Great challenging interview. I like how Deirdre Bosa basically gave one of the biggest web companies the treatment others would give an up-and-coming startup. You can tell Adam Selipsky has a technical background from his responses and at times trying desperately not to go into jargon. These challenging questions to a person who has the technical background to answer them gives us some great insight.
There aren’t just one player H100 from Nvidia. Even for Nvidia GPUs, the most popular ones are probably still V100s since their price to performance ratio is good! Popular LLM models such as BLOOM is trained on V100. For inference, there are already a lot of players from edge to server side inference. Even Apple, with neural engine on every new Macs and iPhone, is a strong contributor to this space. There are lots of opportunities in the server inference space for Amazon to grow.
Companies should ask the question; do they need a large monolithic chip to perform the duties they require? It wasn't that long ago that an AMD GPGPU could run laps around nVidia's part at significantly less cost. nVidia's change in architecture recently won back their compute crown as both AMD and nVidia consolidated the way their GPUs work. There is a use case for a large monolithic die as there is a drop in performance when multiple dies are used. But it comes at a cost of lower yields and higher price for the die. AMD is more than happy not venturing into the large dies as they get the yields they want. Recently Intel announced they have a competitive iGPU. Not sure how that will play out on the larger scale.
Yes, if we are talking raw capabilities, H100 is the best, but he just couldn't say, yes it's good, our chip targets price/perf for specific use cases.
Adam clearly couldn't lie and embellish that Amazon Chips are as good as NVIDIAs because he knows that's far fetched. Amazon GPUs are no where close to NVIDIAs. Good job Deidre for pointing it out and getting him off guard and asking the right tough questions.
H100 got new Transformer Engine that is supposedly 6-9x faster than A100 for transformer models like GPT and BERT for training. Eventhough H100 is more expensive, the perf/price could be really good if you’re working with transformers. That’s still very new and isn’t going to be mass deployed until next quarter.
“The company that uses the innovation rather than the innovators will benefit and profit” - Peter Lynch. Soo aws, google and Microsoft have the data that has quality. These 3 companies stand to gain
AMZ is a value AI buy.. Wall Street is plowing money into AI start up and AI start ups are plowing money into Cloud utilities. Whos leading cloud computing? Amazon... When there's a gold rush, don't invest in the individual gold prospector, invest in the pick axe company. Cheers.
That's how what I felt too. Amazon seemed to have lost it's way. Microsoft and Google are way advanced in this AI race. Also Microsoft Azure is gaining traction on model training and inference race with medium and AI startups since they are powering ChatGPT and Bing AI. They also are investing in their own chipset and procuring from NVDA. I see MSFT at $460 by end of the year.
@@izdotcartershe really doesn't though. He made a valid point. Reporters may know some things, better sure. But computer nerds are gonna know their thing... And reporters don't know it. Reporters still say crypto is best used for money laundering. She was trying a got ya moment, and if she had understood the subject a little better she could have made her case. Not sure why you expect reporters to know tech well. It moves very fast, and lots of us are very very passionate down to insane detail, that let's you have a family/life, and us... Have hobbies lol.
Amazon sent a salesman to this meeting? He did not have a single substantive technical point to make as to how Amazon will counter Nvidia/CUDA. Just offering a supermarket of mediocre capabilities is not enough when the best models will remain closed source.
They didn't jump on this like everyone else. This was an aws service for years. 5 years old. Aws is the biggest profit driver of Amazon that drives the whole company. So yes.... This absolutely is real and isn't hot air. Amazon has a 25-30%ebitda cagr for a reason. That's why the stock is do valuable. It's still growing at an insane rate. None of this is new. Do you sleep through investor conferences? They explain these plans years in advance. Management's ability to execute on these plans is how they keep their job
@@Tential1 Maybe you should google what "sarcasm" means before you go on to "educate" people on things of which just about everyone is aware. My point is that lately all these companies have been in a race to tell everyone how cool their AI implementations are, which immediately pumps their stock price. NVidia and google have also been in AI for years too, yet it's only now the market seems to care due to the craze around chatGPT. Until now, Amazon had stayed silent and seemed to keep their distance form from all the shameless self-promotion going around (see CRM). Now, they finally bought themselves a ticket for that train, as well.
🪄✨ Made with SummarizeYT 0:22 - AWS launched Sagemaker, their machine learning platform, in 2017 and now have over 100,000 customers doing machine learning 0:50 - AWS scientists have been working on language based generative AI models such as LMS 1:13 - Alexa’s voice responses are driven by large language models 2:14 - AWS has designed its own chips called trainium for doing training and inferentia chip for doing inference 3:25 - Amazon is investing $100 million into a new generative AI innovation center 6:39 - Amazon has a managed service for doing machine learning called Bedrock 9:06 - Different models for different enterprises might need generative AI 11:05 - Amazon has been working on generative AI through different kinds of experts inside the company 13:11 - AWS is working on democratizing access to generative AI for every sector of the economy 14:46 - AWS CEO discusses the company's success as the leader of cloud services, having a big head start before its rivals caught on. 15:21 - AWS CEO believes that AI is going to be the next big thing and that they are focused on generative AI based applications. 23:03 - AWS CEO reveals the company's advantage over others in the GPU capacity market because they design their own chips and can provide additional value. 25:15 - AWS CEO confirms that the company is working with ARM, but there is still an option for optionality in their supply chain.
You can create a chip in theory. The issue is, manufacturing it. Tsmc has deals with apple, amd, Intel, and Nvidia. They always get first dibs. So, you compete at a 30% deficit, as the latest manufacturing process adds a lot. Nvidia and apple go as far as buying excess capacity to ensure you can't use it. Then, the entry cost into designing is around 10B. Essentially, no one really can compete with Nvidia... Hence why they have led their industry for 20 years straight. It doesn't even matter if you do create a chip better than Nvidia. Their software stack is also what matters. Everyone uses Nvidia, so even if a competitor is better, your software won't work with it, you won't get support, etc. Trust me dude, I spent the 15 years trying to avoid Nvidia products. But when you actually have mission critical work, you go with Nvidia. It's more stupid that you think competition will best Nvidia now, just because it fits your time narrative. Nvidia used to have tons of competitors. It has amd now. As the last man standing. There are a lot of bodies in Nvidias past. It's why I invested. I stopped fighting Nvidia, stopped trying to avoid the products, stopped begging for a competitor to dethrone them and their high prices. Instead, I bought the stock and the gains from the stock allow me to afford their very high prices.
The only people who seriously make their own chips are the majors. Amazon , Facebook, Google, Microsoft, apple. So if you want big tech to get bigger, yes, we can cheerlead them. Too cost intensive for anyone else
Why are you investing that much? It's suspicious to me as someone who has been doing this for many years because it doesn't seem like they should need that much with strong technical leadership and a cohesive strategy. Maybe to _serve_ the models but that should pay for itself, since AWS' whole thing is that they charge for compute. I'm also an advocate of separating LLMs from generative AI. It seems lots of people, especially on the business side, conflate the two.
@@dreadfulbodyguard7288 > LLMs are subset of generative AI. LLMs are not a subset of generative AI. LLMs and generative AI are two sets of things that loosely overlap. For example, other popular types of generative AI are diffusion models and GANs. Further, LLMs are more often used for text classification, sentiment analysis, and non-generative tasks. The generative use-case has long been understood, but only very recently has it gotten good enough to be more than a research task/toy. > In future, most generative AI will be multimodal with LLMs being important component. Probably. But you seem to be more sure of this than I am. That's the most-likely short-term future. I think the most impactful contribution long-term will be as planning models for robotic control (think: home robots, and even systems like HAL 9000 but hopefully with moral alignment). > No point in separating them. Disagree. It's nice to be able to say that I fine-tuned an LLM each time I specialize a very large text classifier (happens every so often), while being technically correct, and have the other person think I'm doing something exciting. That's why a bunch of people around me don't want to separate these. But I think this actually does us a disservice because when we _actually_ do the exciting (generative AI / LM) things, people will just say "oh, it's only another LLM".
Yes. Amazon is just talking their book. That said, as an investor, Amazon chips will be useful for us. Not everyone will be able to use Nvidia chips and Amazon can design their chips to work for exact use cases while nvidia's chips are more General. I only wish I had more Nvidia stock but oh well. I've got a lot of Amazon, and they're gonna do well from ai too.
Amazon is living out its principles 1/Frugality 2/Deliver results 3/Hire and Develop the best and 4/invent and Simplify! Amazon (AWS) knows how to do more with less and to leverage the best to win. AWS knows how invent where needed and simplify services to customers for easy use. #CustomerObsession all throughout this video!
The interviewer is pressing the CEO very hard to admit that Nvidia GPUs are the once to use if you are dealing with AI extensively; she is likely correct.
Amazon is already in AI /gen AI in a huge way its just market is yet to realize it's prowess . He is trying hard to explain to market but seems like market is not beleiving him.
Image/video reconstruction. Nvidia can take a 480p video and upscale it to 4k, with my minimal quality loss, which saves tons of bandwidth. 3d modeling for game development, you no longer have to animate every move, the ai learns it. Etc. So it's tons of use cases you'll never think of.... Hence why it's valuable.
Too much lip service, please make Alexa works as good as chatGPT to demonstrate AWS is on the top of GAI. Based on some calculations, AWS stock price shall be $122.00 , it is overpriced already. Today's price is $178.22
Love this interviewer! She's really getting into the heart of the matter.
She cut right into it and he had superior media training but still gave real answers. Great interview
Great challenging interview. I like how Deirdre Bosa basically gave one of the biggest web companies the treatment others would give an up-and-coming startup. You can tell Adam Selipsky has a technical background from his responses and at times trying desperately not to go into jargon. These challenging questions to a person who has the technical background to answer them gives us some great insight.
Great interview and quite insightful …
This is such a great and informative interview!
My Alexa gets half of the requests wrong....
Welcome to Amazon Advertising.
There aren’t just one player H100 from Nvidia. Even for Nvidia GPUs, the most popular ones are probably still V100s since their price to performance ratio is good! Popular LLM models such as BLOOM is trained on V100. For inference, there are already a lot of players from edge to server side inference. Even Apple, with neural engine on every new Macs and iPhone, is a strong contributor to this space. There are lots of opportunities in the server inference space for Amazon to grow.
Companies should ask the question; do they need a large monolithic chip to perform the duties they require? It wasn't that long ago that an AMD GPGPU could run laps around nVidia's part at significantly less cost. nVidia's change in architecture recently won back their compute crown as both AMD and nVidia consolidated the way their GPUs work. There is a use case for a large monolithic die as there is a drop in performance when multiple dies are used. But it comes at a cost of lower yields and higher price for the die. AMD is more than happy not venturing into the large dies as they get the yields they want.
Recently Intel announced they have a competitive iGPU. Not sure how that will play out on the larger scale.
Yes, if we are talking raw capabilities, H100 is the best, but he just couldn't say, yes it's good, our chip targets price/perf for specific use cases.
@@Tential1 who are you with?
Adam clearly couldn't lie and embellish that Amazon Chips are as good as NVIDIAs because he knows that's far fetched. Amazon GPUs are no where close to NVIDIAs. Good job Deidre for pointing it out and getting him off guard and asking the right tough questions.
H100 got new Transformer Engine that is supposedly 6-9x faster than A100 for transformer models like GPT and BERT for training. Eventhough H100 is more expensive, the perf/price could be really good if you’re working with transformers. That’s still very new and isn’t going to be mass deployed until next quarter.
“The company that uses the innovation rather than the innovators will benefit and profit” - Peter Lynch. Soo aws, google and Microsoft have the data that has quality. These 3 companies stand to gain
Great interview!
Is it just me or guy really looks like gavin belson in Silicon valley?
AMZ is a value AI buy.. Wall Street is plowing money into AI start up and AI start ups are plowing money into Cloud utilities. Whos leading cloud computing? Amazon... When there's a gold rush, don't invest in the individual gold prospector, invest in the pick axe company. Cheers.
Microsoft Azure and Google GCP as well.
Clearly you do not understand what Microsoft is doing in this space: Adam’s answer to this question is also a little bit awkward and less convincing.
he said nothing but a dream of Amazon ~~~~
That's how what I felt too. Amazon seemed to have lost it's way. Microsoft and Google are way advanced in this AI race. Also Microsoft Azure is gaining traction on model training and inference race with medium and AI startups since they are powering ChatGPT and Bing AI. They also are investing in their own chipset and procuring from NVDA. I see MSFT at $460 by end of the year.
100 milly investment for Amazon is like investing $20 bucks for the average person
Solid interview ngl
This anchor is good, calling him on his BS
The interviewer is missing a few points of knowledge on this topic. Trying too hard to play gotcha too.
She knows more than you
@@izdotcartershe really doesn't though. He made a valid point. Reporters may know some things, better sure. But computer nerds are gonna know their thing... And reporters don't know it. Reporters still say crypto is best used for money laundering. She was trying a got ya moment, and if she had understood the subject a little better she could have made her case. Not sure why you expect reporters to know tech well. It moves very fast, and lots of us are very very passionate down to insane detail, that let's you have a family/life, and us... Have hobbies lol.
1st We need a set of rules for AI asap. 2nd if this is not a AWS ad, i don't know what it is.
😅
It’s not an aws ad. The lady was giving him some hard questions
Amazon sent a salesman to this meeting? He did not have a single substantive technical point to make as to how Amazon will counter Nvidia/CUDA. Just offering a supermarket of mediocre capabilities is not enough when the best models will remain closed source.
you are watching CNBC..What did you expect
@@CHANTI8947 True! But Deirdre Bosa was quite persistent but not obnoxious.
Gavin Belzson strikes again
Right? The angle really makes the guy look like gavin belson
He's reading a teleprompter... Right?
Amazon finally followed in everyone else's footsteps and jumped on the "AI" train to Hotairbaloonville and their stock price went up. How nice ...
They didn't jump on this like everyone else. This was an aws service for years. 5 years old. Aws is the biggest profit driver of Amazon that drives the whole company. So yes.... This absolutely is real and isn't hot air. Amazon has a 25-30%ebitda cagr for a reason. That's why the stock is do valuable. It's still growing at an insane rate. None of this is new. Do you sleep through investor conferences? They explain these plans years in advance. Management's ability to execute on these plans is how they keep their job
"We're using supercomputers to do things better than humans and it's super super awesome". Stock goes up.
@@Tential1 Maybe you should google what "sarcasm" means before you go on to "educate" people on things of which just about everyone is aware.
My point is that lately all these companies have been in a race to tell everyone how cool their AI implementations are, which immediately pumps their stock price. NVidia and google have also been in AI for years too, yet it's only now the market seems to care due to the craze around chatGPT.
Until now, Amazon had stayed silent and seemed to keep their distance form from all the shameless self-promotion going around (see CRM). Now, they finally bought themselves a ticket for that train, as well.
@@milanmilanov348 LOL your original statement was NOT sarcasm so you're probably the one who needs to go Google sarcasm LOL.
@@krisp1878 Sounds to me you need to google what google means, then what lol means and only then we can discuss "sarcasm"
🪄✨ Made with SummarizeYT
0:22 - AWS launched Sagemaker, their machine learning platform, in 2017 and now have over 100,000 customers doing machine learning
0:50 - AWS scientists have been working on language based generative AI models such as LMS
1:13 - Alexa’s voice responses are driven by large language models
2:14 - AWS has designed its own chips called trainium for doing training and inferentia chip for doing inference
3:25 - Amazon is investing $100 million into a new generative AI innovation center
6:39 - Amazon has a managed service for doing machine learning called Bedrock
9:06 - Different models for different enterprises might need generative AI
11:05 - Amazon has been working on generative AI through different kinds of experts inside the company
13:11 - AWS is working on democratizing access to generative AI for every sector of the economy
14:46 - AWS CEO discusses the company's success as the leader of cloud services, having a big head start before its rivals caught on.
15:21 - AWS CEO believes that AI is going to be the next big thing and that they are focused on generative AI based applications.
23:03 - AWS CEO reveals the company's advantage over others in the GPU capacity market because they design their own chips and can provide additional value.
25:15 - AWS CEO confirms that the company is working with ARM, but there is still an option for optionality in their supply chain.
how do i use summariseYT
Instant regret by the interviewer for trying to correct a tech CEO 🤣LLM Generative AI
why is CNBC so obsessed with Nvidia? They act like no one else can create a chip.
You can create a chip in theory. The issue is, manufacturing it. Tsmc has deals with apple, amd, Intel, and Nvidia. They always get first dibs. So, you compete at a 30% deficit, as the latest manufacturing process adds a lot. Nvidia and apple go as far as buying excess capacity to ensure you can't use it. Then, the entry cost into designing is around 10B. Essentially, no one really can compete with Nvidia... Hence why they have led their industry for 20 years straight. It doesn't even matter if you do create a chip better than Nvidia. Their software stack is also what matters. Everyone uses Nvidia, so even if a competitor is better, your software won't work with it, you won't get support, etc. Trust me dude, I spent the 15 years trying to avoid Nvidia products. But when you actually have mission critical work, you go with Nvidia.
It's more stupid that you think competition will best Nvidia now, just because it fits your time narrative. Nvidia used to have tons of competitors. It has amd now. As the last man standing. There are a lot of bodies in Nvidias past. It's why I invested. I stopped fighting Nvidia, stopped trying to avoid the products, stopped begging for a competitor to dethrone them and their high prices. Instead, I bought the stock and the gains from the stock allow me to afford their very high prices.
The only people who seriously make their own chips are the majors. Amazon , Facebook, Google, Microsoft, apple. So if you want big tech to get bigger, yes, we can cheerlead them. Too cost intensive for anyone else
Why are you investing that much? It's suspicious to me as someone who has been doing this for many years because it doesn't seem like they should need that much with strong technical leadership and a cohesive strategy. Maybe to _serve_ the models but that should pay for itself, since AWS' whole thing is that they charge for compute.
I'm also an advocate of separating LLMs from generative AI. It seems lots of people, especially on the business side, conflate the two.
They have to pay employees Einstein.
@@yashagarwal8249 Do they really need a team of ~300 highly-skilled engineers to do this? It seems like 10 of the right people would work.
LLMs are subset of generative AI. In future, most generative AI will be multimodal with LLMs being important component. No point in separating them.
@@dreadfulbodyguard7288 > LLMs are subset of generative AI.
LLMs are not a subset of generative AI. LLMs and generative AI are two sets of things that loosely overlap. For example, other popular types of generative AI are diffusion models and GANs. Further, LLMs are more often used for text classification, sentiment analysis, and non-generative tasks. The generative use-case has long been understood, but only very recently has it gotten good enough to be more than a research task/toy.
> In future, most generative AI will be multimodal with LLMs being important component.
Probably. But you seem to be more sure of this than I am. That's the most-likely short-term future. I think the most impactful contribution long-term will be as planning models for robotic control (think: home robots, and even systems like HAL 9000 but hopefully with moral alignment).
> No point in separating them.
Disagree. It's nice to be able to say that I fine-tuned an LLM each time I specialize a very large text classifier (happens every so often), while being technically correct, and have the other person think I'm doing something exciting. That's why a bunch of people around me don't want to separate these. But I think this actually does us a disservice because when we _actually_ do the exciting (generative AI / LM) things, people will just say "oh, it's only another LLM".
We use a H100 :)
Deirdre is 10/1
Selipsky doesn't give me confidence on Amazon Web Services' future in these new circumstances.
Tesla Dojo: challenge accepted!! See yall in couple of years.
AI still need a human to generate new material.
Generative AI equals degenerative human employment.
AWS !!
The ceo is laughing in his head … why he has to go through these s questions lol😅
ok I admit it. I have no idea what AI is and how it is connected to Amazon. Does it help you find product on shelves? No clue here!
I prefer Jeff
greed is what got us at the end........lol
Thanks to Microsoft and Chatgpt Hollywood will no longer be the center of movie making.
Pls guys put a water food on roof top for birds🐔🕊🦅🐤🦜 and local surroundings for animals❤ pls do awareness share everyone like this🙏
Great advice
Come on let’s not kid ourselves nvidia chips > Amazon chips wtv they claim they have.
Yes. Amazon is just talking their book. That said, as an investor, Amazon chips will be useful for us. Not everyone will be able to use Nvidia chips and Amazon can design their chips to work for exact use cases while nvidia's chips are more General. I only wish I had more Nvidia stock but oh well. I've got a lot of Amazon, and they're gonna do well from ai too.
all talking points - literally a duplicate of the last video I watched on Bloomberg (AWS CEO Selipsky on the AI Boom)
Amazon is living out its principles 1/Frugality 2/Deliver results 3/Hire and Develop the best and 4/invent and Simplify!
Amazon (AWS) knows how to do more with less and to leverage the best to win. AWS knows how invent where needed and simplify services to customers for easy use. #CustomerObsession all throughout this video!
This ai boom
generating revenue for artificial intelligence by illegally forcing customers to join prime?
The interviewer is pressing the CEO very hard to admit that Nvidia GPUs are the once to use if you are dealing with AI extensively; she is likely correct.
Amazon research publications are comparatively nonexistent relative to piers (Google, OpenAI, and even Facebook)
He’s giving me SBF vibes
Amazon is already in AI /gen AI in a huge way its just market is yet to realize it's prowess . He is trying hard to explain to market but seems like market is not beleiving him.
AMD with AWS
Technology does not necessarily make our lives easier or better. There will always be problems, issues, and troubles no matter what.
Gavin belson in parallel universe
What the big hype over AI? The only real uses I see for it is for targeted AD’s, cybersecurity, analyzing data…
Image/video reconstruction. Nvidia can take a 480p video and upscale it to 4k, with my minimal quality loss, which saves tons of bandwidth. 3d modeling for game development, you no longer have to animate every move, the ai learns it. Etc. So it's tons of use cases you'll never think of.... Hence why it's valuable.
This interviewer has no real grasp on the topic
Makes her seem a bit unprepared for her task
Too much lip service, please make Alexa works as good as chatGPT to demonstrate AWS is on the top of GAI. Based on some calculations, AWS stock price shall be $122.00 , it is overpriced
already. Today's price is $178.22
finally
Thats just not very much money for generative AI.
100 M over 1.3 T market cap company is nothing.
All aboard the the AI bandwagon, next stop is Amazon to pick up AWS 😂
First viewer
Only advertisements for AWS…. Is not an independent news or interview…would say bought interview!
Waste
Nvidia is the leader
feeling very sorry for this person.. its jst another catch up.. its OAI and Microsoft period now!