, which they need to justify today’s astronomical valuations. Otherwise a financial bubble popping will trigger a broader economic downturn. Even with an exponential curve, not all of the Frontier LLM companies are sure to survive. So continue to expect financial mania and VC hype.
But regardless of the financial curve, this is not a technological bubble because there is utility in even the current models which will still reshape our physical and economic landscape. LLMs are important because they are accelerating work in math, physics, chemistry and biology, plus managing processing information within companies and government, plus generating media. They are in the early days of being embodied in the world via drones, robots and driverless vehicles (really all the same thing).
So are we on an exponential curve, or an S-curve?
Whether we are on an exponential curve, or an S-curve will only become apparent well after the fact. Unless we experience a true hard takeoff.
True exponential curves don’t exist in the physical world; in the physical world, smooth exponential curves are an illusion; all sustained revolutions are relay races of stacked S-curves hitting and solving bottlenecks.
Semiconductor technology, as demonstrated by Moore’s Law, and energy technologies like solar, batteries, and electrification, are thought of as being exponential curves. However each faced bottlenecks of hitting physical limits, integrating new innovations, building new or replacing infrastructure, and overcoming institutional resistance. They are actually a series of S-curves, even if they were very steep curves, before handing off to the next paradigm.
So, progress could be stuck in a single S-curve, or a series of S-curves over the next couple of years because of limits of the current LLM approaches, or bottlenecks of supply and/or demand.
There are supply side bottlenecks in memory, in making chips that are faster / denser / cooler, in building data centres. There are also minor supply bottlenecks in training data, having scooped up the publicly available internet.
And there are demand side bottlenecks in getting new medicines through trials; changing business in regulated industries; manufacturing at scale; disasters (man-made and natural) affecting manufacturing, infrastructure and supply chain. These will flatten the curve, or at minimum make it spiky.
Even within a single S-curve, we will have a decade of inventions in materials, chemistry and biology, as we spend a decade of adapting our companies, governments and society. Or it could be the first in an series of S-curves, that would be even more as transformative.
Regardless, we are in store for a decade of change as dramatic as the previous three decades.
Think of the world prior to 1995 and the first commercial web browser. Then we had e-commerce, blogs, cellphones and fibre optics; then we had smartphones, pocket GPS, social media, Airbnb and Uber. Then we had large flat screen televisions, 3D printers, and Netflix, Instagram and TikTok. Now we have video conferencing and chat (FaceTime/Zoom/Teams/Slack/WhatsApp), tap to pay, driverless vehicles, military drones, quadcopter taxis, and wireless internet anywhere on the planet. In 1995 we could foresee e-commerce but not Airbnb or TikTok or Netflix. Few people in that pre-1995 era would believe or understand the world of 2026.
Some things of the next decade have been around but never widespread or affordable or didn’t make it out of the laboratory before. But even this current state of LLMs is already changing the velocity of existing products, and generating new theories, and affecting the emergence from the laboratory into the everyday world.
The first-order effects include:
- discoveries in math, physics, chemistry and biology,
- automation of processes (which we still call paperwork) in business, law, software, and government.
- LLM designed objects, 3D printed.
The second-order effects include:
- automation of testing theoretical models in laboratories, both biological and material.
- automation of manufacturing to the point of dark factories with mobile and fixed robotics instead of workers.
- real-time audits of processes to ensure effectiveness and catch fraud or unauthorized access.
Third-order effects include:
- driverless delivery of goods right to, or even in, your home.
- entertainment with fully immersive and interactive narratives (for all practical reasons a game) streaming generative standalone or multiplayer
- experiences directly to screens and VR headsets.
- consumers adopting functional medicine, personalized nutrition and real-time metabolic monitoring.
All of these are actively being built today, and their convergence will unlock possibilities beyond current imagination.
Regardless of the financial bubble we are on a technological curve and one steeper than any we have seen before, driven by LLMs.
As William Gibson famously observed: “The future is already here – it’s just not evenly distributed.” It is about to become even less evenly distributed.
Welcome to living on the curve. Buckle up. Put on your breathing mask first. And expect things to get spiky, strange and steep.