Lead times for Nvidia's AI GPUs narrow, suggesting expanded production capacity

[Shawn Knight]

Why it matters: Wait times for Nvidia GPUs used in AI applications have narrowed significantly in the past few months, which could be an early sign that growth is peaking. Depending on your view, that can either be good or bad news.

In a note to investors, analysts with UBS said lead times – how long a customer has to wait for delivery after placing an order – have come down considerably as of late. Seeking Alpha saw the note and said lead times are now in the 3-4 month range, down from 8-11 months late last year. So, what's happening?

The publication highlights two possibilities. Either Nvidia is planning for future production capacity increases to outstrip future orders, or the company is currently shipping products in excess of the orders it has already received. Given the significant drop in lead times in such a short period of time, Seeking Alpha suggests both options are likely.

What does it all mean? Well, it somewhat depends on your point of view.

Inching closer to eliminating lead times means Nvidia is also getting closer to the end of its AI boom, and that's something neither Nvidia nor investors want. However, if the decline in lead times is indeed coming from increased production capacity, that means more products are being shipped now and more revenue is being generated – great news for the company and shareholders.

At the current rate, it would seem as if Nvidia will work through its backlog of orders in the next two to three quarters. Once that happens, they will only be shipping against new orders and revenue will likely take a bit of a hit.

Whatever the case, UBS raised its price target on Nvidia to $850.

Earlier this week, if you recall, Morgan Stanley bumped its price target from $603 to $750 per share, which is starting to look conservative. As of writing, Nvidia shares are trading at just over $739, giving the company a higher market cap than both Amazon and Alphabet.

Nvidia's quarterly earnings report is due on February 21 after the bell.

Permalink to story.

https://www.techspot.com/news/101919-lead-times-nvidia-ai-gpus-narrow-suggesting-expanded.html

 

[opckieran]

This has to be increased production/shipping. “The end of the AI boom” is wildly overstated. AI shows no signs of slowing down, and OpenAI’s Sora that was just announced is going to need at least an order of magnitude or two more processing power than something like Stable Diffusion. Workloads are only going to increase exponentially, hence large contracts in the 10,000s or even 100,000s of $30k units being awarded to nVidia.

 

[hahahanoobs]

One could argue OLED TV's have peaked.
Sales could slow, but AI isn't going anywhere.

 

[godrilla]

TSMC's growth seems to be also affected by Nvidia's success. TSMC is the one helping them improve production.
https://www.digitimes.com/news/a20231213PD217/tsmc-nvidia-ai-chips-china.html

https://www.cnbc.com/2024/02/15/tsm...igh-after-morgan-stanley-upgrades-nvidia.html


Update on Blackwell rumors being 60% better in Rasterization and 2.5x in rt but may launch in q2 2025 now. Take with witha whole server room of salt.

NVIDIA RTX 5000 GPUs could be 60% faster - and 2.5x for RT - but may not arrive until Q2 2025

NVIDIA is focusing on ray tracing with a rumored 2.5x gen-on-gen performance boost as its target - but next-gen GPUs may turn up later than we thought.

www.tweaktown.com

 

[Watzupken]

This has to be increased production/shipping. “The end of the AI boom” is wildly overstated. AI shows no signs of slowing down, and OpenAI’s Sora that was just announced is going to need at least an order of magnitude or two more processing power than something like Stable Diffusion. Workloads are only going to increase exponentially, hence large contracts in the 10,000s or even 100,000s of $30k units being awarded to nVidia.

You may have over simplified the issue. You need to understand that a huge chunk of orders in the past were to China. Hence the timing of this news vs when the sanction was in place is not a coincidence. Nvidia will prioritise orders to China knowing that they can sell at exorbitant rates, but will not deter Chinese buyers who were hoarding these cards in anticipation of further sanctions. Huawei did that when they were sanction from using TSMC’s cutting edge node. So no surprises here.
The other thing about expanding AI processing power is not just about the ability to buy enough hardware. These hardware draws insane amount of power and it does not mean your power grid can miraculously accommodate the surge in power usage. So whatever expansion will not be happening overnight.
I do feel that demand will slow down due to the fact that most companies looking to dabble in AI, would have secured some hardware. Unlike 2 years back where AI hardware may not be common. Therefore, it should not be surprising for demand to slow down.

 

[opckieran]

You may have over simplified the issue. You need to understand that a huge chunk of orders in the past were to China. Hence the timing of this news vs when the sanction was in place is not a coincidence. Nvidia will prioritise orders to China knowing that they can sell at exorbitant rates, but will not deter Chinese buyers who were hoarding these cards in anticipation of further sanctions. Huawei did that when they were sanction from using TSMC’s cutting edge node. So no surprises here.
The other thing about expanding AI processing power is not just about the ability to buy enough hardware. These hardware draws insane amount of power and it does not mean your power grid can miraculously accommodate the surge in power usage. So whatever expansion will not be happening overnight.
I do feel that demand will slow down due to the fact that most companies looking to dabble in AI, would have secured some hardware. Unlike 2 years back where AI hardware may not be common. Therefore, it should not be surprising for demand to slow down.

I’d love to see the studies that show this alleged new overnight overburdening of power grids due to new AI data centers. You’re talking a tenth of a percent at best, and ignoring that fact that such purchases could actually be REDUCING the amount of power required due to phasing out old systems in favor of new. In fact, most of the biggest customers (M$, Google, Meta, Amazon, Netflix) for these datacenters already have or are on track to powering their datacenters entirely with renewable energies.


As for China, nVidia is already producing skus which adhere to the sanctions for the Chinese. So China isn’t even going to be lost as a customer any time soon, they will just be sold a different set of products. Now, maybe in 6-10 years when China manages to catch up to the rest of the world, nVidia MIGHT have something to worry about.

In the end I think it’s highly speculative to assume that these companies’ appetites for AI (compute) have been satiated for the time being. I again point to the newly announced tools which will require much more processing power to prove my point. And that’s just what has been made public. We haven’t even scratched the surface of what can be done yet. What we the public know about and have access to is barely the tip of the iceberg. Netflix isn’t offering $900K salaries for AI leads just to run public code.

 

[umeng2002]

Other than making interesting pictures and videos, has AI actually done something?