AI adoption follows the J-curve path of general-purpose tech, like early US factory electrification, requiring years of investment before noticeable ROI gains
A framework to understand your firm's AI transformation — I had tea with a senior exec at a well-known public tech company last month.
Exponential View
Context & Ripple Effects
Related coverage frames this as an adoption problem rather than simply a model-capability race: AI vendors are expanding forward-deployed engineering teams to help enterprises put models into production.
The backdrop is a capital-intensive AI buildout with thin reported margins, while earlier coverage argued that initial enterprise deployments may center on labor-saving implementations. The J-curve framing explains why spending and visible returns can remain out of sync during that transition.
First-order effects
Companies pursuing AI transformation must absorb an extended period of investment before they can expect material, observable ROI; near-term performance assessments will therefore be a weak test of the technology's eventual value.
AI vendors and enterprise buyers face immediate pressure to focus on implementation support, not just access to increasingly capable models.
Second-order effects
Demand for forward-deployed engineers and other deployment-oriented services should rise as customers need help adapting workflows before model spending produces measurable gains.
Big Tech's already asset-heavy AI buildout becomes harder to justify on short-term returns alone, intensifying scrutiny of utilization, margins, and the path from infrastructure spending to enterprise revenue.
Third-order effects
If the pattern holds, AI competition will increasingly be decided by organizational integration and distribution into business processes, rather than by model capability alone.
The sector may bifurcate between organizations able to fund and manage the investment trough and those that limit deployments because delayed returns are too difficult to sustain.
The trend: AI is moving from a model-access market into a general-purpose-technology rollout in which complementary investment and workflow change determine when value appears.
GPTs are GPTs, and deploying them in a way that maximizes productivity growth is highly non-trivial This is the exact idea behind the FDE team at OpenAI - we aim to drive maximum possible returns for our clients by reimagining workflows from the ground up with agents in mind
Invention and diffusion are separate concepts. We will have 10-20 years of work to realize the benefits of the AI that's already here. It won't be overnight. Spending a bunch of money on tokens does not a process re-engineer.
“Erik Brynjolfsson calls this the productivity J-curve: general-purpose technologies are a drag in their early years because firms have to make complementary intangible investments before the gains materialize.” [image]
This is an interesting theory, but one may worry that the inefficient and broken sectors of the economy will simply eat up any dark economic surplus, the same way they did during the computer revolution. https://newsletter.semianalysis.com/ ... [image]
“We are at risk of having an event on the scale of the Industrial Revolution where most of the new output is invisible even as businesses spend increasingly large amounts on AI services.”
AI Dark Output: The Visible Cost of Invisible Output Why AI's increasing output is going to be one of the hardest economic measurement problems in history. AI “Dark Output” could end up being the majority of economic activity, but a challenge to measure. https://newsletter.semian…
GitHub Copilot's switch to token-based billing reveals the hidden economics of AI code completion. Flat monthly pricing was masking massive cost variance: power users consuming 10-100x more tokens than casual users. The new model makes unit economics visible but breaks the user […
Resting the entire economy on the rock-solid foundation of an imaginary metric that even its believers admit they can't substantiate. — Great idea. This one's going in the history books. Under “comically stupid boondoggles”.
There is no other industry, surely, where this mealy mouthed paragraph, void of any real measure, is presented as proof of ‘value’ and/or success? [embedded post]
Meanwhile, tensor processors are on a 3 year depreciation schedule and are nearly 40% of data center costs. Tick tock! — My heart will burst with joy if adoption takes so long, the worst humans ever to live lose absolutely existential amounts of money, and the infrastructure h…