3 ms·
I think they're planning for a world where half their customers (hyperscalers) just use GPUs and CUDA while the other half (the long tail) use more profitable,
by wmf 3y ago
I think they're planning for a world where half their customers (hyperscalers) just use GPUs and CUDA while the other half (the long tail) use more profitable, higher-level parts of the platform. They don't have the leverage to force customers one way or the other. It would be easier to just sell GPUs, but they know that sophisticated customers can switch to other chips while the platform provides lock-in for smaller customers.
- neverokay 3y agoDoesn’t matter how good your in-house tech team was, your company still outsourced it to cloud infra. That’s what Nvidia faces. Doesn’t matter how good the current in-house teams are using direct hardware, the trend in corporate is shift to a vendor (Google/AWS). Nvidia can watch this inevitable shift or get ready to offer itself as a platform too.
- joshellington 3y agoI get it, the service model always shines the brightest in the eye of the revenue calculator. But I have immediate skepticism they’ll be able to execute at a competitive level. Their core competency has always been manufacture and production, not service-based things. It’s a big rock to push up a tall hill.
- neverokay 3y agoHey, if it doesn’t work out they can always return back to what they are, a consumer first company but now with world class hardware/software. AI graphics service that renders games in the cloud (photorealistic), concluding its epic journey of being an amazing graphics card company. Kinda cool when you think about it.
- omnimus 3y agoThis doesnt seem to be true. They have been running Geforce Now for a long time and its one of the best gaming streaming services. It seems they are doing it in partnership with other regional companies but nobody says they cant use same partners. Running games with small latency seems more complicated than llms on cuda.
- bboygravity 3y agoGenuine question (I don't know much about cloud stuff): how is providing a cloud service/platform (at scale) even remotely as hard as designing, manufacturing and selling GPU's (including drivers and firmware) at massive scale? It feels like reading that setting up something like Facebook would be extremely challenging for a company like SpaceX.
- westhanover 3y ago[dead]
- komadori 3y agoI share your intuition, perhaps unfairly, that it's indeed not as hard in absolute terms. However, it certainly requires a different set of skills and organisational practices. Just because an organisation is extremely good at one thing doesn't mean it can easily apply that to another field. I would guess that SpaceX probably does have the talent on hand to throw together a Facebook clone, but equally I think they would struggle to actually complete with Facebook as a business at scale.
- avidphantasm 3y agoWell, motivation would be an obvious thing lacking. People who want to work on rockets, I would guess, would find working on facebook to be a “boring” solved problem.
- throwaway2037 3y agoI like this question. You raise a very good point. Occam's Razor tells me the simplest explanation is "core competency". Running AI-SaaS is just a very different business from creating GPUs (including the required software ecosystem). As a counterpoint: Look at Microsoft. Traditionally, they have been pretty good at writing (and making money from) enterprise software, but not very good at hardware. (XBox is the one BIG exception, I can think of.)
- fragmede 3y ago
- xbmcuser 3y agoThey have been building partnerships with ISP's and service providers all over the world with their GeForce now game streaming service. They could continue and expand this by providing a similar backend for LLM services.
- bayindirh 3y agoThere are already platforms and projects which provide resources for long tail of science. EU is supporting projects which brings together resource providers with researchers and private companies in some cases. NVIDIA is not entering an empty marketplace here. Also there's already enough know-how to make things run on cloud, OpenStack, HPC, etc. Unless they make things very difficult for independent platforms, they can't force their way in that much, from my perspective.
- danpalmer 3y agoThe hyperscalers aren't going to be on CUDA for long, Nvidia are taking too much of a cut. Google run tensor chips, Amazon and Microsoft both have their own accelerators in the works. They are massively incentivised to do so right now. Now they do all re-sell Nvidia GPUs in their cloud businesses, but there the cost will be passed very directly on to infrastructure customers, who will see the competing higher level services from those cloud providers (hosted models) likely at lower or competitive prices, and it's going to be harder to justify renting CUDA cores for custom software.