Loading summary
Transcript19 lines
- [00:01]
A
Exchanges on the M and A and IPO landscape Exchanges on the dynamics affecting global trade. For the sharpest analysis on finance, business and the economy, count on exchanges the Goldman Sachs Podcast Listen now. Welcome to Tech News briefing. It's Tuesday, February 17th. I'm Julie Chang for the Wall Street Journal. The AI boom has seen a surge in capital spending for things like new gigantic data centers. And that rapid growth is making it harder to analyze major tech companies earnings. Then a leading expert on the intersection of technology and work shares his perspective on how AI is reshaping the labor market. But first, five major tech companies, Alphabet, Amazon, Meta, Microsoft and Oracle are expected to spend a combined total of $3 trillion on property and equipment in the next four years. But the results won't just be new data centers. We'll also see depreciation expenses soaring. Yet investors can't find those expenses listed on the income statements of big tech companies. Nor is there any consistency in how to report these costs. That's according to Jonathan Weil, who covers finance for the Journals Heard on the street column. He's with me now. Jonathan, can you just briefly tell us what are depreciation expenses for people who might not know?
- [01:29]
B
Depreciation is the expense that companies report as part of their earnings when they go out and build huge plants or they go out and buy lots of equipment, whether it's semiconductor chips. And they don't immediately expense the cost on their income statement of buying that, they put it on their books as an asset and then they try to estimate how long that plant or how long that equipment is going to last. And then they gradually write down the cost of that over time. And the problem for investors is that these depreciation expenses, which are going to be ballooning in the next several years, you can't see how much gets sprinkled into cost of revenue or how much gets sprinkled into R and D or sales and general administrative expenses or other categories of operating expenses. You can find a disclosure that tells you what the total depreciation expense is. You can find that either elsewhere in the financial statements, whether in a footnote, sometimes on the cash flow statement, but where investors really need to be able to see it so they can project out how big the depreciation expenses are going to be and also how it affects all those other categories. They need to see how it's allocated across the income statement.
- [02:40]
A
How does that differ from the way other industries account for depreciation expenses?
- [02:46]
B
This is actually a common problem widely among lots of industries, but there are exceptions where they will show you expressly on the income statement what depreciation expense is. One example is railroads, a classic asset heavy industry. They will tell you the depreciation expense on the face of the income statement and the real service that they're doing there is that you know then that all their other categories of operating expenses don't have depreciation expense buried inside of them.
- [03:15]
A
And you write in your column that rule changes are already on the way. What are those changes and when do they take effect?
- [03:21]
B
The board that sets generally accepted accounting principles in 2024 approved a new set of rule changes and what they'll require is that all those different types of expense categories on the income statement will have to be broken out somewhere in the financial statements into five different categories. And that disaggregation will be helpful to understand what the depreciation cycles for these investments are lacking the effect they have on earnings. The problem for investors who want this information now, because capital expenditures are exploding now, is that those rules won't take effect for most companies until financial statements that they file in 2028 for the year before. So, you know, there's an opportunity for companies to be more transparent and start giving those disclosures earlier. But there hasn't been any sign yet that any of the really big tech companies for which this is a crucial problem are going to be adopting those new rules early.
- [04:18]
A
You know, there's been a lot of talk about a potential AI bubble. Does this give us any insight into just how frothy this market might be?
- [04:25]
B
Not necessarily, but what it does, it gives us insight into blind spots that would keep investors from being able to evaluate the companies and how frothy things may be. Because if you're looking at, say, a company's cost of revenue and you can't really tell what the components are of it, it's really hard to project out what a company's gross margins are going to be if you can't tell how much depreciation expense got baked into there. The same thing with research and development or the other expense lines. If people already have a concern that there's a bubble, what could help put investors mind at ease or confirm? Perhaps their thesis is actual information and visibility, and the less you have of that, the more it leaves people guessing and openly wondering. The question Is this a bubble that.
- [05:12]
A
Was hurt on the Street? Columnist Jonathan Weil, what do you make of major spending on the AI buildout? If you're a listener on Spotify, leave us a comment with your thoughts. Coming up. How is the tech sector preparing for the next wave of automation. And what does that mean for the future of your job? Stay with us. That's after the break.
- [05:36]
C
Trading at Schwab is now powered by Ameritrade. Unlocking the power of thinkorswim, the award winning trading platforms loaded with features that let you dive deeper into the market. Visualize your trades in a new light on thinkorswim desktop with robust charting and analysis tools all while you uncover new opportunities with up to the minute market news and insights. ThinkOrSwim is available on desktop, web and mobile to meet you where you are. It's built by the trading obsessed to help you trade brilliantly. Learn more@schwab.com trading.
- [06:09]
A
After a few years of artificial intelligence experimentation in the workplace, we're starting to see tangible effects of AI on the nature of work from from productivity gains and evolving processes to layoffs and hiring freezes. Erik Brynjolfsson is a professor at Stanford's Institute for Human Centered AI and a co founder of research and software company workhelix. At the recent WSJ Technology Council summit, he sat down with the WSJ Leadership Institute's Wendy Bounds to discuss how AI is really influencing the labor market from the difference between using AI to automate jobs versus augment them, what he calls the Turing Trap, and more. Here are highlights from their conversation.
- [06:52]
D
When we're thinking about the people driving this, you've spoken a lot about the power law adoption of AI, where you got like a small group of rock stars who have these that see outsized gains, where the rest of the company is just kind of like trickling along. What should everybody in this room be doing to get the longer tail of people on board with this?
- [07:13]
E
At my company, we go in and work with a lot of companies and understand how people are using AI, particularly internal LLMs. And what we find in basically every company is there are a few like super users that are doing great and then there's this long tail of the typical users. So one of the things we were doing to help them level up the rest of the people was understanding what are those power users doing that is really effective and then creating templates so other people could do it. They wouldn't necessarily have the exact same thing that those power users were doing, but showing them that this is possible. We used to just tell people here are some best things you can be doing based on first principles. But now when we can go inside the organization and point to one of their colleagues that's actually doing it, we find it's much more credible. It's somebody who's already doing it in the same company, in the same function. And so we just had this learning organization where it's not like once or twice a year we do this. It's a continuous process, constantly identifying these power users and creating templates for other people to level up. And it's just, it's like a lot of low hanging fruit. It's really easy to get wins that way.
- [08:22]
D
So we've had a bunch of different discussions here about entry level workers and declines in employment for certain types of entry level workers. And I want you to talk about what you're seeing there and the difference between the AI tasks that are automating work and ones that augment it and how that's affecting that employment.
- [08:40]
E
Okay. Yeah. The government data actually was not fine grained enough. We were lucky to be able to partner with adp, the world's largest payroll processor. They have a partnership with the Digital Economy Lab, which I run at Stanford University. And if you look at the top line, you don't see all that much, you know, a little bit kind of flat employment like we saw in the BLS reports. But then when you slice it into these different groups, what we did was we ranked all the occupations based on how exposed they are to AI using this methodology that one of my co founders Work Helix developed. And you can rank all the occupations and then when you look at them, you start seeing something, little glimmers. And then when you slice it one more way, when you slice it by age, you start seeing something very pronounced. Specifically in the most exposed occupations, like coding call centers, parts of sales, parts of administrative work. For the youngest workers, the ones aged 22 to 26, there was about a 13% decline. That was when we first published the paper. We now have three more months of data and now that's become a 16% decline for those particular workers. The more experienced workers were doing fine. And then you could subdivide it based on the prompts and what kinds of questions they were asking. One way we classified it was people who were mainly using the prompts to automate and replace some of their work and others that were mainly using it to augment to do something new, to extend what they were able to do. And if you look at those two groups, the first group had even steeper falling employment. But the second group, which sadly was a minority of them, had overall growing employment. So the employment growth was only in that group that was using the LLMs to augment what they were doing?
- [10:16]
D
Tell us about the Turing Trap and why we all need to be worrying about that a little bit.
- [10:21]
E
You guys have all heard of the Turing Test, this iconic thing? 1950, Alan Turing said, hey, what's the test of AI? It's if you can't tell the difference between the AI and the human if it can perfectly imitate. He literally called it the Imitation Game. So that was his test. Kind of cool. I thought it was neat when I was like, 12 years old and first read about it, but actually, now I think it's a terrible idea. It is not what you want AI to be doing, and it leads to a trap, which I wrote this paper called the Turing Trap. The reason this is a trap is for a couple of things. First off, we want the machines to be able to able us to do new things that we never could have done before. Extend our possibilities, not just mimic what we're doing. Secondly, it's a sociological nightmare because if you tell your employees where our goal is to replace you, they're not going to embrace your challenge to use AI more. They're going to say, I'm not sure I want to be on board with this. And then, last but not least, I think societally it's not a good thing. From an economist's perspective, if you make a machine that is an imitation or a close substitute for a human, it tends to drive down wages. That's what substitutes do. They create competition. But if you make a machine that's a complement, that extends what humans can do, then it drives up wages. And so I've been saying, look, think about ways we can use AI to extend what we can do instead of having these black box benchmarks. How well can the AI do it alone? I want them to use what I call Centaur benchmarks, half human, half machine, where they look at how well the team can do to solve some new problem. And I think that's the path towards having a higher ceiling, having more people embrace it, and ultimately having not just prosperity, but shared prosperity.
- [11:59]
A
That was Stanford professor and Work Helix co founder Eric Brynjolfsson speaking with the WSJ Leadership Institute's Wendy Bounds. We'll link the full talk in the show notes. And that's it for Tech News Briefing. If you're a listener on Spotify, leave us a comment. Today's show was produced by me, Julie Chang, with supervising producer Katie Ferguson. We'll be back later this morning with TNB Tech Minute. Thanks for listening.
- [12:29]
E
How are artists, musicians and filmmakers using AI tools to help bring storytelling and creative expression to life. To find out, search for Google's new series A New Era of American Innovation. Wherever you get your podcasts.