Long(View) from London: “AI Capex – How Economically Significant?”

Key Quote re: AI Data Center Expenditure

“Using Nvidia’s datacenter revenue as a reference for US AI capex, @JensNordvig estimates that AI will make up ~2% of US GDP in 2025 given a standard multiplier, implying an AI contribution to GDP growth of 0.7% in 2025.” – see table below,

Source: Jens Nordvig, Exante LINK (via Ed Conard)

Stat of the Week: ‘Impact & Size of AI Capex’

Table 1: Estimates of the impact of AI capex on US GDP

Next week, Nvidia is set to report Q2 quarterly earnings (after hours on Wednesday). The consensus is for earnings of $1.00 up from $0.68 (last year) and sales revenue of $45.6 billion (up from $30 billion last year).

A quick glance at Nvidia’s revenue growth (chart below) illustrates why it’s such a stock market darling. Expected quarterly revenues of $45.6 billion are up 50% on last year; & approximately triple the level of the same quarter 2 years ago in 2023.

Fig 1: Nvidia quarterly revenue ($bn)

Whilst most investors will be watching to see how the company fares relative to expectations, and with respect to its outlook, economists will be interested in the capex signals they can glean from the earnings release and the earnings conference call.

According to Jens Nordvig, (Chief Economist at Exante), AI capex will increase US GDP by 0.7% in 2025 (after 0.8% in 2024) – see table 1. That is calculated by grossing up Nvidia’s AI data center related revenue (by their market share) and then applying a capex GDP multiplier. Capex is considered by economists to have the best multiplier effect on GDP (i.e. in this example, $1 of capex translates into $1.5 to $2 of GDP).

But the question is: How credible (or indeed relevant) is that estimate? And is it incremental (or, simply instead of other spending)?

Paul Kedrosky (see below) takes that capex number and demonstrates its importance relative to prior bubbles. He calculates it’s currently bigger than telco capex spending in 2000 (during the dotcom mania) – i.e. 1.2% vs. 1.0% of GDP for telcos in 2000. It is, though, substantially below his estimate for the railroad boom of the 1880s (i.e. at 6% of GDP – although a quick ChatGPT search suggests railroad capex in the 1880s was running at around 1 – 3% of GDP per annum, substantially below his 6% estimate).