Rise of AI Conference 2025 · Berlin · Applied AI Stage

Making AI Useful

Thomas Schmidt · Amber AI

At Rise of AI 2025 in Berlin, Thomas Schmidt, co-founder of Amber AI, an Nvidia elite partner founded in 2006, argues that the industry must move from AI hype to real, useful impact. He traces the field from AlexNet in 2012 through Nvidia's 2016 shift into AI-focused hardware and the 2022 ChatGPT moment, citing figures that generative AI returns 3.5 US dollars per dollar spent on average, and up to 8 times for the best-performing companies. He argues much enterprise infrastructure is still outdated CPU-based hardware, and that moving to current-generation GPUs, especially Nvidia's Blackwell architecture, cuts power use and cost while shifting workloads from training toward inference. Schmidt presents a human-centric design framework for making AI actually get used inside companies, emphasizing starting from real business pain points rather than technology demos, and describes three concrete use cases: accelerating patent-lawyer document review, optimizing logistics and cargo loading, and using computer vision to identify chick sex in eggs before hatching to stop unnecessary culling. He closes with a call to prioritize trust, impact and workflow integration over model sophistication.

Key takeaways

  • Schmidt cites generative AI ROI of 3.5x on average, and up to 8x for leading companies, per dollar spent.
  • He argues moving legacy CPU workloads to current-generation GPUs cuts both cost and energy use.
  • A patent-review use case cut document analysis time from one to two weeks down to minutes or hours.
  • A computer-vision use case determines chick sex from eggs before hatching to eliminate unnecessary culling.
  • He warns companies most often fail by over-engineering solutions and running endless pilots instead of fixing one real pain point.

Detailed notes

Amber AI and Nvidia's early GPU-for-AI bet

Schmidt introduces Amber, an Nvidia elite partner founded in 2006, when one of his co-founders, then at the Technical University of Munich, was among the first to code on GPUs and worked directly with Nvidia's Jensen Huang on early projects with BMW. Amber, then called Fluidyna, was acquired by the US company Altair; in 2022 the AI team spun back out as Amber to chase Nvidia and AI as the next big opportunity, while Altair itself was later acquired by Siemens. He notes Amber was Nvidia's 2024 Rising Star partner and won 2025 Nvidia Partner of the Year for Central Europe.

From AlexNet to ChatGPT: the state of AI and its ROI

Schmidt sketches the timeline from AlexNet in 2012, through Nvidia's 2016 decision to build dedicated AI hardware after Dell and HPE had not yet seen the opportunity, to transformers and the 2022 ChatGPT moment that brought AI into mainstream attention. He cites figures that generative AI delivers an average return of 3.5 US dollars per dollar invested, with leading companies reaching up to 8 times return, that an estimated one-third of companies expect to double monetization using generative AI, and that by 2028 much more coding and software testing will be done through generative AI. He stresses generative AI is only part of what AI can do.

Infrastructure economics: from legacy CPUs to Blackwell

He argues many customers, from small firms to large corporations, still run legacy CPU infrastructure that could be GPU-accelerated, and that moving workloads to GPUs would cut cost, energy use and heat generation even as AI overall consumes more energy. He contrasts a decade-old Titan-based system worth roughly 130 million US dollars with a DGX/HGX-class system from two years ago worth about 400,000 US dollars, delivering similar power at a fraction of the cost and energy. He notes most current deployments still use Nvidia's Hopper architecture, but that production is shifting to Blackwell, which he says Jensen Huang himself describes as leaving no real reason to keep buying Hopper, particularly because Blackwell's efficiency gains are larger for inference than for training, and the industry overall is shifting from training toward inference workloads.

A human-centric framework for making AI useful

Schmidt argues that past the peak of AI hype, many pilot projects went nowhere because they were built as tech demos rather than around real customer pain points. He proposes a human-centric design framework: understand the specific problem, engage the humans who will use the system, and iterate based on their feedback rather than aiming for 100% accuracy. As an example, he describes a customer-service chatbot now in its 53rd iteration, continually improved based on user complaints, until it reaches a point where users prefer it to alternatives. He adds that in Europe, AI also needs to be explainable rather than a black box, and should be accessible to every relevant worker rather than a narrow team.

Three use cases: legal, logistics, and hatchery ethics

Schmidt describes working with patent lawyers, where customized models for text and drawings, combined with human validation, cut document-review time on some projects from one to two weeks down to minutes or hours. In logistics, Amber helped a customer optimize the packing of large concrete blocks onto lorries and used computer vision to inspect cargo safety, reducing fuel and vehicle needs. A third example addresses the culling of male chicks in the poultry industry: using deep learning and computer vision to determine chick sex from eggs before hatching, eliminating the need to hatch and then cull males at all.

Q&A highlights

Asked about data storage and networking as a bottleneck, Schmidt says customers often underinvest here relative to compute, comparing it to buying a Ferrari and fitting it with tiny wheels, and names Dell, DDN and NetApp as partners working to keep pace with Nvidia's networking speed. Asked about the biggest pitfall in shifting from 'clever' to genuinely useful AI, he says companies commonly over-engineer solutions instead of simplifying, should prioritize quality data over quantity, and often get stuck running pilot after pilot without management involvement to fix real pain points and move projects into production.

Quotes

AI should have a focus of human-centric design framework.

Thomas Schmidt

You're not buying the Ferrari and then put tiny wheels on it and not get the power on the street.

Thomas Schmidt

A lot of companies over-engineer. Sometimes you don't have to have the perfect, super accurate black box that will fix everything.

Thomas Schmidt

Questions answered in this talk

What return on investment does generative AI deliver?

According to Thomas Schmidt at Rise of AI 2025, generative AI delivers an average return of 3.5 US dollars for every dollar spent, with leading companies achieving up to 8 times return.

Why is data storage and networking often a bottleneck in AI projects?

According to Thomas Schmidt at Rise of AI 2025, customers frequently budget heavily for compute but overlook storage and networking, comparing it to buying a Ferrari and fitting it with tiny wheels, then having to request additional budget for networking and storage later.

What is the biggest pitfall companies face when trying to make AI useful?

According to Thomas Schmidt at Rise of AI 2025, the biggest pitfalls are over-engineering solutions instead of solving one real problem simply, prioritizing data quantity over quality, and running repeated small pilots without management commitment to scale them.

Mentioned

Generative AI ROI, GPU-accelerated computing, Nvidia Blackwell, Nvidia Hopper, Human-centric design framework, Inference vs training, AI Factory, Computer vision

Keep exploring

More from Thomas Schmidt

More from Rise of AI Conference 2025

More on Enterprise AI & Adoption

Transcript

Auto-generated transcript, not edited. · 4,561 words

Show the full transcript

Actually, I don't I don't mind the discussion that went on afterwards. We're actually a Dell partner, so that's perfect. And I can confirm a lot of the stuff that that has just been said. Um, and I'll actually get to the point um about pain points and how to make AI useful actually. And I've I've added from hype to real impact because that's that's what we've been trying to to do in the last few months and weeks. Which button? This one. Yeah. Just a few quick words about Amber. Um, Amber is an Nvidia elite partner. We've been founded by my two co-founders in 2006 as a different um kind of beast back then. One of my co-founders is a professor at the technical university of Munich and he was one of the first who just saw GPUs and he decided let's code on GPUs and back then Nvidia was a totally different beast that it is today.

Um, and and actually he went together with Jensen to BMW and and stuff like that. So it was a very different time. Very cool. But um that that kind of put us into the forefront of GPU and AI development very very very early. Um so we became actually one of the first NVIDIA elite partners in 2016. Uh this means we had to learn everything about AI, how to how to GPU accelerate a lot of um software and and and help our customers. Um, Fluidina was then bought by a US company, Alter, because they do simulations as well, and they were we're on the same trend. And a few years ago, 2022, we actually merged out of uh Alter because, well, we saw NVIDIA and AI as the next big thing that we want to want to chase.

Um, and Alter had a bit of a different view on this. Um, who knows who's going to be right in the end, but Alter actually has been bought by Seammens. So, it's coming back home to Germany kind of again. So that's a fun story. Um, and we have been Nvidia's rising star partner last year. That's usually like the the the new startup. Um, which is funny because we've been around since the very very beginning. Uh, but but uh, according to the new merge out, we we were eligible for the rising star and this year we won the Nvidia partner of the year award for central European partners. Yes, it was very fun. Um, so I I I think most people here know the AI story, but uh you never know who's who's listening in.

Uh, so to just have a very very quick catch up where we are. Um, the state of AI, I mean, it really started in 2012 with AlexNet, right? And what else could AI do back then but, you know, find out what are cats in pictures. That was the internet back then, I think. Really fitting. Uh, but that kind of just just kickstarted the whole AI journey. Um uh and and since then I I wouldn't even call it a roller coaster. I would call it a rocket. It has only been going up. It has been it has been accelerating so much over the last few years especially uh since 2016 when Nvidia decided to build their own um AI focused hardware because unfortunately Dell, HPE, all of these big players, they didn't see the market yet.

Nvidia saw it, they did it and they convinced everybody globally I think. Uh so we are where we are today. Um then of course we had transformers everything just accelerated more and more until 2022 happened and all of a sudden chatgpt changed everything and AI was you know in everybody's mouth. Um for us it was funny because we were talking about this thing that's been happening for the last few years but ChatGPT kind of you know brought it to everybody's attention and mind and since then I I think everybody has the same problem. It's hard to keep up. every few weeks there's a new model, there's a new technology, there's a new um best practice.

We've moved from fine-tuning to rag. Uh we're now seeing multimodal model model uh mod models. We're just seeing amazing developments constantly and with that we see a lot of uh startups emerging, a lot of um technology just appearing and use cases being brought up and Jensen always says we have a huge market which we can all address together. Um, and you know, I personally I'm not a fan to always only talk about Gen AI if uh if we're talking about AI in general, but Gen AI is the big thing in the last few months or in the last year, I would say. And what we've seen in Genai and what the numbers show is that it's worth it.

It's worth spending money on Genai. The return of invest is amazing with 3.5 for every dollar spent. And some of the key players out there who really understand what they're doing, uh, they get even up to eight% of an R ROI on their Genai projects. Um, estimations say that a third of all companies will basically use business models to double their moni monetization with Genai. Um, and by 2028, we'll see a lot more coding being done and software tests uh coming in via Genai as well. But again, and I will get back to that later, Genai is an amazing technology is one part, but it's not everything AI can do. Um, what's very important for us as well, AI is to to make AI useful, I think it has to be useful both on the sustainability, it has to be useful on an economic side.

Uh what we still see with a lot of customers and we are talking from small-time customers the backbone of Germany up to the big ducks companies everything globally um we still see a lot of legacy hardware CPU uh systems being used out there right it has been bought one day and people just want to continue using it and don't understand me wrong not everything can be GPU accelerated but a lot of things can be GPU accelerated which aren't today and um everybody sees the headline lines all the time. AI is eating our energy, is heating up the planet. True, it needs a lot of energy, but if we would actually move a lot of this legacy infrastructure to GPU accelerated um infrastructure, we would save money, we would save energy, we would save heat, and it would help us solve a lot of the problems.

And I it's always fascinating when we talk to new customers, even big customers, and they they want to build an AI factory. Um, and maybe they don't always start with the very big uh size, but they are used to the left picture, which is just rows and rows and rows of of racks of servers. And then they ask us uh how how what do we need to run our use case? And then they get the right picture. And that's not very good for PR and for for LinkedIn. But it is what it is, right? Um, and I personally think we should we should turn the view away from we have this huge data center that's just full of GPUs of of CPU servers to to look what we can achieve with such a small actual system.

And um, if we're talking about accelerated computing, it's not just important to just move it from CPU to GPU. It's also important to keep up with the times. Nvidia is doing some amazing stuff in developing their GPUs. And if we actually compare it's a decade old system Titan with just a one DJX or HDX system from from two years ago. Um we see you mean I mean 130 million compared to 400K half a lot less power a lot less power not just half a lot less GPUs. Um so that that just showcases uh you need to go with the times. You need to actually stick uh with the with the newest generations of GPUs. Otherwise you will uh lose again a lot of power that is unnecessary.

You will you will lose money and this just doesn't make sense. Um, and for a more current comparison of Hopper to Blackwell, for example, I mean, we still see most deployments right now uh are with Hopper, but everything is moving towards Blackwell as production is um catching up. Um, and for good reason. I mean, the new Blackwell GPUs again will use much less power. Uh, will need less space. Um, are an absolutely incred incredible systems. We will soon see ultra blackwell very soon. So the GB300 for example or or B300 um that will be even better. But what this does not show is what is very important nowadays is we're moving away from training. This is just showing training right now for ch for a 1.8 trillion model.

Um but we're moving into inference a lot and Blackwell is such an improvement on the inference side that there is and this is what Jensen says himself there is no no real reason to buy Hopper anymore. Um it's available. Please keep buying Hopper, but Blackwell is the way to go. Um and of course if you can go with liquid, we can we can with Dell and everybody we can we can uh as well help you there in getting this done. But this is this is just from the e economic side and from the power side. If you want to make AI useful, if you want to get um your money's worth back, then you need to make sure you you bet on the right infrastructure.

So where are we today? 2022 ChatGPT the explosion started 2023 we basically saw all of these startups coming up all of these use cases everything developing very very quickly and 2025 is the year we're seeing stuff just move into production actually um being used by companies g getting the money back um and this means as well going away from training these large models to actually using them going towards the inference and if we look at the computational requirements we just see an enormous growth over the last decade and we don't see this this stopping anytime soon. So to the actual point of the presentation, why AI should be useful? Why what does it even mean for AI to be useful?

I think everybody knows the technological the technology hype cycle. Um I see us getting past the peak. Everybody was super excited about AI two years ago. It was on every conference. Everybody was talking about AI constantly and um we saw a lot of pilots exactly what we what we just heard Dell say. We saw a lot of pilots which moved nowhere. Um and then and then we got to this point where uh is AI actually what we should use in our company. Is it something that can help us? So we we saw a lot of this excitement uh die down. AI was everywhere but it wasn't very helpful for a lot of customers. We saw tech demos everybody everybody saw tech demos.

I could ask the same question. Who saw techs on some laptop uh which never actually developed anywhere? We saw them all the time. Um so what we found out is you need to talk to customers and you need to understand why they should be using AI which with with which problems can can AI actually help understand the pain that the customer is going through. understand how AI can help them be a competitive and how they can get a competitive advantage towards others and how it can help uh key problems um that they're facing. And of course, AI has to be useful. It doesn't have to be for a lot of customers, it doesn't have to be 100% accurate.

It needs to be usable by the users within the company. So that's exactly the problem we're seeing that AI gets developed and then the teams are saying I I'm going to use it. I'm going to do the training. I don't want to use it actually. So it needs to be very practical. It needs to be focused on the humans using it. It needs to have the UIs and and and make them understand why it actually helps them. It needs to be responsible and it needs to be trusted by everybody in the company um to actually be useful and to be used by the people inside it. And that's why we always say um AI should have a focus of humanentric design framework.

That means um you need to you need to plan around the problems within your company. You need to get input from humans. You need to empathize empathize with them. Uh you need to engage them. You need to find out how does it help them, where does it not help them, which problems arise through um through AI for example. Uh we have customers who want to use AI for customer service, right? And they use it and then they they get feedback from their customers and say customer service your your chatbot is horrible. I I don't want to use it. Why is it horrible? You need to figure that out because some questions aren't asked by the chatbot for example because it just uh gives um you know bad bad answers and you can figure that out.

You can improve and iterate and just yesterday I talked to to someone who told me we have this one project with a customer guess what stage the chatbot is in 53. So it's the 53rd in uh version that they're working on right now and it just keeps on getting better and at some point you get to the inflection point where customers and the users actually um prefer it. They prefer using this compared to other solutions or sometimes even humans and and this needs to be a loop. you need to constantly uh improve and understand how your AI systems work within the company and how you can improve it and then um make it more and more useful for everybody.

And again, the main thing that everybody should take from this is um there are there are amazing models, there are amazing things you can do with the AI, but what you really need to figure out is what do you need within your company? What problems can you solve? And this can be very broad. Uh I will go towards some use cases but um think about all of the demographic change we have in our society. We have people getting older. We need new people working. Um a lot of people are scared for their jobs but at the same time we actually need AI to make sure we are keeping our productivity, keeping our efficiency uh and can actually make sure that our trains will still run in 10 years for example, right?

The train conductors are missing. Uh this this means this is a huge pain point for example which needs to be solved which cannot be solved by anything else but AI unless you pour a ton of money into it and and make it uh make them make them rich like pilots. AI of course also needs to be reliable, safe and consistent. We are in Europe um and I prefer our approach. I don't think we should overregulate everything, but um I don't think we we can we should do it like in the US where we just say hands up, let's just do whatever you want. Um we need to make sure that the eye AI we're using is understandable and it's explainable.

AI should not just be a black box where we put something in and something comes out and we're saying, "Oh, okay. This is great." Um and it needs to be integrated into the workflows of the of the company. I mean, right now we're we're hearing a lot about Agentic AI as well, where it can take over entire workflows. Um, but let's let's get to the point where we're actually integrating the tools into the workflows and then keep on going from there. And again, it needs to be accessible within the entire ecosystem of your company preferably, right? You want to have every worker being able to use the AI tools um within your company. uh just a few use cases which I found interesting which we've worked on in the last years uh which maybe give you a little bit of a hint uh what what's what what I mean if I say uh think about the problems within your own company for example I thought this one was very interesting um we were working with

some lawyers and um especially patent lawyers um who have a whole ton of very complex and hard to read documents um which takes a huge amount of time to to solve. It has sensitive data um and is very very expensive. Lawyers generally are uh so this is this is a point where we it's perfect for AI to to find a solution to help you with that. Um so you you could use customized highly customized models um with with dedicated models for the text and the drawings to understand these documents. Uh then let them work together with humans uh to validate uh the findings and this actually helped the customers um you know reduce time for use cases for for projects from one to two weeks to minutes or hours.

Uh a huge boost in productivity, a huge boost within the the the companies. Um and and of course you understand this this helps them be competitive towards uh against others. Um I personally I really like the everything around logistics like I said uh it's not always all about genai. Um I really like everything supply chain management related logistics because this has a direct impact on on our on the energy required not just to run the systems but for example if you can optimize your warehouses if you can optimize your lries if you can optimize um how much how much traffic you're generating this this this helps our planet at the same time uh while you're also accelerating um the the uh your your workloads.

So this is another challenge we had uh with a customer who says right uh we need to we need to plan very large concrete um blocks for example within lries we need to plan them for warehouses how can we optimize this via AI AI is great for this um and at the same time once it is all on the lorry we want to have some computer vision to inspect whether it is safe to move um and uh of course this is also a computer vision uh thing you can do which was done and helped them a lot in just um you know scanning the cargo pieces. How big are they? How many will fit on the on the uh on the lorry.

This will this will reduce the the um fuel they will need the the amount of lorries they will need. It makes them more competitive towards their um yeah competitors. Third uh and last more detailed at least um use case which I also very like because it's more something on the ethical side and the um yeah again stopping waste in in other areas. Um for for example we have a customer who's working on stopping the um chick calling from male chicks. As you may know that uh male chicks are not really needed. So they they are usually killed after they are hatched. Um, and they found um via computer vision again a way to analyze the the eggs while they're still just um well not not hatched yet.

Um and that that helped uh just eliminate this entire problem. Uh these male c male chicks don't have to be called anymore. Um deep learning image processing was used for it. Um and and no more breeding is required on on on these. And again, a problem. You would probably not immediately think this is an AI problem. This is not a chatbot I'm using to fix this. Um, but it is an AI problem. You can you cannot fix it via any you you couldn't put humans on the on the line and then put a light in there to understand which are the mail chicks. Uh, you need to use AI um to fix a problem like this.

And this is exactly where where you need to look and and and work on. There are a thousand more use cases um in every vertical in everything um in every industry whether you want to talk about predictive maintenance education. Education is going to be a very very fun and huge uh thing that that can be empowered by AI. Um another favorite of mine is everything healthcare and farmer related. Drug discovery, personalized drug discovery will be a big thing. This is actually Genai. It's just a different form of of language model um than of course um X-ray scans, computer vision. Again, there are just tons and tons of tons of things you can do. And pretty much every single company out there, every single enterprise, um they they have pain points that can be fixed via AI that need to be understood.

Call to action uh for everybody out there. Well, please continue building clever AI. This is great. continue building better and better models, but also think about how do you build useful AI. It doesn't have to be uh the model that can do everything. Maybe it can do a very small thing perfectly fine for you. Um this will fix the amount of power it needs. This will fix a lot of things. Um start with real problems. Take a look at your company. Take a look at your customers and and figure out with them how to fix their problems. Prioritize trust, impact, and integration. And of course collaborate between the techies and the humans. If you if you if you can manage to combine these two areas, combine the uh domain expertise of people with the technical expertise, uh you will find the the moonshot um solutions that you're looking for.

And of course um reach out to Nvidia, reach out to us, reach out to Dell. There are a lot of people in this ecosystem who can help you. There are a lot of things that have already been built that you can use uh very quickly. Uh you will get support in building a solution that is fit for yourself. Um just reach out, use the ecosystem that's already here. That's it. Thank you. Thank you, Thomas. Any questions? Oh, I mean we don't have much time but there is one question and there's a microphone coming to you. Sure. Thank you. So I have a short question. So Nvidia is faster and faster and faster. That's good. What's about data storage which is usually a bottleneck which is very slow and and accessing the data in general.

is uh so data storage and networking um both things that a lot of customers don't have in mind when they're planning for AI clusters and AI use cases in general. It's very important. Um I always like to say you're not buying the Ferrari uh and then put tiny wheels on it and not get the power on the street. It's very important. There are different solutions. We work with pretty much everybody and again also a lot of development there whether it be from Dell, DDN, NetApp, everybody is is constantly developing but it is hard to keep pace with Nvidia and especially their their uh networking speed but they are working on it. So again everybody please think about storage as well because it's usually oh I have this budget I want to get as much compute in there as possible and then half a year later oh I need more budget for the networking and the storage and that's usually a pain but um we try to we try to um tell the customers early on

to think about it as well. I obviously also prepared a question and you talked about uh useful and clever. Well, if both together, that's amazing. What's the biggest pitfall company face um to shift to the useful part and not just because it's cool technology? Yeah, a lot of a lot of companies over engineer. So, sometimes you don't have to have the perfect super accurate blackbox that that that will fix everything. Like I said, try to try to simplify. Try to find a problem and fix it and don't try to make it super super complicated. Um data is very important. Um go for quality data instead of quantitative. Don't just throw everything in there, but make sure you you have the data that's actually useful for your use cases.

Um yeah, and don't and don't get don't get into the pitfall of just I'm running pilot after pilot after pilot, right? Uh so it's important to get started but um there you need management to really be involved and for that you need to solve their pains and if you just do some small pilots then this usually does not develop into something bigger. Talking about pilots you talked about um car um trains and buses I can tell you from um insider knowledge pilots are also short. Yeah, we don't have enough. And in 2030, some of the companies and airlines are talking about a co-pilot that might be AI, an actual co-pilot, an act like the actual co-pilot AI, which pilots fear because um there is a reason why there are two people in the cockpit.

Yeah. Um we'll see where that leads to. Of course, the next 10 years will be very interesting. Um I don't think we should fear the developments that are coming. I think uh we need to embrace them. They are coming either way. and they are needed to fix the problems we're facing. Yeah. Yeah. Um short labor here. Um and thank you for bringing the um the story of GPUs because in 2018 I was moderating a conference in Los Angeles with Nvidia and they brought it to Los to Hollywood. So from Silicon Valley to Hollywood and they said GPUs will be the future. We are living in the future. Thank you so much. Thank you as well.

Thanks. Oh hang on. I need your the torch or the to or you give it to see I wanted to remember that word but now the torch has been has to be handed over to two people on

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.