1000 times compute in 8 years - that should give global warming a kick along.View attachment 59355 View attachment 59356
GTC March 2024 Keynote with NVIDIA CEO Jensen Huang
Watch NVIDIA CEO Jensen Huang’s GTC keynote to catch all the announcements on AI advances that are shaping our future.Dive into the announcements and discove...www.youtube.com
BRN Discussion Ongoing
- Thread starter TechGirl
- Start date
Evermont
Stealth Mode
I couldn't find the tempest emoji so fire will have to do.
They do mention some gains in efficiency, 25 x doesn't seem commensurate though.
"...They would also deliver 25 times the energy efficiency, NVIDIA said, a key claim when the creation of AI is criticised for its ravenous needs for energy and natural resources when compared to more conventional computing."
Just to make sure I'm not misunderstanding, you typically expect a huge event from a company that costs hundreds of thousands of dollars to showcase progress in their technology that they're still refining? But because we're a startup, you'll let it slide? (Besides, it's Edge Impulse with our Akida, but okay...)
And by the way in the next set of Quartals, you're probably going to be the first one to ask why we have -1.5 million in expenses.
🫵
Bloodsy
Regular
Anyone seen the new Dyson hairdryer that learns a users preference - on device?
https://www.dyson.com.au/products/hair-care/hair-dryers/dyson-supersonic-nural
"It will remember your go to settings, and select your preferred heat and airflow preferences"
I wonder......
The next generation if hairdryers!
https://www.dyson.com.au/products/hair-care/hair-dryers/dyson-supersonic-nural
"It will remember your go to settings, and select your preferred heat and airflow preferences"
I wonder......
The next generation if hairdryers!
Bravo
If ARM was an arm, BRN would be its biceps💪!
Edge Impulse Brings Nvidia’s Tao Toolkit To TinyML Hardware
By Sally Ward-Foxton 03.18.2024 0Share Post
Share on Facebook
Share on Twitter
Edge Impulse and Nvidia have collaborated to bring Nvidia’s Tao training toolkit to tiny hardware from other silicon vendors, including microcontrollers from NXP, STMicro, Alif and Renesas, with more hardware to follow. Embedded design teams can now easily train and optimize models on Nvidia GPUs in the cloud or on-premises using Tao, then deploy on embedded hardware using Edge Impulse.
“We realized AI is expanding into edge opportunities like IoT, where Nvidia doesn’t have silicon. So we said, why not?” Deepu Talla, VP and GM for robotics and edge computing at Nvidia, told EE Times. “We’re a platform, we have no problem enabling all of the ecosystem. [What we are able] to deploy on Nvidia Jetson, we can now deploy on a CPU, on an FPGA, on an accelerator—whatever custom accelerator you have—you can even deploy on a microcontroller.”
As part of the collaboration, Edge Impulse optimized almost 88 models from Nvidia’s model zoo for resource-constrained hardware at the edge. These models are available from Nvidia free of charge. The company has also added an extension to Nvidia Omniverse Replicator that allows users to create additional synthetic training data from existing datasets.
Tao toolkit
Tao is Nvidia’s toolkit for training and optimizing AI models for edge devices. In the latest release of Tao, model export in ONNX format is now supported, which makes it possible to deploy a Tao-trained model on any computing platform.Integration with Edge Impulse’s platform means Edge Impulse users get access to the latest research from Nvidia, including new types of models like vision transformers. Edge Impulse’s integrated development environment can handle data collection, training on your own dataset, evaluation and comparison of models for different devices, and deployment of Tao models to any hardware. Training is run on Nvidia GPUs in Edge Impulse’s cloud via API.
Nvidia Tao toolkit now has ONNX support so that models can be deployed on any hardware. (Source: Nvidia)Why would Nvidia make tools and models it has invested heavily in available to other types of hardware?
“Nvidia doesn’t participate in all of the AI inference market,” Talla said, noting that Nvidia’s edge AI offerings, including Jetson, are built for autonomous machines and industrial robotics where heavy duty inference is required.
Beyond that, in smartphones and IoT devices: “We will not participate in that market,” he said. “Our strategy is to play in autonomous machines, where there’s multiple sensors and sensor fusion, and that’s a strategic choice we made. The tens or hundreds of companies developing products from mobile to IoT, you could say they are competitors, but it’s overall speeding up the adoption of AI, which is good.”
Making Tao available for smaller AI chips than Jetson isn’t an altruistic move, Talla said.
“A rising tide lifts all boats,” he said. “There’s a gain for Nvidia because…IoT devices go will go into billions if not tens of billions of units annually. Jetson is not targeting that market. As AI adoption grows at the edge, we want to monetize it on the data center side. If somebody’s going to use our GPUs in the cloud to train their AI, we have monetized that.”
Users will save money, he said, because Tao will make it easier to train on GPUs in the cloud, increasing time to market for products.
“It’s beneficial to everyone in the ecosystem,” he said. “I think this is a win-win for all of our partners in the middle, and end customers.”
Nvidia went through a lot of the same challenges facing embedded developers today in creating and optimizing models for Jetson hardware seven to either years ago. For example, Talla said, gathering data is very difficult as you can’t cover all the corner cases, there are many open-source models to choose from and they change frequently, and AI frameworks are also continuously changing.
“Even if you master all of that, how can you create a performance model that is going to be the right size, meaning the memory footprint, especially when it comes to running at the edge?”
Tao was developed for this purpose five to eight years ago and most of it was open sourced last year.
“We want to give full control for anybody to take as many pieces as they want, to control their destiny, that’s why it’s not a closed piece of software,” Talla said.
The technical collaboration between Nvidia and Edge Impulse had several facets, Talla said. First, the teams needed to make sure models being trained in Tao were in the right format for silicon vendors’ runtime tools (edge hardware platforms typically have their own runtime compilers to optimize further). Second, Nvidia regularly updates its model zoo with state of the art models, but backporting those models to older frameworks is extremely challenging—the challenge, he said, is figuring out “whether we can keep the old models with the old frameworks despite adding newer models, something we’re still trying to figure out together.”
Model zoo
As part of the collaboration, Edge Impulse has optimized almost 88 models for the edge from Nvidia’s model zoo, Daniel Situnayake, director of ML at Edge Impulse, told EE Times.“We’ve selected specific computer vision models from Nvidia’s Tao library that are appropriate for embedded constraints based on their trade-offs between latency, memory use and task performance,” he said.
Models like RetinaNet, YOLOv3, YOLOv4 and SSD were ideal options with slightly different strengths, he said. Because these models previously required Nvidia hardware to run, a certain amount of adaptation was required.
“To make them universal, we’ve performed model surgery to create custom versions of the models that will run on any C++ target, and we’ve created target-optimized implementations of any custom operations that are required,” Situnayake said. “For example, we’ve written fast versions of the decoding and non-maximum suppression algorithms used to create bounding boxes for object detection models.”
Further optimizations include quantization, scaling models down to run on mid-range microcontrollers like those based on Arm Cortex-M4 cores, and pre-training them to support input resolutions that are appropriate for embedded vision sensors.
“This results in seriously tiny models, for example, a YOLOv3 object detection model that uses 500 kB RAM and 1.2 MB ROM,” he said.
Models can be deployed via Edge Impulse’s EON compiler or using the silicon vendor’s toolchain. Edge Impulse’s EON Tuner hyperparameter optimization system can help users choose the optimum combination of model and hyperparameters for the user’s data set and target device.
Nvidia Omniverse Replicator integration with Edge Impulse allows users to generate synthetic data to address any gaps in their datasets. (Source: Nvidia)Edge Impulse has also been working with Nvidia on integration with Omniverse Replicator, Nvidia’s tool for synthetic data generation. Edge Impulse users can now use Omniverse Replicator to generate synthetic image data based on their existing data for training—perhaps to address certain gaps in the dataset to ensure accurate and versatile trained models.
Edge Impulse’s integration with Nvidia Tao is currently available for hardware targets including NXP, STMicro, Alif and Renesas, with Nordic devices next in line for onboarding, the company said.
Last edited:
CrabmansFriend
Regular
Edge Impulse Brings Nvidia’s Tao Toolkit To TinyML Hardware
By Sally Ward-Foxton 03.18.2024 0
Share Post
Share on Facebook
Share on Twitter
Edge Impulse and Nvidia have collaborated to bring Nvidia’s Tao training toolkit to tiny hardware from other silicon vendors, including microcontrollers from NXP, STMicro, Alif and Renesas, with more hardware to follow. Embedded design teams can now easily train and optimize models on Nvidia GPUs in the cloud or on-premises using Tao, then deploy on embedded hardware using Edge Impulse.
“We realized AI is expanding into edge opportunities like IoT, where Nvidia doesn’t have silicon. So we said, why not?” Deepu Talla, VP and GM for robotics and edge computing at Nvidia, told EE Times. “We’re a platform, we have no problem enabling all of the ecosystem. [What we are able] to deploy on Nvidia Jetson, we can now deploy on a CPU, on an FPGA, on an accelerator—whatever custom accelerator you have—you can even deploy on a microcontroller.”
As part of the collaboration, Edge Impulse optimized almost 88 models from Nvidia’s model zoo for resource-constrained hardware at the edge. These models are available from Nvidia free of charge. The company has also added an extension to Nvidia Omniverse Replicator that allows users to create additional synthetic training data from existing datasets.
Tao toolkit
Tao is Nvidia’s toolkit for training and optimizing AI models for edge devices. In the latest release of Tao, model export in ONNX format is now supported, which makes it possible to deploy a Tao-trained model on any computing platform.
Integration with Edge Impulse’s platform means Edge Impulse users get access to the latest research from Nvidia, including new types of models like vision transformers. Edge Impulse’s integrated development environment can handle data collection, training on your own dataset, evaluation and comparison of models for different devices, and deployment of Tao models to any hardware. Training is run on Nvidia GPUs in Edge Impulse’s cloud via API.
Nvidia Tao toolkit now has ONNX support so that models can be deployed on any hardware. (Source: Nvidia)
Why would Nvidia make tools and models it has invested heavily in available to other types of hardware?
“Nvidia doesn’t participate in all of the AI inference market,” Talla said, noting that Nvidia’s edge AI offerings, including Jetson, are built for autonomous machines and industrial robotics where heavy duty inference is required.
Beyond that, in smartphones and IoT devices: “We will not participate in that market,” he said. “Our strategy is to play in autonomous machines, where there’s multiple sensors and sensor fusion, and that’s a strategic choice we made. The tens or hundreds of companies developing products from mobile to IoT, you could say they are competitors, but it’s overall speeding up the adoption of AI, which is good.”
Making Tao available for smaller AI chips than Jetson isn’t an altruistic move, Talla said.
“A rising tide lifts all boats,” he said. “There’s a gain for Nvidia because…IoT devices go will go into billions if not tens of billions of units annually. Jetson is not targeting that market. As AI adoption grows at the edge, we want to monetize it on the data center side. If somebody’s going to use our GPUs in the cloud to train their AI, we have monetized that.”
Users will save money, he said, because Tao will make it easier to train on GPUs in the cloud, increasing time to market for products.
“It’s beneficial to everyone in the ecosystem,” he said. “I think this is a win-win for all of our partners in the middle, and end customers.”
Nvidia went through a lot of the same challenges facing embedded developers today in creating and optimizing models for Jetson hardware seven to either years ago. For example, Talla said, gathering data is very difficult as you can’t cover all the corner cases, there are many open-source models to choose from and they change frequently, and AI frameworks are also continuously changing.
“Even if you master all of that, how can you create a performance model that is going to be the right size, meaning the memory footprint, especially when it comes to running at the edge?”
Tao was developed for this purpose five to eight years ago and most of it was open sourced last year.
“We want to give full control for anybody to take as many pieces as they want, to control their destiny, that’s why it’s not a closed piece of software,” Talla said.
The technical collaboration between Nvidia and Edge Impulse had several facets, Talla said. First, the teams needed to make sure models being trained in Tao were in the right format for silicon vendors’ runtime tools (edge hardware platforms typically have their own runtime compilers to optimize further). Second, Nvidia regularly updates its model zoo with state of the art models, but backporting those models to older frameworks is extremely challenging—the challenge, he said, is figuring out “whether we can keep the old models with the old frameworks despite adding newer models, something we’re still trying to figure out together.”
Model zoo
As part of the collaboration, Edge Impulse has optimized almost 88 models for the edge from Nvidia’s model zoo, Daniel Situnayake, director of ML at Edge Impulse, told EE Times.
“We’ve selected specific computer vision models from Nvidia’s Tao library that are appropriate for embedded constraints based on their trade-offs between latency, memory use and task performance,” he said.
Models like RetinaNet, YOLOv3, YOLOv4 and SSD were ideal options with slightly different strengths, he said. Because these models previously required Nvidia hardware to run, a certain amount of adaptation was required.
“To make them universal, we’ve performed model surgery to create custom versions of the models that will run on any C++ target, and we’ve created target-optimized implementations of any custom operations that are required,” Situnayake said. “For example, we’ve written fast versions of the decoding and non-maximum suppression algorithms used to create bounding boxes for object detection models.”
Further optimizations include quantization, scaling models down to run on mid-range microcontrollers like those based on Arm Cortex-M4 cores, and pre-training them to support input resolutions that are appropriate for embedded vision sensors.
“This results in seriously tiny models, for example, a YOLOv3 object detection model that uses 500 kB RAM and 1.2 MB ROM,” he said.
Models can be deployed via Edge Impulse’s EON compiler or using the silicon vendor’s toolchain. Edge Impulse’s EON Tuner hyperparameter optimization system can help users choose the optimum combination of model and hyperparameters for the user’s data set and target device.
Nvidia Omniverse Replicator integration with Edge Impulse allows users to generate synthetic data to address any gaps in their datasets. (Source: Nvidia)
Edge Impulse has also been working with Nvidia on integration with Omniverse Replicator, Nvidia’s tool for synthetic data generation. Edge Impulse users can now use Omniverse Replicator to generate synthetic image data based on their existing data for training—perhaps to address certain gaps in the dataset to ensure accurate and versatile trained models.
Edge Impulse’s integration with Nvidia Tao is currently available for hardware targets including NXP, STMicro, Alif and Renesas, with Nordic devices next in line for onboarding, the company said.
My favorite quotes:
Bravo
If ARM was an arm, BRN would be its biceps💪!
NVIDIA have announced DRIVE Thor SoC will be based on Blackwell GPU architecture.
And here's something I think is VERY INTERESTING! Thor will be paired with a "YET TO BE NAMED GPU"!!!
EXTRACT
www.hardwarezone.com.sg
And here's something I think is VERY INTERESTING! Thor will be paired with a "YET TO BE NAMED GPU"!!!
EXTRACT
Blackwell GPU power onboard NVIDIA Drive Thor will unleash Gen AI experiences in your upcoming car
Your next car might be more chatty, and context-aware thanks to generative AI and LLM processing capability brought on by NVIDIA Drive Thor and its Blackwell GPU power. #nvidia #drivethor #genai #blackwellgpu #autonomousvehicle
Last edited:
CrabmansFriend
Regular
I am just trying to get up to date about the Nvidia GTC news.
Nvidia's Blackwell looks really impressive but power consumption for these ML systems is becoming mind boggling if you add up all instances/data centers etc.
A german tech journal had a nice quote (translated via deepl.com)
Nvidia's Blackwell looks really impressive but power consumption for these ML systems is becoming mind boggling if you add up all instances/data centers etc.
A german tech journal had a nice quote (translated via deepl.com)
NVIDIA have announced DRIVE Thor SoC will be using Blackwell GPU architecture.
View attachment 59364
And here's something I think is VERY INTERESTING! Thor will be paired with a "YET TO BE NAMED GPU"!!!
EXTRACT
View attachment 59365
Blackwell GPU power onboard NVIDIA Drive Thor will unleash Gen AI experiences in your upcoming car
Your next car might be more chatty, and context-aware thanks to generative AI and LLM processing capability brought on by NVIDIA Drive Thor and its Blackwell GPU power. #nvidia #drivethor #genai #blackwellgpu #autonomousvehicle
Bravo
If ARM was an arm, BRN would be its biceps💪!
It looks like NVIDIA are focussing on power consumption with Blackwell.
Stable Genius
Regular
Anyone seen the new Dyson hairdryer that learns a users preference - on device?
https://www.dyson.com.au/products/hair-care/hair-dryers/dyson-supersonic-nural
"It will remember your go to settings, and select your preferred heat and airflow preferences"
I wonder......
The next generation if hairdryers!
Nice find @Bloodsy
May not be us included but shows the direction the market is headed; and where we could help/fit in. I thought a few years back we were tentatively linked to Dyson but can’t recall how now.
I interestingly the hairdryer is called Nural which is trademarked so I’m guessing Nural might appear in a host of their products.
Admittedly not sure how we get into a product without announcing a partnership at least. Unless the nature of our relationship is commercially sensitive (which it would be) and Dyson would lose a commercial advantage to competitors (I think that’s how it goes). But when revenue comes in I think we then have to make a statement to that effect on the ASX (whilst still protecting some specifics). A tangled web. Time will tell!
Bravo
If ARM was an arm, BRN would be its biceps💪!
Extract
GTC Wrap-Up: ‘We Created a Processor for the Generative AI Era,’ NVIDIA CEO Says
Kicking off the biggest GTC conference yet, NVIDIA founder and CEO Jensen Huang unveils NVIDIA Blackwell, NIM microservices, Omniverse Cloud APIs and more.
blogs.nvidia.com
CrabmansFriend
Regular
Yes it's impressive, but in the end it probably always leads to a "Jevons Paradox", but that's not Nvidia's fault (alone).
Bravo
If ARM was an arm, BRN would be its biceps💪!
Peter Van Der Made once said "4 bits are enough".
hothardware.com
NVIDIA Unveils Powerful Blackwell GPU Architecture For Next-Gen AI Workloads At GTC
NVIDIA CEO Jensen Huang opened GTC 2024 by announcing Blackwell, the most powerful AI chip ever made.
Pom down under
Top 20
Smoothsailing
Regular
Nice
Smoothsailing
Regular
Incredible insight
Fullmoonfever
Top 20
Here's an article from late last year where this author does various reviews on companies, their strategies, Dev etc.
Had to split into 3 posts as too many characters.
He seems to have the same line of thinking as the rest of us as to what they need....guess it's just how they might go about it and implement
Covered in part 3 if want to skip analysis.
www.howtheygrow.co
JARYD HERMANN
SEP 07, 2023
Welcome to How They Grow, my newsletter’s main series. Bringing you in-depth analyses on the growth of world-class companies, including their early strategies, current tactics, and actionable product-building lessons we can learn from them.
Hi, friends
Get comfy, and hold onto your seats.
I won’t mince words here. Nvidia—the most valuable company by market cap that we’ve tackled so far—is not just a behemoth, but one of the most important companies on the planet. Analyzing a company this size means I’m either going to get caught with my pants down trying to tell this story well enough, or, if pulled off, I imagine this will be one of the more successful deep dives I’ve done.
As always, you’ll be the judge of that.
We all know Nvidia has taken the spotlight in the news recently, and for good reason. On the surface, it’s because their stock has been on an absolute tear.
Just take May 25th, when they added $184 billion to their market cap in a single day. That’s more than the value of SHEIN, Stripe, and Canva combined.
And it’s not just a price run with nothing behind it. In their latest second-quarter earning report, their revenue was $13.51 billion, up 101% YoY and up 88% from Q1. They also have $16B in cash/liquid assets. That’s wild.
Of course, this growth has been because Nvidia is playing a huge role in this AI revolution. But I don’t need to tell you that with a 30-year-old company valued at $1.2 trillion—there’s more to the story.
So, my curiosity sent me on a 30-hour researching and analyzing spree, going over everything I could get my hands on to bring you my armchair analysis of the world's 6th most valuable company.
I’ll summarize this 7K word deep dive for you in three words: a strategic masterclass.
Jensen Huang—the Taiwanese-born, leather-jacket-rocking, and pimped-up-Toyota-Supra-driving founder and CEO of Nvidia—is a man who’s bet the entire future of Nvidia not once, but three times. And, he may well do it again.
Jensen is a seismic wave-spotting genius, and arguably one of the best strategic executors I’ve come across. He has a resume of making deep and accurate analyses of Nvidia’s competitive and macro environment, developing very clear and well-defined strategic visions, and addressing all potential strategic bottlenecks with policies and coherent actions that guide Nvidia forward.
But, there’s a good chance you’ve never heard his name until now because he doesn’t poke at other CEOs about cage-fighting, and he doesn’t get people all riled up. He thinks, strategizes, and he humbly gets to work building for the future.
Jensen Huang
I learned so much from studying him and Nvidia. I hope you will too.
Now, just to tease what we’ll be getting much deeper into…
Nvidia’s invention of the Graphical Processing Unit in 1999 pioneered one of the most significant developments in modern technology: GPU-accelerated computing. With it, they also sparked the growth of the PC gaming market, redefined modern computer graphics, and revolutionized parallel computing; allowing computers to do way more.
In the past few years, Nvidia saw their first big growth surge with the crypto wave, where their computational hardware was essential for mining. But, after the hype cycle, revenue fell as crypto crashed and the gaming industry slummed.
Then along came ChatGPT, producing a tidal wave of interest in the AI sector, especially Generative AI. This ignited a race at various layers of the AI stack. While Google, Microsoft, Meta, OpenAI, Anthropic, and other AI players battle it out, Nvidia rakes in the cash because their chips power pretty much all of them. And they’re benefiting from this increased demand more than anybody else because Nvidia’s specialized chips can process large amounts of data more efficiently and cost-effectively than traditional semiconductors. Essentially, they are the ground for AI research.
You might be thinking that Nvidia won the lotto here by having GPU chips so perfectly suited for being the engine behind AI. Except, thanks to Jensen, being in this position was a decade in the making. The company’s chips broke through the AI community back in 2012 with AlexNet Neural Network, and it’s clear Nvidia made a massive bet on AI long before other companies did. As Jensen said in a keynote at the National Taiwan University in May, “We risked everything to pursue deep learnings.”
So, while Nvidia gets to ride the AI wave, it’s their GPU-powered deep learning that actually ignited modern AI in the first place, with their GPUs acting as the brains of computers, robots, and self-driving cars.
Simply, through a pick-and-shovel play (more on this soon), Nvidia is tightly coupled with most high-growth sectors, like AI, machine learning, VR/AR, big data, cybersecurity, gaming, crypto, and autonomous travel, as they all rely on Nvidia’s chips. Thus, when one of these high-growth sectors is running hot, so too will Nvidia.
In short, Nvidia’s products are at the center of the most consequential mega-trends in technology. And I have no doubt that they will continue to be at the forefront of the next revolution. At the end of this piece, we’ll touch on what that might be for Nvidia: Hint: Neuromorphic Computing.
On that note, let's get to it.
There are a lot of places I could start telling it, but I think the most logical place is in a dingy diner covered in bullet holes, in the then-sketchy neighborhood of East San Jose, in April 1993. Three entrepreneurial electrical engineers—Jensen Huang, Chris Malachowsky, and Curtis Priem—sat drinking bottomless burnt coffee, deep in conversation, anticipating that the next wave of computing would be graphics-based.
Nobody would have thought their conversation about chips was laying the foundation for a company that was about to define computing—and still does.
As Chris recalled:
So, buzzed on coffee and their strong conviction that the next big opportunity in technology was in accelerated computing— a parallel processing approach that frees the computer’s Central Processing Unit (CPU) and gives the heavy lifting of data processing to other processors, like the GPU—they left Denny’s Diner to go and found Nvidia.
Things were in motion. But it took a few years for them to find their feet.
The first product they built was a multimedia card called NV1, which was introduced in the market in 1995.
The founders leaned on the expected multimedia revolution at the time and tried to ride the wave and establish an industry standard. However, the NV1 was a bit of a flop. It wasn’t considered better than the competitors (Intel, and AMD), and it was hard to program. As a result, the games that were created for the NV1 weren’t great and had performance issues.
They got to work on a second version, the NV2, but ended up canning it before it went live.
Why?
Because they were not feeling a pull from the market, and while NV2 was in development, Jensen was running a strategic analysis that took into account the team’s, and company’s, existing strategic advantages, as well as crucial market observations from media industry experts like Ed Catmull, co-founder of Pixar.
He realized their current path was not the way, and he made the first big bet on Nvidia’s future.
So, four years in and with an actual product in the market paying the bills, he made a major strategic pivot and focused on the development of a product that he believed would be better suited for a new, exponentially growing, market.
A good lesson on the exercise of niching down to find your sweet spot in the market.
They decided this was the market they needed to focus on, the real wave, and set out to design a Graphic Processing Unit (GPU) that would render 3D graphics. They dropped their initial approach to graphics they had created with the NV1 and adopted the then quickly-rising (and today’s standard) in 3D graphics processing—the graphics pipeline. This approach was in fact developed by a competitor, SGI.
So, why did the team decide to compete in this fast-paced and highly competitive arena of computational power? Why did they think this move would be successful? Two reasons:
This meant there was a massive TAM and a big enough slice for Nvidia to take.
The reason was the emerging video game industry, where (1) games had the most computationally challenging problems, (2) computational processing power was a competitive advantage for online video gamers, and (3) they believed video there would be incredibly high sales volume.
So, the better your GPU as a gamer, the better your performance, and the more bragging rights you’d have about your gaming skills.
This presented gaming as a strong wedge into the 3D graphics computational arena—which as we’ll see—became the company's core flywheel to reach other adjacent markets and fund huge R&D to solve massive computational problems.
When Nvidia broke into this market, the main dog was Big Blue: Intel.
Intel's incumbent strategy was to minimize the CPU’s costs. AKA, to make the single horse pulling the chariot more efficient and more profitable to sell.
Nvidia’s strategy was to maximize the GPU’s performance. AKA, to add another type of workhorse to share the load. It increased the cost of the machine, but their go-to-market was to cater to the constantly growing need for faster graphics rendering.
To give a bit more context as to why Jensen’s bet was such an innovative and bold one, it’s helpful to understand the status quo at the time in the industry. In short, progress in computational power depended on decreasing the size of transistors and reducing consumed power. So, the first chip player who found a way to design a significantly smaller transistor and a way to manufacture it would gain a huge advantage in the market.
Sounds good, except that the rate of progress seemed to follow an observation by Gordon Moore, Intel’s co-founder, known as Moore’s law. It basically stated that significant progress in chips would occur every 2 years.
https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https://substack-post-media.s3.amazonaws.com/public/images/d0f2bdbe-c71f-4110-a2a1-3ab2e3fd43d0_1500x1000.png
The whole industry relied on that law, accepting that nobody could really break it. At least not consistently. Which is kind of funny when you think about it: the leader of the incumbent company coining a law that in a way fends off competition.
But, Nvidia’s radically different positioning in the computational arena was based on Jensen’s belief that Nvidia could actually provide an equivalent improvement of computational power in one-third of the industry’s cycle by focusing on increasing the number of transistors in their GPUs (not CPUs), taking full advantage of the graphics pipeline approach.
This approach was revolutionary, and their GPU strategy was so successful at setting up Nvidia’s advantage for 2 big reasons:
The first big bet that paid off.
But, unfortunately, the physical limit in the chip world does exist, as you can only make a transistor so small.
So, despite forking in the cash and watching their public company’s stock price climb, Jensen saw the limit approaching. He saw Nvidia losing their coveted advantage. And being a badass who never rests on his laurels, he went back to the strategy board and maneuvered the ship once again.
And this stage was foundational, not just for the company, but for human progress. Simply, because GPUs set the stage to reshape the computing industry.
But in 2006, they opened Pandora’s box to how GPUs could be used by launching CUDA—a general-purpose programming model that allowed developers to use their Nvidia GPUs for other things.
This was Jensen and Nvidia’s first platform strategy.
To drastically oversimplify it, this made the power of accelerated computing fully accessible, allowing smart people to unlock and harness their GPU’s processing ability beyond games. This breakthrough allowed for multiple computations at a given time instead of sequentially, resulting in much faster processing speeds—a game-changing move in the development of applications for various industries, like aerospace, bio-science research, mechanical and fluid simulations, climate modeling, finance, professional visualization, energy exploration, autonomous driving, and AI, which all require big number crunching to work.
In other words, CUDA was a monumental development in the industry that took Nvidia from a hardware product for gaming to an industry-opening platform.
That flywheel was, and still is, fueled further by two other elements:
AKA, Nvidia is poised to provide for multiple gold rushes.
A testament to the power of solving hard problems in the background.
What you can do with this
When demand is up, build your position in the arena from a different point of view
Study the market to spot emerging opportunities, don’t rest on your laurels, and study your competitors to understand their most important strategic decisions.
Then build a coherent strategy based on the following:
Had to split into 3 posts as too many characters.
He seems to have the same line of thinking as the rest of us as to what they need....guess it's just how they might go about it and implement
Covered in part 3 if want to skip analysis.
How Nvidia Grows: The Engine for AI and The Catalyst Of The Future
A pure strategic masterclass from Nvidia's trillion-dollar accelerated computing empire
www.howtheygrow.co
How Nvidia Grows: The Engine for AI and The Catalyst Of The Future
A pure strategic masterclass from Nvidia's trillion-dollar accelerated computing empire
JARYD HERMANN
SEP 07, 2023
Welcome to How They Grow, my newsletter’s main series. Bringing you in-depth analyses on the growth of world-class companies, including their early strategies, current tactics, and actionable product-building lessons we can learn from them.Hi, friends
Get comfy, and hold onto your seats.
I won’t mince words here. Nvidia—the most valuable company by market cap that we’ve tackled so far—is not just a behemoth, but one of the most important companies on the planet. Analyzing a company this size means I’m either going to get caught with my pants down trying to tell this story well enough, or, if pulled off, I imagine this will be one of the more successful deep dives I’ve done.
As always, you’ll be the judge of that.
We all know Nvidia has taken the spotlight in the news recently, and for good reason. On the surface, it’s because their stock has been on an absolute tear.
Just take May 25th, when they added $184 billion to their market cap in a single day. That’s more than the value of SHEIN, Stripe, and Canva combined.
And it’s not just a price run with nothing behind it. In their latest second-quarter earning report, their revenue was $13.51 billion, up 101% YoY and up 88% from Q1. They also have $16B in cash/liquid assets. That’s wild.
Of course, this growth has been because Nvidia is playing a huge role in this AI revolution. But I don’t need to tell you that with a 30-year-old company valued at $1.2 trillion—there’s more to the story.
So, my curiosity sent me on a 30-hour researching and analyzing spree, going over everything I could get my hands on to bring you my armchair analysis of the world's 6th most valuable company.
I’ll summarize this 7K word deep dive for you in three words: a strategic masterclass.
Jensen Huang—the Taiwanese-born, leather-jacket-rocking, and pimped-up-Toyota-Supra-driving founder and CEO of Nvidia—is a man who’s bet the entire future of Nvidia not once, but three times. And, he may well do it again.
Jensen is a seismic wave-spotting genius, and arguably one of the best strategic executors I’ve come across. He has a resume of making deep and accurate analyses of Nvidia’s competitive and macro environment, developing very clear and well-defined strategic visions, and addressing all potential strategic bottlenecks with policies and coherent actions that guide Nvidia forward.
But, there’s a good chance you’ve never heard his name until now because he doesn’t poke at other CEOs about cage-fighting, and he doesn’t get people all riled up. He thinks, strategizes, and he humbly gets to work building for the future.
Jensen Huang
I learned so much from studying him and Nvidia. I hope you will too.
Now, just to tease what we’ll be getting much deeper into…
Nvidia’s invention of the Graphical Processing Unit in 1999 pioneered one of the most significant developments in modern technology: GPU-accelerated computing. With it, they also sparked the growth of the PC gaming market, redefined modern computer graphics, and revolutionized parallel computing; allowing computers to do way more.
In the past few years, Nvidia saw their first big growth surge with the crypto wave, where their computational hardware was essential for mining. But, after the hype cycle, revenue fell as crypto crashed and the gaming industry slummed.
Then along came ChatGPT, producing a tidal wave of interest in the AI sector, especially Generative AI. This ignited a race at various layers of the AI stack. While Google, Microsoft, Meta, OpenAI, Anthropic, and other AI players battle it out, Nvidia rakes in the cash because their chips power pretty much all of them. And they’re benefiting from this increased demand more than anybody else because Nvidia’s specialized chips can process large amounts of data more efficiently and cost-effectively than traditional semiconductors. Essentially, they are the ground for AI research.
You might be thinking that Nvidia won the lotto here by having GPU chips so perfectly suited for being the engine behind AI. Except, thanks to Jensen, being in this position was a decade in the making. The company’s chips broke through the AI community back in 2012 with AlexNet Neural Network, and it’s clear Nvidia made a massive bet on AI long before other companies did. As Jensen said in a keynote at the National Taiwan University in May, “We risked everything to pursue deep learnings.”
So, while Nvidia gets to ride the AI wave, it’s their GPU-powered deep learning that actually ignited modern AI in the first place, with their GPUs acting as the brains of computers, robots, and self-driving cars.
Simply, through a pick-and-shovel play (more on this soon), Nvidia is tightly coupled with most high-growth sectors, like AI, machine learning, VR/AR, big data, cybersecurity, gaming, crypto, and autonomous travel, as they all rely on Nvidia’s chips. Thus, when one of these high-growth sectors is running hot, so too will Nvidia.
In short, Nvidia’s products are at the center of the most consequential mega-trends in technology. And I have no doubt that they will continue to be at the forefront of the next revolution. At the end of this piece, we’ll touch on what that might be for Nvidia: Hint: Neuromorphic Computing.
On that note, let's get to it.
Here’s what you can expect in today’s analysis:
- How Nvidia Started: Denny’s Diner, Stumbled Steps, GPUs, and CUDA
- Right direction, wrong product
- Spotting an opportunity in gaming, crafting a winning GPU strategy, and building their base layer
- Expanding beyond games, and unlocking industries with the CUDA platform
- How They Grow: Powering The Next Stage Of The Internet
- The Nvidia Data Center: One platform, unlimited acceleration
- Chips & Systems: AIaaS— a classic Pick-and-Shovel play
- Nvidia Omniverse: The platform for the useful metaverse
- Seeding their own multi-market ecosystem: A lesson on expanding your TAM
- What could be next for Nvidia? Neuromorphic Computing?
How Nvidia Started: Denny’s Diner, Stumbled Steps, GPUs, and CUDA
The story of Nvidia is one that shows how engineering breakthroughs really do push our world forward.There are a lot of places I could start telling it, but I think the most logical place is in a dingy diner covered in bullet holes, in the then-sketchy neighborhood of East San Jose, in April 1993. Three entrepreneurial electrical engineers—Jensen Huang, Chris Malachowsky, and Curtis Priem—sat drinking bottomless burnt coffee, deep in conversation, anticipating that the next wave of computing would be graphics-based.
Nobody would have thought their conversation about chips was laying the foundation for a company that was about to define computing—and still does.
As Chris recalled:
The wave they saw coming was the nascent market for accessible GPUs, and how there was an opportunity to create plug-in chips for PCs that could provide realistic 3D graphics for games and films.
So, buzzed on coffee and their strong conviction that the next big opportunity in technology was in accelerated computing— a parallel processing approach that frees the computer’s Central Processing Unit (CPU) and gives the heavy lifting of data processing to other processors, like the GPU—they left Denny’s Diner to go and found Nvidia.
Things were in motion. But it took a few years for them to find their feet.
Right direction, wrong product
Even though Jensen implemented a highly successful strategy from 1998 to 2008—setting their trajectory to own the global chip market— Nvidia stumbled in their first steps.The first product they built was a multimedia card called NV1, which was introduced in the market in 1995.
The founders leaned on the expected multimedia revolution at the time and tried to ride the wave and establish an industry standard. However, the NV1 was a bit of a flop. It wasn’t considered better than the competitors (Intel, and AMD), and it was hard to program. As a result, the games that were created for the NV1 weren’t great and had performance issues.
They got to work on a second version, the NV2, but ended up canning it before it went live.
Why?
Because they were not feeling a pull from the market, and while NV2 was in development, Jensen was running a strategic analysis that took into account the team’s, and company’s, existing strategic advantages, as well as crucial market observations from media industry experts like Ed Catmull, co-founder of Pixar.
He realized their current path was not the way, and he made the first big bet on Nvidia’s future.
So, four years in and with an actual product in the market paying the bills, he made a major strategic pivot and focused on the development of a product that he believed would be better suited for a new, exponentially growing, market.
Spotting an opportunity in gaming, crafting a winning GPU strategy, and building their base layer
Jensen and the team noticed an attractive opportunity in the field of 3D graphics, a sub-niche of their initial multimedia focus.A good lesson on the exercise of niching down to find your sweet spot in the market.
They decided this was the market they needed to focus on, the real wave, and set out to design a Graphic Processing Unit (GPU) that would render 3D graphics. They dropped their initial approach to graphics they had created with the NV1 and adopted the then quickly-rising (and today’s standard) in 3D graphics processing—the graphics pipeline. This approach was in fact developed by a competitor, SGI.
So, why did the team decide to compete in this fast-paced and highly competitive arena of computational power? Why did they think this move would be successful? Two reasons:
1. The market’s demand for GPUs was expected to be “unlimited”
Unlimited in the sense that it would surpass supply by a significant margin and wouldn’t slow down anytime soon.This meant there was a massive TAM and a big enough slice for Nvidia to take.
The reason was the emerging video game industry, where (1) games had the most computationally challenging problems, (2) computational processing power was a competitive advantage for online video gamers, and (3) they believed video there would be incredibly high sales volume.
So, the better your GPU as a gamer, the better your performance, and the more bragging rights you’d have about your gaming skills.
This presented gaming as a strong wedge into the 3D graphics computational arena—which as we’ll see—became the company's core flywheel to reach other adjacent markets and fund huge R&D to solve massive computational problems.
2. They took a radically different position in the computational market
Remember, accelerated computing is all about improving a computer's processing power. The better it can process, the more advanced feats it can pull off. Back in the '90s and early 2000s, the main place processing was taking place was the PC's Central Processing Unit (CPU).When Nvidia broke into this market, the main dog was Big Blue: Intel.
Intel's incumbent strategy was to minimize the CPU’s costs. AKA, to make the single horse pulling the chariot more efficient and more profitable to sell.
Nvidia’s strategy was to maximize the GPU’s performance. AKA, to add another type of workhorse to share the load. It increased the cost of the machine, but their go-to-market was to cater to the constantly growing need for faster graphics rendering.
To give a bit more context as to why Jensen’s bet was such an innovative and bold one, it’s helpful to understand the status quo at the time in the industry. In short, progress in computational power depended on decreasing the size of transistors and reducing consumed power. So, the first chip player who found a way to design a significantly smaller transistor and a way to manufacture it would gain a huge advantage in the market.
Sounds good, except that the rate of progress seemed to follow an observation by Gordon Moore, Intel’s co-founder, known as Moore’s law. It basically stated that significant progress in chips would occur every 2 years.
https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https://substack-post-media.s3.amazonaws.com/public/images/d0f2bdbe-c71f-4110-a2a1-3ab2e3fd43d0_1500x1000.png
The whole industry relied on that law, accepting that nobody could really break it. At least not consistently. Which is kind of funny when you think about it: the leader of the incumbent company coining a law that in a way fends off competition.
But, Nvidia’s radically different positioning in the computational arena was based on Jensen’s belief that Nvidia could actually provide an equivalent improvement of computational power in one-third of the industry’s cycle by focusing on increasing the number of transistors in their GPUs (not CPUs), taking full advantage of the graphics pipeline approach.
This approach was revolutionary, and their GPU strategy was so successful at setting up Nvidia’s advantage for 2 big reasons:
- It enabled multiple lightning strikes within its category—driving big attention waves. Being able to announce a GPU that was 2 times as fast as the previous one, and 3X faster than competitors, meant they got a ton of large-scale marketing. This built their brand and allowed them to steal market share from Intel.
- It enabled them to build a data and experience flywheel much faster. Thanks to their ultra-fast iteration cycles, Nvidia’s team gained experience in how to turn their technological designs into sellable products three times faster than their competitors. And since the available human capital in the industry was limited, experienced engineers were another benefit that amplified Nvidia’s competitive advantage—a cornered resource superpower.
The first big bet that paid off.
But, unfortunately, the physical limit in the chip world does exist, as you can only make a transistor so small.
So, despite forking in the cash and watching their public company’s stock price climb, Jensen saw the limit approaching. He saw Nvidia losing their coveted advantage. And being a badass who never rests on his laurels, he went back to the strategy board and maneuvered the ship once again.
Expanding beyond games, and unlocking industries with the CUDA platform
From 1999 to 2006, Nvidia’s bread and butter was servicing the gaming market with their newly minted GPU hardware.And this stage was foundational, not just for the company, but for human progress. Simply, because GPUs set the stage to reshape the computing industry.
But in 2006, they opened Pandora’s box to how GPUs could be used by launching CUDA—a general-purpose programming model that allowed developers to use their Nvidia GPUs for other things.
This was Jensen and Nvidia’s first platform strategy.
To drastically oversimplify it, this made the power of accelerated computing fully accessible, allowing smart people to unlock and harness their GPU’s processing ability beyond games. This breakthrough allowed for multiple computations at a given time instead of sequentially, resulting in much faster processing speeds—a game-changing move in the development of applications for various industries, like aerospace, bio-science research, mechanical and fluid simulations, climate modeling, finance, professional visualization, energy exploration, autonomous driving, and AI, which all require big number crunching to work.
In other words, CUDA was a monumental development in the industry that took Nvidia from a hardware product for gaming to an industry-opening platform.
Multi-market platforms and their network moat
Nvidia’s first platform had three key players:- Software developers: GPUs are specialized hardware and need highly skilled hardware programmers to code. However, CUDA’s brilliance is that it brings the convenience of software development to specialized hardware products. When they launched in ‘06, Nvidia went all in chasing them down and getting them accustomed to the platform—knowing they were essential to the platform flywheel.
- Hardware manufacturers: CUDA is all software, but it needs hardware to run with it. Jensen decided to keep this component closed, meaning only Nvidia Taiwanese-made GPUs can run CUDA. This turned out to be a fantastic move in (1) creating defensibility, and (2) providing superior performance to users as it allowed Nvidia to rapidly iterate on better designs and bring the best-in-class hardware-software integration to the developers.
- Consumers: This includes all the customer profiles who would need fast computing.
That flywheel was, and still is, fueled further by two other elements:
- Strong partnerships: Nvidia’s partnerships with other big tech companies (Microsoft, Amazon, IBM, etc.) are another competitive advantage. This was especially true in their GTM phase of CUDA, as it helped them reach new customers and markets. They’ve also partnered very closely with the game developer community, and this tight feedback loop has given them a massive advantage in the gaming industry, as is evident by Nvidia being the preferred choice by a landslide for game developers.
- Robust sales and distribution channels: Nvidia sells their chips through a bunch of channels, including direct sales, original equipment manufacturers (OEMs), and e-commerce platforms.
AKA, Nvidia is poised to provide for multiple gold rushes.
A testament to the power of solving hard problems in the background.
What you can do with this
When demand is up, build your position in the arena from a different point of view
Study the market to spot emerging opportunities, don’t rest on your laurels, and study your competitors to understand their most important strategic decisions.
Then build a coherent strategy based on the following:
- An understanding of where you are
- A different point of view about what the big opportunity ahead is
- An honest analysis of the challenges/problems that stand in your way
- A fair assessment of your internal capabilities, and your advantages
Last edited:
How many gazillions is 2,000 trillion?NVIDIA have announced DRIVE Thor SoC will be based on Blackwell GPU architecture.
View attachment 59364
And here's something I think is VERY INTERESTING! Thor will be paired with a "YET TO BE NAMED GPU"!!!
View attachment 59366
EXTRACT
View attachment 59365
Blackwell GPU power onboard NVIDIA Drive Thor will unleash Gen AI experiences in your upcoming car
Your next car might be more chatty, and context-aware thanks to generative AI and LLM processing capability brought on by NVIDIA Drive Thor and its Blackwell GPU power. #nvidia #drivethor #genai #blackwellgpu #autonomousvehicle
Fullmoonfever
Top 20
Part 2
Now, if you’ve learned anything about Jensen so far—the man who has no problem betting the farm for a village—you know another big company gamble is coming along.
Here’s a quote by Jensen. It’s somewhat random to insert here, but I think it paints a great picture of who Nvidia’s captain is.
Now let’s see, pun absolutely intended, how the chips landed with his 2012 bet.
Today, it’s Nvidia’s biggest driver of growth, and coupled with their Omniverse and AI platforms, it plays a crucial role in the building of the 3D Internet (what Nvidia calls the metaverse) and AI development.
For instance, back in 2012, Nvidia sparked the era of modern AI by powering the breakthrough AlexNet neural network. They knew AI was coming long before it was hot, and they knew the type of horsepower the engine for it would demand.
So, seeing that (1) the limit was nearing, and (2) there would be demand for computational power far beyond what single PCs or even servers could provide, he leaned into Nvidia’s expertise to invest heavily in a new, third, class of programmable processors:
And I trust him.
https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https://substack-post-media.s3.amazonaws.com/public/images/ea781045-7798-4192-8d29-abf115184e15_828x552.png
Elon sending it.
Essentially, Nvidia realized the future of accelerated computing was going to be a full-stack challenge, demanding a deep understanding of the problem domain, optimizing across every layer of computing, and all three chips—GPU, CPU, and DPU.
But as we’ve seen multiple times in this newsletter—selling products alone doesn’t get you to the 4-comma club. Although, selling a chip with unparalleled demand at $30K a pop won’t hurt you.
No, only building an integrated platform will get you there.
Which is exactly what they’ve done. In partnership with AWS, they’re building out the world's most scalable, on-demand cloud AI infrastructure platform optimized for training increasingly complex large language models (LLMs) and developing generative AI applications.
But Nvidia’s massive data center strategy isn’t their only platform play besides CUDA.
An homage to HBO’s Silicon Valley and the iconic data center
Nope. So let’s talk about the other two:
The latter folks not being in the business of caring whether they found it or not, since selling the tools was a sure-fire way of making money without the risk of investing in finding gold, which had huge uncertainties.
In the world of AI, Nvidia is in the same supply chain business. And not just by providing the hardware and data centers (Chips) to support all the AI players, but by selling builders access to their computation services and software to actually build and train their AI models (Systems).
It’s called the Nvidia DGX Cloud, and it’s an AI supercomputer accessible from your browser.
The DGX Cloud service already includes access to Nvidia AI software, AI frameworks, and pre-trained models. Through it, Nvidia is also providing tools for building, refining, and operating custom large language and generative AI models.
At a high level, their full-stack AI ecosystem looks like this:
Very clearly, they are at the outset of building an AI-as-a-service (AIaaS) business component, and it is having a fundamentally transformative impact on the business, and more significantly, the entire AI industry.
As The Motley Fool wrote, this early leadership could turn into a lasting competitive advantage, creating a defensible flywheel for:
Now, combining their AIaaS with Nvidia Omniverse (below), we see that Nvidia has created a two-piece platform-as-a-service play.
What you can do with this
Subscribe
It’s a platform (also a chips-and-systems play) for virtual world-building and simulations, focused on enterprise customers. It’s based on Universal Scene Description (USD) technology, which, I know, means nothing to either of us. All you need to understand is that it’s the solution to everything 3D-related. Originally invented by Pixar, USD is well positioned to become the open standard that enables 3D development.
Reading up on their Omniverse platform, and it’s clear they have 3 main objectives here:
Rather, they’re betting on the more practical and realistic metaverse by looking at players like:
One epic example of this is how Nvidia is creating a digital twin of the planet—named Earth-2—to help predict climate change decades in advance. It’s well worth the 1 minute 29s watch of Jensen explaining how their supercomputer is helping us save the world.
Super cool.
Like I said, Nvidia is at the forefront of some of the most consequential initiatives and is probably the best example we’ve seen of how engineering innovations really do move the world forward.
What you can do with this
When your category evolves, double down on your capabilities. Just because your previous strategic advantage can’t be maintained, it doesn’t mean it’s worthless.
The benefit of being a category queen is that:
Okay, so just to regroup real quick, we’ve seen 3 key platforms (all relying on their hardware) that Nvidia uses to drive their growth.
But as we’ve seen already with winning platforms like Stripe (read deep dive), Shopify (read deep dive), and Epic Games (read deep dive)—the best of them don’t just capture markets super effectively, but, they actively work to make their markets bigger.
Shocker, but with a market cap of 7X those three companies combined, Nvidia is no different.
Let’s run through them real quick.
All in, they have over 15K startups that have been incubated in this program. And what’s interesting given you’ve never heard of it, is that those startups have raised an on-par amount of funding as the market cap of TechStars’s portfolio companies. (~$100B)
This is a powerful investment strategy that remains largely under-appreciated, as it drives adoption of their own hardware and software (locking in the next wave of startups to Nvidia), as well as gives Nvidia exposure to startups now more likely to flourish and push the boundaries of AI, data science, and HPC.
And then, as part of Nvidia’s “seed the ecosystem” platform, they recently launched an arm focused on driving bigger investments into these startups.
As you can tell, Jensen loves a platform.
The creation of their Deep Learning Institute (DLI) has proven to be a super valuable initiative with lots of benefits for Nvidia’s business and the wider community. Most significantly, it’s become a key driver in advancing knowledge and expertise in AI, accelerated computing, data science, graphics, and simulation.
Yet again, ensuring Nvidia’s flag is right there at the forefront of breakthrough thinking and innovation across most high-growth, deep-tech, sectors.
Through these three hidden gems, Nvidia demonstrates that they have been brilliant at creating scale for their business through partnering, investing, co-creating, and teaching their customers and partners their technologies to solve real-world problems.
What you can do with this
It’s rare to be in a position where you can grow your TAM through investing and incubating startups. Your business needs to be substantially large to do that. But here are a few other takeaways Nvidia bring us that are more realistic for most of us reading:
Now, if you’ve learned anything about Jensen so far—the man who has no problem betting the farm for a village—you know another big company gamble is coming along.
Here’s a quote by Jensen. It’s somewhat random to insert here, but I think it paints a great picture of who Nvidia’s captain is.
What a legend. 🫡
Now let’s see, pun absolutely intended, how the chips landed with his 2012 bet.
How They Grow: Powering The Next Stage Of The Internet
Over a decade ago, in a more suave setting than a bullet-ridden Denny’s Diner, the Nvidia founders reviewed the landscape yet again and spotted the next big wave: The data center.Today, it’s Nvidia’s biggest driver of growth, and coupled with their Omniverse and AI platforms, it plays a crucial role in the building of the 3D Internet (what Nvidia calls the metaverse) and AI development.
The Nvidia Data Center: One platform, unlimited acceleration.
So far, we’ve covered how there were two programmable (AKA, follows specific instructions) processors:- The CPU: Intel’s main competency, with transistors focused on linear computing. For years, it was the sole programmable element of a computer.
- The GPU: Nvidia’s expertise, with transistors focused on parallel computing. Initially used for rendering 3D graphics, GPUs’ processing efficiencies make them ideal for things like crypto and AI.
For instance, back in 2012, Nvidia sparked the era of modern AI by powering the breakthrough AlexNet neural network. They knew AI was coming long before it was hot, and they knew the type of horsepower the engine for it would demand.
So, seeing that (1) the limit was nearing, and (2) there would be demand for computational power far beyond what single PCs or even servers could provide, he leaned into Nvidia’s expertise to invest heavily in a new, third, class of programmable processors:
- The DPU (Data Processing Unit). To avoid the technical mumbo-jumbo here, just know that these chips are ultra-efficient in data functions and are the building blocks of hyper-scale data centers. The DPU is at the center court of the computational arena today, with insane demand since it powers AI.
And I trust him.
https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https://substack-post-media.s3.amazonaws.com/public/images/ea781045-7798-4192-8d29-abf115184e15_828x552.png
Elon sending it.
Essentially, Nvidia realized the future of accelerated computing was going to be a full-stack challenge, demanding a deep understanding of the problem domain, optimizing across every layer of computing, and all three chips—GPU, CPU, and DPU.
But as we’ve seen multiple times in this newsletter—selling products alone doesn’t get you to the 4-comma club. Although, selling a chip with unparalleled demand at $30K a pop won’t hurt you.
No, only building an integrated platform will get you there.
Which is exactly what they’ve done. In partnership with AWS, they’re building out the world's most scalable, on-demand cloud AI infrastructure platform optimized for training increasingly complex large language models (LLMs) and developing generative AI applications.
But Nvidia’s massive data center strategy isn’t their only platform play besides CUDA.
An homage to HBO’s Silicon Valley and the iconic data center
Nope. So let’s talk about the other two:
- Nvidia’s AI Platform-as-a-Service
- Nvidia’s Omniverse Platform
Chips & Systems: AIaaS— a classic Pick-and-Shovel play
The Pick-and-Shovel strategy is a nod to the California gold rush in the 1800s, where there were two types of people: Those who looked for gold, and those who sold the tools (picks and shovels) to mine it.The latter folks not being in the business of caring whether they found it or not, since selling the tools was a sure-fire way of making money without the risk of investing in finding gold, which had huge uncertainties.
In the world of AI, Nvidia is in the same supply chain business. And not just by providing the hardware and data centers (Chips) to support all the AI players, but by selling builders access to their computation services and software to actually build and train their AI models (Systems).
It’s called the Nvidia DGX Cloud, and it’s an AI supercomputer accessible from your browser.
The DGX Cloud service already includes access to Nvidia AI software, AI frameworks, and pre-trained models. Through it, Nvidia is also providing tools for building, refining, and operating custom large language and generative AI models.
At a high level, their full-stack AI ecosystem looks like this:
- AI Supercomputer: An all-in-one AI training service that gives enterprises immediate access to their own cloud-based supercomputers
- AI Platform Software: The software layer of the Nvidia AI platform, Nvidia AI Enterpriseﾠpowers the end-to-end workflow of AI. Simply put, it streamlines the development and deployment of production AI.
- AI Models and Services: Nvidia AI Foundations are cloud services within the supercomputer for customizing and operating text, visual media, and biology-based generative AI models. Think of them like Shopify plugins that make the Shopify platform more flexible and useful.
Very clearly, they are at the outset of building an AI-as-a-service (AIaaS) business component, and it is having a fundamentally transformative impact on the business, and more significantly, the entire AI industry.
As The Motley Fool wrote, this early leadership could turn into a lasting competitive advantage, creating a defensible flywheel for:
Looking at Nvidia’s success and rigor in being strategically nimble, I’m betting this is exactly what will happen.
Now, combining their AIaaS with Nvidia Omniverse (below), we see that Nvidia has created a two-piece platform-as-a-service play.
What you can do with this
- Anticipate industry shifts: Nvidia foresaw the limitations of CPUs and GPUs and invested in the DPU before the demand skyrocketed. Always be forward-thinking and anticipate where your industry is headed.
- Diversify your offerings: Nvidia didn't just rely on selling chips; they built an integrated platform. Diversifying your product or service offerings can open up new revenue streams and make your business more resilient.
- Leverage partnerships: Nvidia's partnership with AWS allowed them to build a scalable AI infrastructure platform. Find the right industry leaders or complementary businesses to amplify your reach and capabilities.
- Platforms, not just products: Platform plays, as we’ve seen time and again, can lead to exponential growth. Platforms create ecosystems where third-party developers or businesses add value, leading to huge network effects.
- Solve multiple Jobs-to-be-done: Nvidia's full-stack AI ecosystem is designed to simplify the AI development process for developers. Always think about how you can remove friction and make it easier for your customers to achieve their goals.
- Build for the long-term: Nvidia's investments in AIaaS and their platform strategy are long-term plays that could give them a lasting competitive advantage. A great example of how to apply long-term thinking is here: Ants & Aliens: Long-term product vision & strategy
- Stay nimble: Despite being massive, Nvidia has shown agility in their strategic decisions and ability to pivot into new areas. Regardless of size, be mindful of the traps that lead companies into the innovator’s dilemma.
Subscribe
Nvidia Omniverse: The platform for the useful metaverse
If Nvidia AI is their answer to every artificial intelligence question, then Omniverse is Nvidia’s answer to every metaverse question—the 3D evolution of the internet.It’s a platform (also a chips-and-systems play) for virtual world-building and simulations, focused on enterprise customers. It’s based on Universal Scene Description (USD) technology, which, I know, means nothing to either of us. All you need to understand is that it’s the solution to everything 3D-related. Originally invented by Pixar, USD is well positioned to become the open standard that enables 3D development.
Reading up on their Omniverse platform, and it’s clear they have 3 main objectives here:
- Make Omniverse the go-to platform for Augmented Reality and Virtual Reality (AR/VR) development
- Sign up major partners to use Omniverse
- Create a strong and defensible ecosystem around Omniverse
- In order to become the prevalent system behind every 3D project, they’re focusing on offering the best simulation tech to the biggest companies and supporting every business that builds applications on top of Omniverse.
- In order to make 3D content creation and sharing as easy as possible, they’re focusing on individual and team collaboration tools for real-time interaction with partners and clients, world-building tools, a library of prebuilt objects and environments, compatibility with other sources of graphics and formats, and the leading game engines, Unreal and Unity.
- They’re securing partnerships with major cloud service providers (Azure, AWS, Google Cloud) to adopt Nvidia’s tech and architecture, essentially forcing the competition to use their technology, in turn, almost eliminating it.
Rather, they’re betting on the more practical and realistic metaverse by looking at players like:
- Artists of 3D content
- Developers of AIs trained in virtual worlds
- Enterprises that require highly detailed simulations
One epic example of this is how Nvidia is creating a digital twin of the planet—named Earth-2—to help predict climate change decades in advance. It’s well worth the 1 minute 29s watch of Jensen explaining how their supercomputer is helping us save the world.
Super cool.
Like I said, Nvidia is at the forefront of some of the most consequential initiatives and is probably the best example we’ve seen of how engineering innovations really do move the world forward.
What you can do with this
When your category evolves, double down on your capabilities. Just because your previous strategic advantage can’t be maintained, it doesn’t mean it’s worthless.
The benefit of being a category queen is that:
- You identify your category’s evolution early
- You get to convert your advantage to serve you in the emerging new category
- Complacency: falsely believing that you can sustain your current competitive advantage (see The Innovators Dilemma)
- Overreaching: abandoning your advantage and trying to build new capabilities from zero
Okay, so just to regroup real quick, we’ve seen 3 key platforms (all relying on their hardware) that Nvidia uses to drive their growth.
- Their CUDA platform gives them exposure to various general domains (gaming, automotive)
- Their Data Centers + AI Platform gives them exposure to the rapidly growing AI sector
- Their Omniverse Platform gives them exposure to the professional metaverse
But as we’ve seen already with winning platforms like Stripe (read deep dive), Shopify (read deep dive), and Epic Games (read deep dive)—the best of them don’t just capture markets super effectively, but, they actively work to make their markets bigger.
Shocker, but with a market cap of 7X those three companies combined, Nvidia is no different.
Seeding their own multi-market ecosystem: A lesson on expanding your TAM
Nvidia grows their own market in three powerful ways:- By incubating AI startups
- By investing in companies integrated with the Nvidia ecosystem
- By offering formalized training and expertise development for their chips and systems.
Let’s run through them real quick.
1. Nvidia’s Inception AI Startup Program
Inception is a free accelerator program designed to help startups build, grow, and scale faster through cutting-edge technology, opportunities to connect with venture capitalists and other builders, and access to the latest technical resources and expertise from Nvidia.All in, they have over 15K startups that have been incubated in this program. And what’s interesting given you’ve never heard of it, is that those startups have raised an on-par amount of funding as the market cap of TechStars’s portfolio companies. (~$100B)
This is a powerful investment strategy that remains largely under-appreciated, as it drives adoption of their own hardware and software (locking in the next wave of startups to Nvidia), as well as gives Nvidia exposure to startups now more likely to flourish and push the boundaries of AI, data science, and HPC.
And then, as part of Nvidia’s “seed the ecosystem” platform, they recently launched an arm focused on driving bigger investments into these startups.
As you can tell, Jensen loves a platform.
2. Nvidia’s Venture Capital Alliance Program
Here is what Nvidia said about the program when they launched it in 2021:Again, this furthers the same advantage Nvidia gets from their accelerator—it helps drive the scale of their customers. As they grow, so does Nvidia’s TAM.
3. Nvidia’s Deep Learning Institute
I love it when companies layer in education and training plays. And for a company like Nvidia, which needs more than a thoughtful onboarding experience to get a new customer setup, having meaningful learning material and access to exert resources is crucial.The creation of their Deep Learning Institute (DLI) has proven to be a super valuable initiative with lots of benefits for Nvidia’s business and the wider community. Most significantly, it’s become a key driver in advancing knowledge and expertise in AI, accelerated computing, data science, graphics, and simulation.
Yet again, ensuring Nvidia’s flag is right there at the forefront of breakthrough thinking and innovation across most high-growth, deep-tech, sectors.
Through these three hidden gems, Nvidia demonstrates that they have been brilliant at creating scale for their business through partnering, investing, co-creating, and teaching their customers and partners their technologies to solve real-world problems.
What you can do with this
It’s rare to be in a position where you can grow your TAM through investing and incubating startups. Your business needs to be substantially large to do that. But here are a few other takeaways Nvidia bring us that are more realistic for most of us reading:
- Think beyond direct sales: Nvidia's approach demonstrates that market leadership isn't just about selling products. It's about creating an ecosystem where your products become indispensable. The sweet spot: you’re able to create an environment where your product is deeply integrated into the fabric of an industry.
- Innovate across multiple fronts: Nvidia isn't just innovating in terms of products. They're innovating in terms of how they engage with startups, how they connect with VCs, and how they educate the market. Always be thinking about how you can innovate across multiple fronts to drive growth and market leadership.
- Layer in education: Offering educational resources or training can help customers get more value from your products and can position your company as a thought leader. Consider how you can educate your customers, whether it's through online courses, webinars, or other resources.
- Create alliances with VCs: If you're in a high-growth industry, consider how you can facilitate connections between investors and startups that use or complement your products.
- A rising tide lifts all boats: Nvidia actively works to make their market bigger by supporting their customers' growth. Always be thinking about how you can expand your TAM, whether it's by entering new markets, creating new use cases for your products, bringing non-customers into the market, or driving innovation in your industry that attracts investment.
Similar threads
