You have to wonder if the economics of it all can be justified. With so many in the AI space in a hurry to put up huge data centres, Vembu and IBM CEO Krishna are cautioning that the price tag might not be worth it. Vembu calls it being prudent; he says Zoho is going to put its money where it counts: on verification, curation and reinforcement learning, not on raw scale.
It puts a fine point on what is on the mind of Big Tech and their investors: can you really make the numbers work on an AI build-out of this size, or is the market running ahead of itself when it comes to demand and bottom-line results?
A contrarian bet on the AI value chain
Vembu put some of his own spin on Krishna’s warning in a post on X, to show where Zoho is headed. We’re after the things that make AI doable and you can put your trust in it, not just more compute. He pointed to near-term work on data curation, compiler infrastructure and the like as a way to vouch for what the AI is putting out there.
“We are investing in creating capabilities like data curation, reinforcement learning, and most crucially the compiler infrastructure to ensure AI output can be verified but we will not chase the investment bubble. This is just our normal prudence. To some people that would sound defeatist, but we will talk in 5 years,” was how he put it.
That is a different ballgame from the way some of our competition is throwing good money at data centres and chips. For us, the edge is in having something you can rely on and verify in the field, not in how much capital you can put to work.
Why this matters now
There is a lot of head-scratching over whether the current round of building makes sense in light of how fast (or slow) adoption is actually happening. Vembu is only echoing what Krishna has been saying about the risk of a bubble in the multi-trillion dollar data center market.
The numbers testing the bullish case
It all comes down to cost. Futurism has it that a 1-gigawatt AI data center these days is an $80 billion proposition. If you are in the 20-30 gigawatt range, you are looking at $1.5 trillion.
When you look at the whole industry and the AGI ambitions driving it, you could be talking 100 gigawatts and some $8 trillion in capex. Krishna’s take is you need $800 billion in profit just to service the debt on that kind of deal.
Then you have cloud veteran David Linthicum, who made no bones about it in a piece he called “The Emperor Has No Clothes.” His view is that the math is off because enterprises are moving slower than you’d think and consumer AI is still trying to find its footing.
"The math just doesn't add up,” he wrote, comparing the current fervour to the dot-com and metaverse hypes. He sees hundreds of gigawatts in the pipeline over the next ten years and trillions in spending to go with it.
Competing playbooks and what to watch
Of course, not everyone is on board with the naysayers. There is a feeling that AI is being used in spades, with some power users in front of their tools for hours a day. For them, having the compute and the models is a no-brainer for the long term.
So you have a split. One side is in a race to put together as much capacity as they can. The other is focused on the software and the quality of the data, with the idea that you get better value from a deployment you can count on.
Vembu has been on this before. He has put to rest the notion that AI is the only reason for job cuts, and has said the bubble isn’t going to hold up the economy for ever. You don’t see him buying into the hype around productivity without some hard proof to back it up.
Here is where the debate stands:
– Krishna has his eye on a possible bubble
– Vembu has no interest in the arms race on infrastructure
– Linthicum is not convinced by the projections
– And then there are those who see a surge in adoption
For the CIOs and the money men, it is a matter of seeing some evidence. Can you hit the revenue marks to cover the costs? Or is the way Zoho is doing it, with a focus on what can be verified, the more sensible path to a return?
Vembu has put a date on it: let’s check back in five. Until then, the thing to keep an eye on is whether the profits are as good as the word on the street about the AI infrastructure of today.
