The Next Collapsing Tech Cost Is Software Itself
History can act as a good guide to what the future might look like…
Firstly, are we in an AI bubble?
Marc Andreessen says no — and makes a compelling case.
In a recent a16z podcast, he noted: “At the peak, 97% of the fibre that had been laid was dark. Contrast that with today. There are no dark GPUs.”
Tech waves surge when the expensive turns cheap enough to squander.
He argues that the dot-com bubble stemmed from “dark fibre”… massive overbuilds of internet infrastructure with zero demand to light it up.
Uptake and use simply weren’t there.
Tech waves surge when the expensive turns cheap enough to squander.
AI faces no such demand drought — quite the opposite.
With Work Slop and AI Slop emerging as genuine issues (endless low-value outputs flooding workflows, emails and the web), and AI so dirt-cheap to run, consumption is exploding.
Suddenly AI has become cheap, to the point where people are “wasting” it.
Suddenly, we’re wasting it on everything from meme generators to code tweaks.
Due to AI’s intuitive UI — natural language input — it’s wildly accessible.
Usage is through the roof.
So, back to those technology collapses…
We’ve seen costs plummet across the board. Remember Gmail’s 2004 launch, offering “ridiculous” 1GB of free storage when rivals capped at 4MB?
Or dial-up days when every byte of data felt precious — now we stream 4K in the background without a second thought, and auto-updates eat bandwidth unchecked.
Add CPUs to the list… from room-sized mainframes to pocket supercomputers.
And now software…
Generative AI tools are boosting developer productivity by up to 55%, slashing dev cycles and costs for everything from prototypes to production code.
Marc Andreessen is famous for his widely cited 2011 Wall Street Journal essay: “Software is eating the world.”
For decades, software’s ballooning complexity and sky-high development costs choked production, piling up societal technical debt.
To some extent, software couldn’t eat the world — not when it demanded scarce, expensive resources like skilled programmers to build even basic apps.
Now?
Software is devouring everything because its cost and complexity are collapsing.
Like content went permissionless (YouTube, blogs, Twitter), software is going permissionless too.
Content is now an application; apps are now content.
As content proliferated across mediums and morphed into apps, the lines blurred…content is now an application; apps are now content.
But…
I suspect this shift will play out differently in enterprises.
Legacy systems, compliance hurdles and scale demands won’t vanish overnight — AI might democratise the edges, but the core stays gated.
Software’s complexity & cost impeded production for decades, creating societal technical debt.
Case in point, while consumer-facing permissionless AI thrives via decentralised platforms, enterprise adoption faces hurdles, with 95% of generative AI pilots failing to scale to production due to security and governance gaps.
Chief Evangelist @ Kore.ai | I’m passionate about exploring the intersection of AI and language. Language Models, AI Agents, Agentic Apps, Dev Frameworks & Data-Driven Tools shaping tomorrow.
