The million-dollar AI mistake
Good morning,
If “corporate horror” were a movie subgenre: in 2019, a couple received strange deliveries including live insects and a bloody pig mask as part of a harassment campaign against them. Now, eBay and former execs have agreed to pay ~$56M in a settlement with the couple.
In today’s edition:
- AI growing pains get “catastrophically” pricey
- Flight prices spike for the top corporate routes
- Overheard in tech: token poor
Everyone’s stressing over AI capex, but opex spending is even slipperier
AI infrastructure spend has dominated headlines. Understandable, since investors are sweating over the $2.4T in capex that Big Tech has committed to the data center buildout. But there’s another side of AI spending that’s silent… and it can be very costly.
We’re talking about the opex side, or spending on AI to boost internal productivity. With traditional productivity-boosting software, you don’t usually have to worry about overspending for using the tool. The worst that can happen if you spend too much time in your CRM or ERP software is dry eyes. With AI, the reality is much different:
- The Financial Times reported that Amazon employees identified instances of “catastrophically expensive” cost overruns caused by mistakes in AI deployment.
- In one case, employees were reportedly told that the company spent $1.8M on a failed deployment which involved trying to match author details with listings using AI. That was 860% over the project’s budget and took five months to detect.
Amazon said that “isolated examples where teams are learning from one another” don’t reflect how teams across Amazon are using AI, and that may very well be the case. Still, this story highlights just how hard it can be for any company to track AI-related costs.
Ramp data shows AI costs spike 50% or more about one in every four months for the biggest spenders. Recently, there have been reports of companies blowing through their full-year AI budgets way ahead of schedule.
The visibility issue doesn’t just apply to tech giants. Smaller companies actually have a higher median monthly AI spend per employee, according to the Ramp AI Index.
- Small companies spend about $20 per month per employee on AI, versus $8.33 for medium companies and just $2.86 for larger companies.
- Meanwhile, the top 1% of AI adopters on Ramp spend $4,883.33 per employee per month, compared to a median of $10.66.
The bottom line:
The growing pains stage must eventually end. Because the AI efficiency gap is costly, especially for the biggest adopters. Ramp data shows the median company spends $3.54K a month on AI, but the gap between an efficient team and a wasteful one is more than 4x per token. We’re still in the early stages, and growing pains are a given. But considering the cost, companies big and small will have to start closing the efficiency gap soon. Several are already implementing AI spending caps and restrictions.
High-flying prices likely won’t be descending
Catching flights to catch meetings isn’t what it used to be. Prices for the top five roundtrip routes booked on Ramp all saw double-digit increases in Q2 compared to a year earlier.
- San Francisco (SFO) pairings saw the biggest surges, led by the Seattle (SEA) and SFO pairing, up 41%. The median price for the 2-hour-ish flight was $637. The LAX-SFO pairing wasn’t far behind, up 36% to $374.
Middle East fuel supply disruptions have led to surging airfares, and not just for corporate America: the latest CPI shows U.S. airfares up 26.5% in June from a year earlier.
Carriers like United and American Airlines expect their fuel bills will rise by billions this year, and they’re passing on costs to travelers.
But even though Brent crude prices have cooled from recent multiyear highs, airlines expect fares to stay high this year, even if prices dip.
Execs say bookings are still strong despite the fare hikes, and Delta CEO Ed Bastian said he expects the pricing power to hold.
Term Sheet: “token poor”
A term overheard in tech circles: token poor, a person or company lacking in the AI usage tokens needed to accomplish complex tasks (or really, whatever task they need).
A software engineer might comment that they’re “token poor” after leaving a token-generous company. The engineer might be financially rich, but without the seemingly endless corporate AI budget they’re token poor.
Earlier this year, token budgets were becoming the new corporate perk in tech, advertised alongside benefits like unlimited PTO, catered lunches, and kombucha taps. In March, Nvidia CEO Jensen Huang pitched the idea of giving his company’s engineers a generous AI token budget in addition to their base salary.
Even after the tokenmaxxing comedown, AI is becoming so entrenched in daily workflows that “token poor” is a term we might be hearing more often.
Signals Shortlist
- SpaceX’s post-IPO plunge sets tense backdrop for first earnings report (CNBC)
- NY sues Kalshi, says its prediction markets are illegal gambling (NBC News)
- Citadel Securities sees a $500B chip financing debt binge (Bloomberg via Yahoo Finance)
- How a wunderkind built a $45B AI hedge fund and lost most of it in days (CNBC)
- ISM Manufacturing hits a four-year high, expanding for seventh month (PR Newswire)
- The struggling colleges raiding their endowments to pay the bills (The Wall Street Journal)
🗓️ Leading Events
Tuesday, August 4: JOLTS. Earnings expected from SpaceX, AMD, McDonald’s, Caterpillar, Pfizer, BP, Marathon Petroleum, Arista Networks, Booking Holdings, and Amgen
Wednesday, August 5: ISM Services PMI. Earnings expected from Disney, Eli Lilly, Uber, Shopify, AppLovin, Sandisk, e.l.f., and DoorDash
Thursday, August 6: Weekly jobless claims. Earnings expected from ConocoPhillips, Warner Bros. Discovery, Airbnb, Rigetti Computing, and Datadog
Friday, August 7: July jobs report. Earnings expected from DraftKings, Oklo, and Take-Two
🤝 Like what you read? Share Leading Indicators with a friend.
