|
Here are edited highlights from that part of the discussion.
WSJ Leadership Institute: You argue that technology may be more important than a company’s balance sheet. How so?
Krishna: It’s a simple way of saying, “what's most important for a company to survive?” If you look at the long arc of history, it used to be physical security. Great, I have armies, I can protect myself. I have lots of people. Fast forward, it kind of went to the knowledge economy, the strength of my workforce, the talent that allows us to succeed. I think now it's technology that allows people to succeed...
Pre-internet, technology was probably a productivity (driver) and a cost saver. I don't think it allowed you to triple your business if you could use technology well. Now, you can.
WSJLI: You elected not to join the race to develop frontier large language models. Why?
Krishna: Many will thrive, some will disappoint. So that goes to the first question, is there real value? And the answer here I think is unambiguous. There's absolutely real value. We can use this to make companies far more productive. We can use it to generate a lot more revenue.
Why do I say that some will disappoint? I think that very large models -- you can switch between them. It is a cruel way of saying that it is a commodity. But it is a commodity like gold is a commodity. It's a valuable commodity. But if that's true, that means that the ones who are competing can only derive a certain price, they can't just take price … to the heavens.
So then I just do some simple math.
Over 100 gigawatts have been committed over the next three to five years.
It takes about $80 billion dollars to completely fill up a gigawatt (data) center, and I'm not counting the power investment and all that because that is often outside these companies' balance sheet. But filling up the data center is usually on a tech company's balance sheet.
If you put that much capital in the ground, what do you need to pay for it? So you need on the order of $2 to $4 trillion of revenue per year. Is that all moving out from current tech? Can I have that much incremental? I think that is tough. Maybe $1 trillion could be incremental. So that says the equation or the economics to me, doesn't close fully. The data center is good for probably 15 years, but what's inside you need to refresh every five years. So you need to get your money back in five years.
I for sure think a couple of these people (frontier model developers) will be very, very successful, out of the half dozen or so who are doing the spending. So some will disappoint.
WSJLI: Why isn’t that logic apparent to all of the players in the market?
Krishna: Maybe it is. But it's like this. In my downside scenario, there’s still $1 trillion dollars of extra revenue. If you think you have a chance of getting that trillion, wouldn't it behoove you to spend whatever it takes to get that?
WSJLI: What’s your framework, as CEO, for assessing the right level of risk for IBM?
Krishna: I actually have a core belief. Eighty percent of the time, good management knows what to do. But only 20% of the time is it actually done because you're going to be disturbing people.
You're going to be breaking loyalties. You've got to turn around and say, I need to change the business model. You need to turn around and say, I need to cannibalize my own existing revenue, and people find it really hard. I have a fundamental belief that the riskiest part is taking too little risk. Actually, you have to take risk. But don't make the risk so big that it is existential to survive.
|
