The Tech Sales Newsletter #75: It’s NVIDIA’s world, you are just living in it
Source: FinChat
The majority of the value accrues at the bottom of the stack.
This is one of the most fundamental principles in tech sales, and any investor or a sales rep would be wise not to ignore the implications. As we enter 2025, it's time to turn our eyes towards the intersection of hardware and software at the bottom of the cloud infrastructure technology stack.
As LLMs and Enterprise-grade ML have taken over the industry, one company has become the most influential player that we've ever seen. Many outsiders look at NVIDIA and try to write them off mentally in order to feel better about themselves:
"Oh, it's overvalued."
"Others will catch up with GPU production."
"They made a great bet, but going forward other companies will cut into their business."
"Did you see Cisco back in the day?"
The elephant in the room, of course, is, what happens if they are wrong? What if the company that has been the foundational player in shaping up the next step of computing, well, keeps pushing?
What happens when you combine huge cashflows, big vision and the competence to execute? What happens when the guy that got them here, keeps on pushing harder than ever?
This is a "what if Steve Jobs didn't die" moment in the industry.
The key takeaway
For tech sales: If NVIDIA accomplishes its mission, tech sales as we know it will have to fundamentally change. The company is deeply invested in providing the hardware and software needed to run a "robotic" version of the world. They have all of the momentum, vision, capability and wider ecosystem support to achieve their goals. I would advise you to assess what that can mean for you, based on your current knowledge and performance levels.
For investors: If the big growth for NVIDIA in 2023-2024 was driven by massive demand for H100s that were used for model training purposes, the next two years will be focused on delivering computing power that's highly efficient and capable of servicing large volumes of inference. Smart investors should think along with NVIDIA about who the big winners will be from this big spike in inference usage.
The keynote
In the first section of this article, we will focus on the key themes from the NVIDIA presentation. At the start of the presentation we learn about the concept of the "AI Factory", which NVIDIA first introduced last year.
"AI factories are a new form of computing infrastructure. Its purpose is not to store user and company data or run ERP and CRM applications. AI factories are highly optimized systems purpose-built to process raw data, refine it into models, and produce monetizable tokens with great scale and efficiency,"
The idea here is that general computing data centers predominantly serve as storage and application capacity. The NVIDIA future relies on moving towards a world where AI-driven activity will depend on token generation across local, edge and data center compute. Everything will revolve around the token, as the token powers the intelligence and the intelligence powers the actionable outcomes.
If LLMs kick-started the generative AI era, which is now quickly moving towards an agentic AI, the real value in front of us will be realized when AI can independently influence the physical world. What is quickly becoming clear is that this will not happen based on collecting real-world data alone; simulating as many alternative scenarios as possible is the best path forward for accelerated computing.
The best place to observe this change is the dream of autonomous driving. Tesla is moving closer to launching FSD into the world, and Waymo has been able to prove the viability of the technology for daily usage. The algorithms work and they work well. What's needed is a lot more real-world data to iron out the edge cases - this is where simulated environments will bridge the gap and accelerate the timeline to making this widely available within 2 years.
If autonomous driving proves that we can solve the spatial challenge of using AI for real-world outcomes in geometric space, then the logical next step is to extrapolate this into robotics.
The fundamentals of the hardware have been around for a while, but the software needed to actually make robots functional outside of an assembly line simply didn't exist. What happens if we now have all of the ingredients to build these?
What if we now have the ability to create robotic workflows? Whether these are robot AI agents, robot factories, robot cars or humanoid robots, the core premise of software operating independently and driving outcomes is becoming not just a concept but a very practical reality.
The fireside chat afterwards
Ben Reitzes: First, I just want to say thanks for having us all here. Really grateful for how you treat all of us. Look how many people are here, obviously.
It's very unique and, very appreciated. You made this job fun again.
While the X influencers were spamming useless threads with "the biggest announcements" for engagement bait, the real conversation continued with a select group of analysts and investors. This is an AI-generated transcript, so it's not a perfect outline, but you get the message:
Colette Kress: When we think about the demand that's in front of us, it's absolutely a growth year. How we look at every path in terms of every quarter, we're going to try, of course, our best, but we only guide one quarter at a time at this time. But keep in mind, our focus is additionally now with more advanced systems, bringing those to market. And again, we do believe the growth is in front of us in terms of this new fiscal year in front of us.
Harlan Sur: Your next generation Blackwell Ultra set to launch in late 2025 and in line with the team's aggressive annual product cadence. Help us understand the decision, purchasing dynamics with your customers given that you'll still be ramping prior generation Blackwell solutions at the same time that you'll be rolling out Blackwell UltraWright? How do you and your customers and the supply chain, I mean, manage the simultaneous ramp of 2 product families?
Colette Kress: So a great effort all around. But let's understand in terms of why. Why is this important? We're in one of the largest transitions here. And each and every time that we're coming to market with new product opportunity is additional advancements, additional understanding in terms of where AI is moving towards, and then helping each one of our customers with the path that they have for AI and the types of solutions that they want to bring to market.
We have our own roadmap, but our customers do as well. We are matching their roadmap to our roadmap to say, what project do you want to bring to market at that time? Which one of our systems are going to be enabled for that different project that you have? So this is a great opportunity for some of the most advanced systems and those and some of our future generation things of AI that you'll see moving, to our top, top products. But you also now have the ability to swing our enterprises to create AI factories even with existing and existing libraries and software and NIMs to actually help them to get market.
The primary reason for the NVIDIA presentation was the launch of their Blackwell family of accelerated computing hardware, including the GB200 platform. This is an important moment since it's significantly more optimized for inference (allowing for much faster queries) while offering much more efficient energy consumption. In the discussion above, the analyst is probing on how this is going to impact demand, as if NVIDIA is just waiting on orders from their webshop.
The reality is that they are working on multi-year roadmaps with all of their customers, planning capacity and architecture to fit specific long-term goals and demands from those organizations. Imagine being the rep behind one of these.
Harlan Sur: This whole notion of test time compute inferencing, right? So how does the NVIDIA team think about the continued scaling of models and the requirements for more GPU compute capability going forward?
Colette Kress: Yes, we do. We do believe that there are multiple scaling laws and many of them that you discussed. For a long time, folks just thought about the training and the pre training that you train it once and it's done and you move on. And actually, in the same manner that you're always learning, you're always getting educated, you're also always training. And that training is important to think that that post training can incorporate many different aspects.
And as you discussed, you can move to synthetic data that says, can I approximate something in the future using any synthetic version to also expand the model and improve the model for use? And so you're going to see both the pre and the post training and that verification process continue. But there's a yet a third focus as well. This is the time to token, which is so essential. What you see today is chat GPT, you ask it a question, off we go, and we get an answer.
But there is this part where reasoning and long thinking is going to be essential to the work that is necessary in many of this inferencing and many of the different types of questions that will be there. So this is, again, where a lot more scaling needs to happen. This is an important reason why a system like a GB200, which is not only focused in terms of the ability to train, but a massive improvement in terms of what it can do for inferencing and what it can do in terms of saving money and saving time to token. These are the important parts as this reasoning and scaling of inferencing is going to be an important part.
The big explosion of demand for the H100s was driven by the big leap in performance with GPT3.5 which led to multiple players in the space to deploy large clusters of H100s to do model training. This was an extremely capital intensive investment and in 2024 it become clear that it will not scale to infinity (i.e. you can’t make ChatGPT 5 ten times smarter by training it on ten times the compute). More recently, with the introduction of OpenAI’s o1 reasoning model, the concept of test-time compute become a big focus as the next step up in performance.
Fundamentally, test-time compute requires repeated token generation in order to get to a result. AI Agents will consume a lot of compute, both for one-shot prompting and for reasoning. Robots will consume a lot of compute, in order to communicate with humans and think trough their next autonomous decision.
Colette Kress: : Because even if you think you just have capital, many think about the work in terms of data centers in a 4 to 5 year projection, meaning they are looking at purchasing data center space really that far out. It is a necessity because it's not something that can start up in a month. It is something that needs long thought in terms of where, how much and how to continue to make it some of the most efficient type of data centers. So their needs for data center, their need also in terms of the right types of power. Power for the last several decades has been the same on the grid.
Not much has moved from that. And you're going to see a big movement, a movement in terms of how that grid was used and also new forms in terms of power to move that. But the work that we're doing is not only in terms of the best performance, but what we can also do to lower the cost overall and also the most efficiency that we can get in terms of sustainability as well. That's what accelerated computing and AI brings together. The journey that we're on doesn't see a path of stopping the journey that we see going forward as more to come is probably one of the most unique times where you see the adoption of a very important transition has taken place worldwide at one time.
The Blackwell architecture is focused on efficiency and inference gains. Model training will continue on H100s, but for real-time inference, this architecture will take over as the primary choice for the key players in the industry. Since everybody is constrained by electricity and space challenges, these machines will start replacing regular data centers. This is a good moment to explain accelerated computing:
Analyst: So I think you knew where I was going with that second question. So where do you see the lines between GPU and custom ASIC evolving? Has that changed in your mind at all?
Just curious if kind of given some of the recent reports from some others, how that's evolved in your mind? Thank you.
Jensen Huang: Yes. A couple of things. First thing is, take a step back and even just ask yourself where why is that question important today? And the reason why it is, is because it is now universally clear that accelerated computing is the path forward. I think that's one grand observation, that general purpose computing is done.
Hand coding instructions on CPUs as a way of developing all softwares over. And that the future is likely machine learning accelerated GPUs, neural networks and so on and so forth. And so that's the first observation that we are seeing a across the board recognition by literally every computing company on the planet that machine learning and accelerated computing is a path forward. And so you got to take a step back and just say, okay, what does that imply? Well, that implies that every data center in the future is going to be built different, literally.
But going forward, you know, if you're not putting accelerated computing and machine learning capable systems in your data center, you're doing something wrong. Nobody should be building a data center that is filled with a whole bunch of general purpose computers in the future. Makes no sense.
What the analyst is asking for here is: "Will Microsoft and AWS making their own servers cut into your business?" The answer from Jensen's point of view is that those players have already agreed to operate from his perspective.
Accelerated computing is the use of specialized hardware to dramatically speed up computational tasks through parallel processing, making computing more efficient and powerful. The goal is to have your compute be most effective at the tasks that make the biggest difference - which going forward will be AI-related activity.
By accepting the premise that AI-driven computing is the future of all computing, all important players in hardware and software have taken a position that is fundamentally, well, NVIDIA's game.
Jensen Huang: Rather than a custom ASIC for one company, we're building a computing platform for every developer, for every company. And so we're not building a custom ASIC, we're building a computing platform. And the benefit of a computing platform is that you never know what's the next amazing application you're going to attract. So let me give you an example. All of a sudden, we attracted world foundation models.
All of a sudden, we attracted robotics. All of a sudden, we attracted, the ability to do computational lithography for TSMC. All of a sudden, we attracted all these different things. A computing platform has the benefit of computers, the thing that we all valued about the last 60 years. We don't want to use the same architecture anymore because it's a general purpose, but we loved the concept of a programmable architecture.
In the last couple of 2, 3 years, what has happened? Everything from incredible new ways of doing attention, the attention is the mechanism for transformers, this incredibly computationally intensive thing. The invention of large context windows, speculative decoding, a whole bunch of new technologies are being invented. As a result, we now have this multiple ways of doing scaling. So much of it is invented on NVIDIA GPUs.
And the reason for that is because we're easy to program. If you want to change the algorithm, you want to change the architecture, so be it. And there's state space decoders, fantastic. Hybrid versions, fantastic. And so all kinds of different architectural innovations are happening as a result of the fact that we are a computing platform.
You could use us for data processing. Notice all of a sudden the amount of video that you have to process because of these world models and multimodal models. We train Cosmos, our the foundation model we're just about to release, and it would have taken several years on CPUs to process that data. And so instead, we accelerated the entire pipeline and you do it in hours.
Jensen is not worried that AWS will beat him at the game tomorrow because he literally supports them in building their own custom hardware.
Jensen Huang: But don't forget, the world of computing is large. There are sovereign AI systems to be built there. It's going to AI is going to be part of the national infrastructure the way telco is part of a national infrastructure. Every country has their own telcos. They all have their own banks, they have their own etcetera, etcetera.
And so the ability for every country to be able to build their own national infrastructure, you can't do that with a custom ASIC. And so our opportunity in enterprise, private clouds, in regions, in edge, in robotics and self driving cars, in all of the universal all of the other accelerated computing simulations or I mean computation you have to do, our reach is just a ton broader. If you're going to cut that little tiny sliver of an ASIC, it's going to be hard to do. And then the third point that I'm going to make is this is hard. There's some evidence that it's hard.
Look at the number of startups that have been started and not successful. It's incredibly hard. And the reason for that is because it's not a chip problem. This is definitely a systems problem, and this is definitely a full stack problem. The software stack necessary to make one of these systems useful is not about 10 engineers.
We're talking thousands of engineers, and it's going to take years to make it great. And how many stacks can the world make? That's the next question. How many stacks can the world make? The number of stacks that the world can currently sustain is about 2.5, 3, barely.
And so that kind of puts it in perspective. Now when the world then goes off and builds 20 stacks, is that good for us or less good for us? Surprisingly, the answer is it's better for us. I'd rather compete against 20 competitors in a world where there's limited resources than one, right? It's very logical.
The critical work done by NVIDIA in the last 10 years is not limited to CUDA and the machine learning libraries they supported. It's the hardware, the networking and the software, all on the same stack, scaled on top of tens of thousands of developers and researchers.
How many alternative stacks should really exist? And what would be the benefit of those at a time when the primary stack is clearly enabling a highly accelerated path forward?
Jensen Huang: And and so agentic AI, test time scaling for sure. Fine tuning, not much necessary because you already started with pre trained models, but test time scaling computing intense. Okay? That's one. The second is robotics and AV.
That is going to be data processing and training intense. Look at look at the size of the the AI superclusters that Elon has with Nvidia's hoppers and Ampere's and hoppers. It's gigantic. Every car company in the world will have 2 factories, a factory for building cars and the factory for updating their AIs. Makes sense?
Because every single car company will have to be autonomous or you're not going to be a car company. And so we know now everything that moves will be autonomous. That's a foregone conclusion. We know now that Waymo has got right they have they have turned a corner. I mean, that that's clever.
Right? We think Waymo's turned a corner, and it's accelerating. We know that FSD version 13 is I thought version 12.5 was incredible, and version 13 is incredible. We know that every single EV company in China has AV capability, every single one. If it's EV, it's got AV.
That's going to be the standard. And so that's going to put enormous pressure on every single car company on the planet. If you don't have 2 factories, an AI factory along with your car factory, you're not going to be a car company. And so the amount of data processing is intense. Now we're going to have robotics.
You could they're going to be there can only be so many cars because there's so many humans to move around. But there's no limitations to robots. It could very well be the largest computer industry ever. And the reason for that is we don't need more cell phones than people, but robots, you could build as many as you like. And there's a very serious population and workforce situation around the world, as you guys know.
The workforce population is declining and in some manufacturing countries fairly seriously. And they have they have it's a strategic imperative for some countries to make sure that robotics is stood up and productive in the next several years. Their population is not going to grow, not for foreseeable future. Lots of data processing, lots of training, lots of simulation.
Autonomous driving is clearly illustrating the need for AI factories that are generating the tokens needed for the cars to operate, learn and interact with the wider world. If this is proven to be effective, countries, not just companies, will have to invest in scaling their AI factory infrastructure, similar to the investment made in telecommunications.
All of this will run on NVIDIA hardware, software or a combination of both.
Jensen Huang: Robotics is going to be a massive data problem. Lots and tons and tons of videos and human demonstrations and synthetic data generation, as we showed yesterday, a huge data problem. We're going to take all that data, we're going to train AI models for them. Okay? So the first business opportunity for us in robotics is training.
But I have to enable everybody to be able to train at scale, and everybody lacks data. That's the reason why NVIDIA built Omniverse and Cosmos. So that between Omniverse,which is a physics grounded physics grounded, you guys know large language models when it first came out, the criticism was it hallucinated. Remember that? And and notice that the criticism is starting to decline, and the reason for that is because of retrieval augmented generation.
Isn't that right? Grounding it in facts. Okay? Well, if we had a generative model that can generate physics, which is essentially what Cosmos is doing, physics generator, but it's hallucinates, then I can ground it with physics because Omniverse is physically grounded, physics grounded. Does it make sense?
Omniverse is calculated. It's based on principled solvers, and so its physics is known and believed. Okay? And then I connect that to a generative model. Now I can create Doctor Strange.
"Slowly for a while, then quickly." The idea by Jensen here, that humanoid robots can go to market faster than agentic AI because hallucinations can be controlled more easily, is an interesting one. If true, then if we have autonomous cars by 2027, humanoid robots can be in play by 2030. By then, agentic AI workflows would become dominant in the world of knowledge workers. This is expanded further here:
Jensen Huang: So for example, we've had self driving cars for I want to say 8 years, but it's taken 8 years to reach this level of maturity. And the reason for that is because of risk, safety. And but recommending a movie, recommending a product to you, generating a unicorn flying under the sea surrounded by ocean turtles, that's not gonna hurt anybody.
And so depending on the application, each one of them has a different degree of risk. Agentic AI for enterprise applications is not likely to be high risk depending on the functionality it is. In legal, you're going to have to be careful a bit more careful. In accounting, obviously, you have to be more careful. But in marketing, you could probably take a little bit more risk.
In customer service, it's probably domain specific and you'll guardrail it very specifically. In software engineering, very low risk because human's in the loop. In chip design, very low risk because human's in the loop. And so there's a you have to gauge kind of for each one how you would go to market. In the case of robotics, self driving cars, that's very hard.
In the case of humanoid robotics, less hard. And the reason for that is because a car has to drive all over the world and has to drive all over the United States. It's got bumpy roads.
It's got, you know, you got Vegas, you got Arizona, you got San Francisco. And so the domain variation is very high. But humanoid robot, once you bring them into your facility, the domain adaptation is rather limited, rather narrow. And so I wouldn't be surprised if human robotics is much, much faster to deploy if the technology works well. Okay?
And I'm hoping that I'm looking forward to technology working well. And so it's really about the rate of adoption as a function of the complexity of technology, how you would take it to market, the safety of the system, so on and so forth.
If we extrapolate this a bit further, if humanoid robots performing real-world actions can operate mostly independently on a shorter timeline, then the first serious missions to Mars are likely to be staffed by a very limited number of humans (if any) and many AI machines. The most powerful capability of course would be the fact that we can update them remotely - by the time they get to Mars, another 2 years of progress would have occurred on the software side.
Jensen Huang: Quantum computing can't solve every problem. It's good at small data, big compute big combinatorial computing problems. It's not good at large data problems. It's good at small data problems.
And the reason for that is because the way you communicate with a quantumcomputer is microwaves. And it's terabytes of data is not a thing for them. And so just working backwards, there are some very, very interesting problems that you could use quantum computers for, truly generating a random number, cryptography. So these are problems that are small data, big compute. And working backwards, we're a computing company.
We're an accelerated computing company. And as you know, we work with CPUs. We obviously build Grace. We're not offended by anything around us. And we just want to build computers that can solve problems that normal computers can't.
And so in the case of quantum computing, it turns out that you need a classical computer to do error correction with the quantum computer. And that classical computer better be the fastest computer that humanity can build, and that happens to be us. And so we are the perfect company to be the classical part of classical quantum. And so we are working with just about every quantum computing company in the world is working with us now. And they're working with us in 2 ways.
First, this quantum classical, we call it CUDA Q. So we're extending CUDA to quantum. And they use us for simulating their algorithms, simulating the architecture, creating the architecture itself and developing algorithms that we can use someday. And when is that someday? We're probably somewhere between in terms of the number of cubits, order of five orders of magnitude or six orders of magnitude away.
Two of the next "frontiers" of utilizing computing power are quantum and reversible computing. Who is heavily involved in funding research and providing hardware for these? NVIDIA, of course, it was casually mentioned.
Deal Director, more simple, please…
We are in a transition towards a new stage of computing. That has software and hardware implications.
Jensen Huang: Another way to say it is if you were a computer company, if you're a computing company and you're building a data center and that data center is full of CPUs, general purpose computers, you really ought to self evaluate whether you understand computing at all and whether your company is moving forward. And as you know, all the cloud service providers, whatever their CapEx is, whatever their CapEx is, is going to largely go into accelerated computing going forward. They're not going to go, why build more general purpose computers, data centers when the gross margins of renting a CPU is basically below cost? You're renting it basically for free so that you could get their storage.
Why do that? Why not invest in a modern computer where it's margin accretive, actually generates revenues? Okay. So that's number 1. Number 2, the second thing we have to believe is that that AI is a new layer above the computing stack, that it's a new layer above everything that we've done.
And the reason for that is this. The last layer is called software, and software are tools, tools used by humans that we built. We use them. And that layer give let's use an example. My favorite tool, Outlook. I use it. Excel, I use it. PowerPoint, I use it. You know? And so these tools, you know, cadence, synopsis, we use them. They're tools. Well, what AI is is not a tool. AI is an agent, is a robot that sits on top of tools that use tools.
That's what an AI is. Right? And so what is a self driving car? A self driving car is a digital chauffeur. It's not an FM radio. It's not an operating system. It's a digital chauffeur. And so it's an agent that sits on top of the current stack. That layer has never existed before. Do you guys agree with that?
That's why AI is a growth industry. If we're successful, and there's every evidence we're successful, that AI will be a growth industry. And then here comes the next part. Whereas software runs on general purpose computing, the AI that is created where we are fed by our cafeteria and fed off our paycheck, AI is fed off of an AI factory. This is that factory I was talking about to Joe earlier, right?
The software implication will be that users will embrace "tell the thing to do the thing." This was very clear to me already a year ago when I wrote this in #26 of "The Tech Sales Newsletter":
Key lessons for selling GenAI as of Q1'24
The GenAI “hype” is not limited to LLMs, but it covers ML in general, as well as the cloud ecosystem.
Building a platform like the one in the example will require a vector database, a lakehouse, a custom built front-end for the sales reps with a fine-tuned LLM on top, probably a hybrid deployment of environments between on-prem and public cloud depending on the sensitivity of the data accessed, custom ML algorithms that evolve based on new datasets in order to generate the lead scoring.
This platform will then plug into the larger cloud infrastructure ecosystem, including endpoint protection for the servers, an observability platform to keep high uptime of the tool (we can’t have the reps not be dialling in because the system is down), fraud detection and risk assessment systems that are also built on giant lakehouses and continuously improve those algorithms.
Everything is logged continuously for compliance and optimisation purposes.
The majority of this software stack will be procured on a cloud hyperscaler marketplace. Whether or not Azure will make 40% margin per OpenAI Service token consumption is THE LEAST INTERESTING THING about this from the perspective of selling the outcome.
The 55 year old CEO just wants increased loan upsells to the existing customers (safer loans and increased revenue per customer) and the current workflows are not producing those desired outcomes.
End users just want to tell the thing to do the thing.
Approaching LLM and ML adoption from the perspective of a developer point of view is very limiting. We have the highest adoption of compute power across the world today, yet most of it is spent on underpowered applications with no productivity gains.
The reason why LLMs matter is because it helps non-technical users get improved outcomes from using the available computing power. If those sales reps in the example above actually get connected to more people that want to buy and say compelling things that convert customers while on the call, this is a tremendous jump in productivity vs the current state of things.
LLM adoption will be compostable on many small niches going trough a change.
The software implication will be that users will embrace "tell the thing to do the thing." At the time we hadn't yet seen the variety of agentic workflows that became obvious towards mid-2024. If we follow Jensen's logic, AI agents become the new interface layer that will sit on top of the tools that make things. Over time, most productivity will go through this layer (hence very high inference usage), with very few practitioners still using the actual backend tools.
Now if you can just tell an agent to do stuff and part of this includes the agent creating new software tailored to your needs, where does paid software sit going forward? I see four potential outcomes for software as part of tech sales:
Likely: Software and the compute required to run it are no longer approached separately. Cloud infrastructure software essentially becomes the only SaaS category on the market, as everything else is generated on the fly by agents or sits on top of large platforms gated behind agents. The job becomes a field for the elite only, T-shaped professionals with deep industry knowledge, technical competence, top 20% sales skills and strong personal motivation.
Still possible but becoming less likely by the day: No fundamental shift because we can't scale AI one step further. o3 and similar future models are too expensive to run so they become the equivalent of today's "supercomputers", used for academic and niche purposes. Simple agentic workflows are just a feature set.
Bad: AGI becomes a commodity and tech sales goes through heavy automation on both the buyer and the seller side. The majority of headcount is lost, with humans only performing supervision and escalations. Those that remain in the industry will be VP OF AI AGENT SALES, part-techie, part-sales, part-management gig.
Doomer: We accelerate faster than expected and the whole field of buying/selling software becomes a thing of the past, due to heavy corporatization of all useful assets, open-source AI writing all the other code and universal income schemes. Since economic activity appears hardwired into our psyche, those who do sell will probably be focused on compute, electricity and physical places.
I'm guessing this is not what you expected? Is discussing "ten useful email openers" making you feel better?
Last year I wrote the following:
Average tech sales reps want to be left alone.
Good tech sales reps want to lead.
Great tech sales reps want to know the truth.
The truth is that if we think through the logical outcomes of the forces at play (technology, goals of the key players, politics and economics), the only timeline where tech sales continues as we know it is if the vision for AI fails. While there are a lot of very bearish takes that I've seen from otherwise smart people, the majority of arguments on why this will happen seem to stem from wishful thinking due to politics, rather than practical arguments.
Every major player in the industry in terms of competence, money and vision (Musk, Zuckerberg, Bezos, Jensen, Satya, Sergey Brin) appears to be very convinced that AGI and afterwards ASI is happening. The models themselves as available in the market today are a clear step up in capability and outcomes compared to anything we've used before. Not only that, but the players keep focusing on new big deliverables coming over the next 5-10 years, i.e., they assume that AGI is achieved or will be here shortly. Their opinions are not mere words, they are conviction backed with the largest investment we've seen in hardware, ever.
Which brings us to NVIDIA and the hardware side of things.
If AI workflows are going to become the defining way of how we use computing, then specialized hardware that is the most effective at running those workloads becomes critical. NVIDIA offers essentially the only "complete" package on the market - hardware, software, networking and capability to execute.
They work with everybody. They supply everybody. They coach, fund, provide feedback and work with everybody.
If Apple became the most valuable company in the world by dominating B2C computing in the age of mobile, then NVIDIA overtook them by becoming the leading B2B tech company of the age of AI.
This article only covered a part of their activities and vision. We’ve not touched anything about NVIDIA cloud, their devices and consumer hardware or startup investments. They have massive cash flows and most of the top talent in the world trying to join them. Jensen is a workaholic that is fully dedicated to the mission.
The only thing that gave me pause during the CES presentation was him saying "as long as we live," twice. A lot of things right now are riding on key players who are in their 50s and 60s. We need them to squeeze a couple more years of exceptional performance at the same level.
