Skip to content

Podcast

Three AI Rooms to Be a Fly on the Wall in 2026

The Cloudcast

Source ↗ ← All highlights
  • NVIDIA’s Strategic Expansion Potential
    • NVIDIA sits on massive cash and is positioned to expand well beyond chips into software, hosting, and agentic tools.
    • Their M&A choices will reshape AI stack ownership and ecosystem alignment across hardware and software. Transcript: Brian Gracely So first room, I think the most obvious room would be NVIDIA’s M&A room, right? They’re sitting there at, you know, still roughly, you know, depending on the day, roughly $5 trillion in market capitalization. They are throwing off an enormous amount of cash. We’re looking at how much CapEx, you know, between Amazon and Meta and Google and Microsoft are all just planning to spend in 2026 you know a good chunk of that will go to nvidia so they’re Going to have for the most part what’s going to feel like almost unlimited money or at least you know the top end of the market in terms of money we’ve already seen them you know make make A few acquisitions we saw you know 20 billion dollar acquisition of grok there at the end of last year a grok with q i think that right, Grok with a Q. Yeah, the chip company. But it would be really interesting as we think about where the market’s going from them. We’re seeing some people trying to figure out alternative hardware accelerators. So whether it’s Google TPUs or AWS with Tranium or Microsoft apparently has their own projects to spin stuff up. We know that there are some Chinese alternatives that are spinning up. AMD and Intel are still trying to play in that GPU game to a certain extent. So the hardware space, while NVIDIA still has a dominance, again, we’ve talked about this a hundred times in this show, being the dominant proprietary highest end hardware provider Is a difficult place to stay in for a very long time. And so it would appear that you would think that they’d likely be going after assets of a software stack, right? Assets to begin to make up whether they want to try and dominate more on the enterprise side of AI, whether it is on the consumer side of AI, probably not necessarily their play. We begin to see them starting to release some things around autonomous driving. They made some announcements at a consumer electronics show around that. We’ve seen some things around robotics. So it would be very, very interesting just given the vast amount of resources they have and being able to look into the future beyond chips, you know, beyond CUDA as sort of a infrastructure Layer, you know, where do they see the biggest bang for the buck? And probably most importantly, you know, where do they start to see people that they’re customers today trying to commoditize the stack in order to move it up higher, as we’ve seen, You know, over decades with different trends, you know, where software begins to start to do that. So it’ll be very interesting to see if we see them moving into a set of sort of, you know, agentic tools, if everything they’re doing becomes part of the, you know, the NVIDIA AI factory, If they start to, you know, aggressively start to try and find some of the leaders in whether it’s agentic or rag services, or, you know, if they start to try and get into data services Somehow, you know, there’s a lot of different places they could go to fill out the stack to try and do that. Do they take their software and begin to open source it in order to prevent, become sort of the de facto standard and prevent competitors from getting up in the software stack? Do we see them doubling back down and going back into some of the hosting services, right? Today, they’re sort of the arms dealer for those hosting services, but do they try and become more dominant in that? Or do they feel like they’ve got that market segment pretty well wrapped up at this point? Networking is obviously doing very well for them, but they’re not necessarily like in the storage space. Maybe they make an acquisition in that space. So a lot of places for them to go. It’ll be very interesting to see where that goes because that will have kind of a distinct impact on both architectural decisions that NVIDIA tries to move forward with, as well as, You know, where their ecosystem tries to figure out where NVIDIA is going, where they can be aligned to them, where they can be adjacent to them, versus where they maybe become direct Competitors to NVIDIA and AI factories and so forth. (Time 0:01:44)
  • Hardware Leadership Faces Rising Competition
    • Hardware dominance is hard to sustain and competitors (TPUs, Tranium, Chinese GPUs, AMD, Intel) are emerging.
    • NVIDIA may respond by acquiring software, storage, or hosting capabilities to defend and extend its platform advantage. Transcript: Brian Gracely We’re seeing some people trying to figure out alternative hardware accelerators. So whether it’s Google TPUs or AWS with Tranium or Microsoft apparently has their own projects to spin stuff up. We know that there are some Chinese alternatives that are spinning up. AMD and Intel are still trying to play in that GPU game to a certain extent. So the hardware space, while NVIDIA still has a dominance, again, we’ve talked about this a hundred times in this show, being the dominant proprietary highest end hardware provider Is a difficult place to stay in for a very long time. And so it would appear that you would think that they’d likely be going after assets of a software stack, right? Assets to begin to make up whether they want to try and dominate more on the enterprise side of AI, whether it is on the consumer side of AI, probably not necessarily their play. We begin to see them starting to release some things around autonomous driving. They made some announcements at a consumer electronics show around that. We’ve seen some things around robotics. So it would be very, very interesting just given the vast amount of resources they have and being able to look into the future beyond chips, you know, beyond CUDA as sort of a infrastructure Layer, you know, where do they see the biggest bang for the buck? And probably most importantly, you know, where do they start to see people that they’re customers today trying to commoditize the stack in order to move it up higher, as we’ve seen, You know, over decades with different trends, you know, where software begins to start to do that. So it’ll be very interesting to see if we see them moving into a set of sort of, you know, agentic tools, if everything they’re doing becomes part of the, you know, the NVIDIA AI factory, If they start to, you know, aggressively start to try and find some of the leaders in whether it’s agentic or rag services, or, you know, if they start to try and get into data services Somehow, you know, there’s a lot of different places they could go to fill out the stack to try and do that. Do they take their software and begin to open source it in order to prevent, become sort of the de facto standard and prevent competitors from getting up in the software stack? Do we see them doubling back down and going back into some of the hosting services, right? Today, they’re sort of the arms dealer for those hosting services, but do they try and become more dominant in that? Or do they feel like they’ve got that market segment pretty well wrapped up at this point? Networking is obviously doing very well for them, but they’re not necessarily like in the storage space. Maybe they make an acquisition in that space. So a lot of places for them to go. (Time 0:02:34)
  • Anthropic’s Broader Agentic Vision
    • Anthropic aims to expand agentic systems beyond developer tooling into collaborative, domain-specific workflows.
    • Their vision will shape economics and team composition as humans and agents hybridize work processes. Transcript: Brian Gracely Second room is, and I don’t know if this would be a room necessarily, but it would be really interesting to sort of see what Anthropik’s broader agentic vision looks like, right? Somewhere, I imagine Dario or, you know, a team of folks over at Anthropic have a pretty big view of where they see the future of Claude, of Claude co-work. And, you know, we’re beginning to see more and more things that, you know, Anthropic is talking more and more about how the tools that even they’re putting out have been built by Agentic Tools. But I think, you know, really getting a sense of how do they expect to see Agentic expand beyond sort of the, you know, the coder who says, hey, I run 10 instances of something, right? Where do they expect to see this go in terms of, you know, orchestration of these things? Where do we expect to see them go in terms of being able to collaborate across teams, share learnings within agents, you know, all of those sorts of things. You know, where do we see that in a year from now, two years from now, three years from now? What’s the bigger vision of where they see all this going? You know, I think that would be, I think it would be really, really interesting because it’s going to very much be a big piece of the puzzle as Claude has begun to take a very dominant position In the software developer space. You know, did they move downstream into sort of the no coding type of tools, right? That’s sort of where Cloud Cowork is beginning to go. But, you know, do we see that, you know, start to expand out into spaces? Do they start creating entirely different paradigms of how, you know, vertical specific, you know, lawyers think about how to solve a case or financial teams think about how to build Tooling to help them better predict markets or do stress testing or, you know, do rapid closing of the books at the end of the month, you know, all those sort of, you know, vertical niche Types of things that are sort of technology oriented. They’re not necessarily, you wouldn’t necessarily think of them as like developer oriented, but, you know, there are plenty of, you know, sort of spreadsheet experts within all those Sort of different jobs that could transfer over into something that becomes agentic. Do we see a different type of agentic happen in marketing campaigns? We’re already beginning to see that around advertising, but do we see it in terms of, you know, market understanding, competitive intelligence, all those sorts of things. Very, very fascinating to see where that goes, because the better sense you have of what that looks like, the better sense you have of what are the future economics going to look like Around not only software developers, but sort of software development teams that are hybrid of humans and agents. What does the optimal number of agents or ongoing projects per human software developer look like? What do we start to see that look like? What tools are they going to put in place to allow that to become more optimized, more flexible, allow more experimentation to happen? So I think that would be a fascinating room to be in. (Time 0:05:30)
  • Watch Agent Orchestration And Pricing Signals
    • Observe how Anthropic plans orchestration, collaboration, and shared learning between agents to forecast developer productivity changes.
    • Track experiments and pricing conversations to anticipate economics for hybrid human-agent teams. Transcript: Brian Gracely Where do they expect to see this go in terms of, you know, orchestration of these things? Where do we expect to see them go in terms of being able to collaborate across teams, share learnings within agents, you know, all of those sorts of things. You know, where do we see that in a year from now, two years from now, three years from now? What’s the bigger vision of where they see all this going? You know, I think that would be, I think it would be really, really interesting because it’s going to very much be a big piece of the puzzle as Claude has begun to take a very dominant position In the software developer space. You know, did they move downstream into sort of the no coding type of tools, right? That’s sort of where Cloud Cowork is beginning to go. But, you know, do we see that, you know, start to expand out into spaces? Do they start creating entirely different paradigms of how, you know, vertical specific, you know, lawyers think about how to solve a case or financial teams think about how to build Tooling to help them better predict markets or do stress testing or, you know, do rapid closing of the books at the end of the month, you know, all those sort of, you know, vertical niche Types of things that are sort of technology oriented. They’re not necessarily, you wouldn’t necessarily think of them as like developer oriented, but, you know, there are plenty of, you know, sort of spreadsheet experts within all those Sort of different jobs that could transfer over into something that becomes agentic. Do we see a different type of agentic happen in marketing campaigns? We’re already beginning to see that around advertising, but do we see it in terms of, you know, market understanding, competitive intelligence, all those sorts of things. Very, very fascinating to see where that goes, because the better sense you have of what that looks like, the better sense you have of what are the future economics going to look like Around not only software developers, but sort of software development teams that are hybrid of humans and agents. What does the optimal number of agents or ongoing projects per human software developer look like? What do we start to see that look like? What tools are they going to put in place to allow that to become more optimized, more flexible, allow more experimentation to happen? (Time 0:06:12)
  • Agents Will Reshape Vertical Workflows
    • Anthropic’s agentic direction will affect verticals like law, finance, marketing, and scientific problem-solving.
    • Understanding their path clarifies future labor economics and specialized agent value. Transcript: Brian Gracely It would be a fascinating room to, you know, sort of see not only what their immediate roadmap looks like, but, you know, what are some of the experiments they’re thinking about over The next year or two years, right? What are they doing to bring in industry-specific experts to sort of help them in that domain? What kind of conversations are they having with early customers about what they think the pricing might look like for these services as they evolve over time, you know, as they become, You know, potentially more expert or more able to work for long periods of time on more complex problems. You know, what does it look like for a, you know, sort of a swarm team trying to solve biological problems or chemistry problems or, you know, physics problems or things along those lines. So I think that would be a fascinating, fascinating place to be sort of a fly on the wall. And then my third one for this Q1 kind of falls in line a little bit with the CapEx discussion we talked about with NVIDIAs. But I think the third, you know, most fascinating place would be the TSMC sort of 2028 to 2030 planning room. You know, in their space, you know, they just announced they are basically doubling, you know, their CapEx for, you know, the next couple of years or, you know, for the next year. But obviously they have to think in three and five year increments in terms of building capacity, building space, being able to retrofit to new processor levels in terms of sizing and So forth. So it would be fascinating to sort of get a sense from them what sort of demand are they forecasting? Where do they see potential bumps in their forecast? Where do they see potential changes happening to the technology landscape that might create breakthroughs that are, while they’re, you know, they’re not necessarily going to be The designer of that chipset, but where do they see across their design partners a potentially new innovation happening that could either significantly reduce cost of the accelerator Chipsets or associated chipsets? Where do they see potential yield growth going up and so forth? Are they seeing any bottlenecks that they’re running into that might, you know, restrict the ability or create situations in which, you know, the current sort of forecast for growth Is going to run into bumps. You know, for example, we’ve seen some things where, you know, we’ve seen certain failures for certain chipsets on training, you know, maybe more so than others. And that’s caused some delays in terms of some of the frontier models coming out. So, you know, it’d be very, very interesting to sort of see where they see the broader market, where they’re seeing forecasts for things, where they’re seeing, you know, so both capacity Forecasts, how they’re tracking, you know, usage forecasts to make sure that they’re aligned to what the, you know, their customers are asking for. (Time 0:08:18)
  • TSMC’s Planning Drives AI Capacity
    • TSMC’s multi-year capacity planning (2028–2030) is central to AI hardware supply and industry forecasts.
    • Their capex, yield, and material constraints will determine whether demand forecasts hit or encounter bottlenecks. Transcript: Brian Gracely But I think the third, you know, most fascinating place would be the TSMC sort of 2028 to 2030 planning room. You know, in their space, you know, they just announced they are basically doubling, you know, their CapEx for, you know, the next couple of years or, you know, for the next year. But obviously they have to think in three and five year increments in terms of building capacity, building space, being able to retrofit to new processor levels in terms of sizing and So forth. So it would be fascinating to sort of get a sense from them what sort of demand are they forecasting? Where do they see potential bumps in their forecast? Where do they see potential changes happening to the technology landscape that might create breakthroughs that are, while they’re, you know, they’re not necessarily going to be The designer of that chipset, but where do they see across their design partners a potentially new innovation happening that could either significantly reduce cost of the accelerator Chipsets or associated chipsets? Where do they see potential yield growth going up and so forth? Are they seeing any bottlenecks that they’re running into that might, you know, restrict the ability or create situations in which, you know, the current sort of forecast for growth Is going to run into bumps. You know, for example, we’ve seen some things where, you know, we’ve seen certain failures for certain chipsets on training, you know, maybe more so than others. And that’s caused some delays in terms of some of the frontier models coming out. So, you know, it’d be very, very interesting to sort of see where they see the broader market, where they’re seeing forecasts for things, where they’re seeing, you know, so both capacity Forecasts, how they’re tracking, you know, usage forecasts to make sure that they’re aligned to what the, you know, their customers are asking for. But it would be fascinating to sort of look at all of that, not to mention, you know, how they are hedging, you know, with the challenges going on with rare metals in the marketplace, you Know, the sort of spike in certain commodity prices that have been happening that could impact their thing. So it would be very, very interesting to look at, you know, just kind of the breadth of all of the variables that they’re trying to take into consideration over the next three, four, five Years. (Time 0:09:11)
  • Track Three Core Players For Market Signals
    • Monitor NVIDIA, Anthropic, and TSMC as a trio to understand AI architecture, developer productivity, and capacity constraints.
    • Use their signals to adapt product strategy and partnerships in the evolving AI market. Transcript: Brian Gracely But yeah, I think if we look at the three of those, kind of, you know, where do we see NVIDIA trying to spread out if they’re trying to spread out and move into adjacent markets? Where do we see Anthropic really trying to completely redefine what not only software development looks like, but sort of work, you know, the expansion and productivity of work around Agents. And then finally, you know, what TSMC is planning is sort of the core, most important central piece in terms of, you know, all this capacity that, you know, has been forecasted and demanded And to a certain extent, you know, is being, you know, overwhelming the market in terms of that. So those are my three for Q1. I’ll probably come back and do this as a regular thing. Maybe every quarter we’ll look at, you know, three rooms that we would love to be in and three kind of unknowns that, you know, we would love to get a little bit more clarity on. So with that, I’m going to wrap it up. Thank you all for listening. Thanks for telling a friend and we will be back and we will talk to you next week. Thank you for listening to The Cloudcast. (Time 0:11:47)