I feel that one of the most fundamental aspects of business reality is being skipped in all of these conversations, on and off HN.
Everyone building an AI data center is likely using Nvidia technology. Sure, there's a 20% that is partially using other technology. The bulk of it is Nvidia.
If your project is in the planning for the next, say, two years, you have already placed your orders with Nvidia or are going to in the next few months.
Hardware has real lead times. You don't compile yourself 100K chips. They have to be made and you have to wait in line to get yours. For example, I remember when, during the pandemic, we had to place orders for chips with 40 to 50 week lead times.
This means you have to make decisions today (or you already made them months ago) to get in line.
Changes in training or inference efficiency should not change these orders or plans at all. If someone can train faster, they will benefit from the hardware in the pipeline. If they can make inference more efficient, they will be able to service more requests at reduced transactional costs.
The orders are in the pipeline and will continue to be added to the pipeline. Nvidia isn't going to be shipping half the hardware because Wall Street, overnight, panicked. What the grocery store owner does with their stock portfolio because they panic has nothing whatsoever to do with reality.
The same is true in the other direction. Wall Street has been going nuts with quantum stocks. Companies like Rigetti have exploded from nothing to see insane gains. This does not mean the company went from, well, shipping nothing to shipping real working solutions at scale.
Today's market reaction was nothing less than sheep running scared because someone when "boo!". It has nothing whatsoever to do with business realities on the ground. Go build an AI data center without Nvidia chips (or with 10x less chips) and see how that goes when everyone is loading-up with them.
Everyone building an AI data center is likely using Nvidia technology. Sure, there's a 20% that is partially using other technology. The bulk of it is Nvidia.
If your project is in the planning for the next, say, two years, you have already placed your orders with Nvidia or are going to in the next few months.
Hardware has real lead times. You don't compile yourself 100K chips. They have to be made and you have to wait in line to get yours. For example, I remember when, during the pandemic, we had to place orders for chips with 40 to 50 week lead times.
This means you have to make decisions today (or you already made them months ago) to get in line.
Changes in training or inference efficiency should not change these orders or plans at all. If someone can train faster, they will benefit from the hardware in the pipeline. If they can make inference more efficient, they will be able to service more requests at reduced transactional costs.
The orders are in the pipeline and will continue to be added to the pipeline. Nvidia isn't going to be shipping half the hardware because Wall Street, overnight, panicked. What the grocery store owner does with their stock portfolio because they panic has nothing whatsoever to do with reality.
The same is true in the other direction. Wall Street has been going nuts with quantum stocks. Companies like Rigetti have exploded from nothing to see insane gains. This does not mean the company went from, well, shipping nothing to shipping real working solutions at scale.
Today's market reaction was nothing less than sheep running scared because someone when "boo!". It has nothing whatsoever to do with business realities on the ground. Go build an AI data center without Nvidia chips (or with 10x less chips) and see how that goes when everyone is loading-up with them.