7 AI-related charts

GPU pricing, enterprise budgets, and more hiring data

Publishing note: Random Walk is going to experiment with a different publishing cadence over the next few weeks (more details forthcoming). I will endeavor to trade some frequency for depth, at least some of the time.

• GPU rental index

• AI budgets increasing

• No, AI is (probably) not taking your job. It might make your job, though.

---

👉👉👉Reminder to sign up for the Weekly Recap only, if daily emails is too much. Find me on twitter, for more fun. 

---

AI Charts

It’s Nvidia day, so just a few quick AI-related charts.

GPU rental price index

There’s apparently a new GPU rental cost index, which tracks the cost of renting H100s as a proxy for demand.

With no view on the veracity of this thing whatsoever, this is what the series looks like:

Bloomberg

H100 compute costs got progressively cheaper, and then spiked in January, before leveling off again.

Rental costs apparently plummeted in May, for reasons that are entirely unclear (and may well be a data issue).

As a proxy for supply-and-demand, GPU rental costs make sense (and it’s not terribly surprising to see costs come down). This is something Random Walk has been interested in before.

The hard part, from my understanding, is that prices are hard to ascertain because there’s a lot more nuance than simply $/hr, including what other bells n’ whistles are involved, who the seller happens to be (primary or secondary), and other “strings attached” on usage (e.g. the cost of off-hours spare compute).

More broadly, this reflects the increasing sophistication of compute markets (and I suppose is part of why investors are bullish on CoreWeave).

Enterprise AI budgets increasing

Firms have been generally belt-tightening for a while, which has made tough-sledding for enterprise SaaS companies, but the exception is AI, of course.

A recent Bain survey shows that annual budgets for AI have grown a lot:

Bain

Average budgets have doubled since February 2024, from ~$5M to $10M.

The biggest gains are amongst the largest companies ($5B+) where average annual budgets for AI spending have 5xed.

If there are approximately 5,100 companies with revenues north of $5B, then you get ~$140B in annual AI spend, just from the biggest companies. That’s not nothing.

No, AI did not take your job (reprise)

Incidentally, the reason that companies are spending less on IT is not because “AI took our jobs,”:

Bain

Only 7% of respondents said “AI has made the org smaller” as a reason that IT budgets are under pressure.

To be fair, even if AI has made the org smaller, that’s an odd reason to give for cutting IT budgets (plus the share of respondents did increase from 4% last year), so take this for what you will.

More broadly, the AI took our jobs claim continues to be mostly not-true, as best as anyone can tell. On the other hand, AI as job-maker seems credible for now. Mostly, though, hiring is just slow across the board, tech included.

The “intensity” of AI-related jobs as a share of overall IT job postings has grown a lot:

Sherwood

There are now 24 “AI jobs” for every 100 “IT jobs,” which is about 2.5X the prior ratio.

This is another way of saying that with respect to technical hiring, AI is increasingly the main event. The counterfactual is hard to know, but query whether in the absence of AI hiring, there would be any standout categories, at all. My guess is “no.”

For Tech companies specifically, if anything, hiring has been reoriented around AI (as opposed to displaced by AI).

SignalFire

At Big Tech, hiring has contracted across the core “business function,” while Machine Learning headcount has grown substantially.

There’s all kinds of denominator issues here, but you get the idea.

One thing that does appear to be true is that entry-level roles continue to be hardest hit—again, this is true everywhere, but it’s interesting to see it as applied to tech, and it’s interesting to see the apparent divergence with more senior roles.

For both Big Tech and startups, entry-level hiring is hardest hit, while more senior roles make a moderate comeback:

SignalFire

Again, hiring is down across the board, the most senior roles are closest to breakeven in Big Tech, and even small tech has increased headcount in the 2-5 year seniority range.

For all the talk of “no more middle management” it’s interesting to see senior hires making a comeback, but perhaps these are just very experienced engineers.

Is it possible that AI is replacing the entry-level associate, but not the employee with 1-2 years experience, or 2-5? I guess it’s possible, but it seems more likely that companies are prioritizing “track record”—even just the barest semblance of demonstrated competence—and that they want “track record” to be developed on someone else’s dime.

If you want to blame anyone (and it’s not obvious there is anyone or anything to blame), this seems like (yet another) failing of ‘higher ed.’ Graduating with a fancy degree is obviously not producing enough signal to demonstrate that grads have any useful skills and/or are ready for the workplace.

---

Other reads

• Hope and hype of fusion. This is above my pay-grade, but I like general bullishness on energy innovation. Why not fusion too?

• Subscription line financing all the rage. GPs are borrowing against capital calls to fund investments, and other things. Money is tight, but it hasn’t broken anything yet.

• Future of small business banking. Good rundown of all the fintech and incumbent players competing for different segments of the business lending landscape (and the different ways they’re doing it).

• AI’s energy footprint: the story you haven’t heard. Scrolly-tell from MIT tech review on the energy consumption habits of LLMs. It turns out, it’s a hard question to answer.

• People use more AI than you think. A lot more tokens are being consumed. At the same time, models are getting more efficient. It stands to reason, people are using AI a lot.

• Measuring employer-demand for AI skills. No real surprises here, but continues to reinforce my priors that AI is job-creator, not job-killer.

• Has the Beige Book become disconnected from economic data? The Fed’s anecdata report on business conditions (the Beige Book) has, since the pandemic, tracked substantially lower than hard data. One reason might be because all the economic growth has been in counter- and a-cyclical industries, which are a much smaller part of the Beige Book coverage.

• Shoemaker leans on automation to expand US production. They said it couldn’t be done.

---

Previously, on Random Walk

---

Random Walk is an idea company dedicated to the discovery of idea alpha. Find differentiated data, perspectives and people, and keep your information mix lively. A foolish consistency is the hobgoblin of small minds. Fight the Great Idea Stagnation. Join Random Walk. Follow me on twitter. Follow me on substack: