Post

Premium Newsletter: How Has AI Changed The Economy? Not very much outside of GPU sales and data center construction, with AI software sales and GPU rentals having a negligible effect, even as BLS data overemphasizes software's contribution to real GDP. www.wheresyoured.at/premium-how-...

This week the WSJ reported that AI is expected to account for 3.63% of GDP over the coming six years — an insane figure suggesting that AI's contribution, mostly from AI GPUs and data center construction, is a load-bearing part of the economy. www.wheresyoured.at/premium-how-...

These figures are nothing to celebrate, as they show the economy is dependent on a largely debt-driven capital expenditure explosion, meaning that it will inevitably slow. Underneath, AI software and GPU rental sales aren't doing much to grow GDP. www.wheresyoured.at/premium-how-...

The effect of AI GPU sales and data center construction on GDP is also obfuscating a growing problem with the tech industry - that its contribution to the economy is stagnating on a nominal basis, and AI doesn't appear to have increased real GDP growth. www.wheresyoured.at/premium-how-...

Part of the problem is that software's contribution to real GDP is overstated, due BLS' mistaken belief about the extra "value" added, that software prices are near-identical to 1997, even as every software company raised prices starting in 2022. www.wheresyoured.at/premium-how-...

All of this is to say that the actual impact of AI on the economy may be far smaller than boosters would have hoped. But that too is an estimate, because we really don't have great data to base any conclusions upon. www.wheresyoured.at/premium-how-...

It doesn't help that the waters have been muddied by all sorts of exotic metrics that sound impressive, but don't tell you much, and are themselves based on faulty assumptions. www.wheresyoured.at/premium-how-...

Attempts to measure the economic effects of AI using "cost of intelligence" are inherently flawed, because benchmarks do not equal actual units of work or replicable completions of tasks outside of their rubrics. Everyone's measuring the wrong things. www.wheresyoured.at/premium-how-...

The problem is that we've started with the assumption that AI will be an economic driver (especially when it comes to productivity), instead of starting with a hypothesis to test. Something I blame on the economic context of when ChatGPT launched. www.wheresyoured.at/premium-how-...

Another problem with calculating AI's impact on economic productivity is that, until very recently, AI services were almost uniformly subsidized via all-you-can-eat subscriptions which aren't reflective of any sustainable long-term use. www.wheresyoured.at/premium-how-...

Even with that caveat, which should benefit AI, there's still virtually no data indicating any major economic-wide productivity gains from AI. www.wheresyoured.at/premium-how-...

On an individual business level, there's still no signs that AI's driving any real productivity gains, with most CEOs still talking about AI's productivity benefits in the same hopeful terms as they did in 2023. www.wheresyoured.at/premium-how-...

Similarly, the impact of AI on employment remains a threat rather than a reality -- as does the wide-scale creation of AI-related jobs, which to date tend to be precarious, low-paid data curation gigs. www.wheresyoured.at/premium-how-...

More than a trillion dollars and four years into the AI bubble, it's not unreasonable to expect that we'd have solid data showing that AI was this major economic agent of change. www.wheresyoured.at/premium-how-...