Apple's New Mac Line is Built Around Local AI. The Bet Is You'd Rather Own Than Rent.

Apple just refreshed the entire desktop Apple just refreshed the entire desktop Mac line around local AI. The machines Mac line around local AI. The machines Mac line around local AI. The machines began arriving September 22nd, although began arriving September 22nd, although began arriving September 22nd, although the headline 512 gigabyte configuration the headline 512 gigabyte configuration the headline 512 gigabyte configuration does not arrive until late October. Very does not arrive until late October. Very does not arrive until late October. Very Apple. At first glance, this looks like Apple. At first glance, this looks like Apple. At first glance, this looks like Apple challenging Nvidia for AI compute. Apple challenging Nvidia for AI compute. Apple challenging Nvidia for AI compute. Now, I know what you're thinking. 512 GB Now, I know what you're thinking. 512 GB Now, I know what you're thinking. 512 GB still will not hold every enormous still will not hold every enormous still will not hold every enormous frontier model that people may imagine frontier model that people may imagine frontier model that people may imagine that they want to run. A data center can that they want to run. A data center can that they want to run. A data center can give an agent much more compute. It can give an agent much more compute. It can give an agent much more compute. It can give it much more context, more copies give it much more context, more copies give it much more context, more copies of itself in a newer model that updates of itself in a newer model that updates of itself in a newer model that updates constantly versus the computer on your constantly versus the computer on your constantly versus the computer on your desk. The memory math videos that I've desk. The memory math videos that I've desk. The memory math videos that I've seen coming out are absolutely right seen coming out are absolutely right seen coming out are absolutely right about that. Look at the strange wrinkle about that. Look at the strange wrinkle about that. Look at the strange wrinkle here. Apple put the new M6 generation at here. Apple put the new M6 generation at here. Apple put the new M6 generation at the bottom of the desktop line while the the bottom of the desktop line while the the bottom of the desktop line while the more powerful Mini and Studio are still more powerful Mini and Studio are still more powerful Mini and Studio are still on the M5. Why did they do that? What on the M5. Why did they do that? What on the M5. Why did they do that? What does that tell us about AI and where does that tell us about AI and where does that tell us about AI and where Apple thinks it's going? There's no M6 Apple thinks it's going? There's no M6 Apple thinks it's going? There's no M6 Pro here, right? There's no Max or ultra Pro here, right? There's no Max or ultra Pro here, right? There's no Max or ultra that appears with this launch. I think that appears with this launch. I think that appears with this launch. I think that Apple shipped the memory that that Apple shipped the memory that that Apple shipped the memory that people can use now because memory is so people can use now because memory is so people can use now because memory is so short instead of waiting for the chip short instead of waiting for the chip short instead of waiting for the chip family to line up as neatly as it family to line up as neatly as it family to line up as neatly as it usually does. Apple doesn't have to own usually does. Apple doesn't have to own usually does. Apple doesn't have to own the moving frontier to make money when the moving frontier to make money when the moving frontier to make money when frontier capability is this urgent. We frontier capability is this urgent. We frontier capability is this urgent. We should read the chipsets as Apple saying should read the chipsets as Apple saying should read the chipsets as Apple saying everyone is desperate for local compute.

everyone is desperate for local compute. everyone is desperate for local compute. We are going to provide it. people are We are going to provide it. people are We are going to provide it. people are going to figure out how to get agents on going to figure out how to get agents on going to figure out how to get agents on there and increasingly frontier agent there and increasingly frontier agent there and increasingly frontier agent capability is not going to be important capability is not going to be important capability is not going to be important to ordinary people. You as an individual to ordinary people. You as an individual to ordinary people. You as an individual are going to need to decide which future are going to need to decide which future are going to need to decide which future you believe in. Do you believe in the you believe in. Do you believe in the you believe in. Do you believe in the cloud future with frontier agents where cloud future with frontier agents where cloud future with frontier agents where you're going to be able to rent that you're going to be able to rent that you're going to be able to rent that future? You're going to rent the future? You're going to rent the future? You're going to rent the intelligence. You're going to rent the intelligence. You're going to rent the intelligence. You're going to rent the work. ultimately a lot of your own work. ultimately a lot of your own work. ultimately a lot of your own memory and files and everything else is memory and files and everything else is memory and files and everything else is going to live with that frontier lab or going to live with that frontier lab or going to live with that frontier lab or do you believe in the future where a lot do you believe in the future where a lot do you believe in the future where a lot of it is going to be local and you're of it is going to be local and you're of it is going to be local and you're going to have a souped-up computer and going to have a souped-up computer and going to have a souped-up computer and you're going to have your own GPU and you're going to have your own GPU and you're going to have your own GPU and your own local models that do a lot and your own local models that do a lot and your own local models that do a lot and you'll go to the frontier labs when you you'll go to the frontier labs when you you'll go to the frontier labs when you need frontier intelligence. That is the need frontier intelligence. That is the need frontier intelligence. That is the question a lot of serious AI workers are question a lot of serious AI workers are question a lot of serious AI workers are going to face in the next 90 to 120 going to face in the next 90 to 120 going to face in the next 90 to 120 days. That's a real bet and we're going days. That's a real bet and we're going days. That's a real bet and we're going to talk about it. If you're new here, to talk about it. If you're new here, to talk about it. If you're new here, I'm Nate B. Jones. I've spent the last I'm Nate B. Jones. I've spent the last I'm Nate B. Jones. I've spent the last 20 years in tech, but I've spent the 20 years in tech, but I've spent the 20 years in tech, but I've spent the last few years focusing on helping last few years focusing on helping last few years focusing on helping leaders use AI in their businesses. I leaders use AI in their businesses. I leaders use AI in their businesses. I talk about news on here. I talk about talk about news on here. I talk about talk about news on here. I talk about news for AI, why it matters to you, and news for AI, why it matters to you, and news for AI, why it matters to you, and how you can use it to confidently build how you can use it to confidently build how you can use it to confidently build the life you want. That's what I care the life you want. That's what I care the life you want. That's what I care about. And today, that means figuring about. And today, that means figuring about. And today, that means figuring out which intelligence is worth owning out which intelligence is worth owning out which intelligence is worth owning and which is worth renting. Is renting and which is worth renting. Is renting and which is worth renting. Is renting you the entire daily computing you the entire daily computing you the entire daily computing experience. The new Mac Mini comes with experience. The new Mac Mini comes with experience. The new Mac Mini comes with either M6 or M5 Pro. Apple calls it a either M6 or M5 Pro. Apple calls it a either M6 or M5 Pro. Apple calls it a machine for running models on device and machine for running models on device and machine for running models on device and for always on deskside agentic

for always on deskside agentic for always on deskside agentic computing. Apple has figured out how to computing. Apple has figured out how to computing. Apple has figured out how to talk about agents people. The M6 version talk about agents people. The M6 version talk about agents people. The M6 version goes from 16 to 32 GB of unified memory goes from 16 to 32 GB of unified memory goes from 16 to 32 GB of unified memory while M5 Pro goes harder, right? It while M5 Pro goes harder, right? It while M5 Pro goes harder, right? It reaches 64 gigs and 307 gigs per second reaches 64 gigs and 307 gigs per second reaches 64 gigs and 307 gigs per second of memory bandwidth. The Mac Studio is of memory bandwidth. The Mac Studio is of memory bandwidth. The Mac Studio is just going to push more power up there, just going to push more power up there, just going to push more power up there, right? M5 Max reaches 128 gigs of right? M5 Max reaches 128 gigs of right? M5 Max reaches 128 gigs of memory. M5 Ultra reaches 512 gigs and memory. M5 Ultra reaches 512 gigs and memory. M5 Ultra reaches 512 gigs and 1.2 terabytes per second of bandwidth. 1.2 terabytes per second of bandwidth. 1.2 terabytes per second of bandwidth. Apple says this is enough to run really Apple says this is enough to run really Apple says this is enough to run really big language models entirely on device big language models entirely on device big language models entirely on device with obviously no token meter, no cost with obviously no token meter, no cost with obviously no token meter, no cost per token, no cloud bills, just the cost per token, no cloud bills, just the cost per token, no cloud bills, just the cost of electricity. The machines are going of electricity. The machines are going of electricity. The machines are going to start shipping on September 22nd. Uh to start shipping on September 22nd. Uh to start shipping on September 22nd. Uh although the really monster one, the 512 although the really monster one, the 512 although the really monster one, the 512 gig one, that doesn't arrive until late gig one, that doesn't arrive until late gig one, that doesn't arrive until late October. Very typical for Apple. But October. Very typical for Apple. But October. Very typical for Apple. But these machines are arriving just as the these machines are arriving just as the these machines are arriving just as the most capable agents are moving in the most capable agents are moving in the most capable agents are moving in the other direction onto persistent other direction onto persistent other direction onto persistent computers in the cloud. So there's a computers in the cloud. So there's a computers in the cloud. So there's a larger story and a larger question here. larger story and a larger question here. larger story and a larger question here. Why is Apple making its largest local AI Why is Apple making its largest local AI Why is Apple making its largest local AI bet? Now I think the answer explains why bet? Now I think the answer explains why bet? Now I think the answer explains why Apple has become much smarter about AI Apple has become much smarter about AI Apple has become much smarter about AI lately and why this market is big enough lately and why this market is big enough lately and why this market is big enough for multiple winners. It's big enough for multiple winners. It's big enough for multiple winners. It's big enough for Apple to win, for Jensen and Nvidia for Apple to win, for Jensen and Nvidia for Apple to win, for Jensen and Nvidia to win, and it's even big enough for to win, and it's even big enough for to win, and it's even big enough for OpenAI to win in that space at the same OpenAI to win in that space at the same OpenAI to win in that space at the same time. That doesn't mean that everyone's time. That doesn't mean that everyone's time. That doesn't mean that everyone's going to win in the same way. And we're going to win in the same way. And we're going to win in the same way. And we're going to get into that. One thing that going to get into that. One thing that going to get into that. One thing that popped out in this whole Apple launch is popped out in this whole Apple launch is popped out in this whole Apple launch is that Apple is not quiet about AI that Apple is not quiet about AI that Apple is not quiet about AI anymore. Apple is not just adding an a anymore. Apple is not just adding an a anymore. Apple is not just adding an a neural engine and hoping that nobody

neural engine and hoping that nobody neural engine and hoping that nobody notices. Apple is explicitly selling the notices. Apple is explicitly selling the notices. Apple is explicitly selling the Mac as the computer where an AI agent Mac as the computer where an AI agent Mac as the computer where an AI agent can live. Like this is much more can live. Like this is much more can live. Like this is much more explicitly AI native AI local than Apple explicitly AI native AI local than Apple explicitly AI native AI local than Apple has ever been before. I look I've been has ever been before. I look I've been has ever been before. I look I've been saying for months that Apple could be saying for months that Apple could be saying for months that Apple could be the company that's serious about local the company that's serious about local the company that's serious about local AI and this launch represents a real AI and this launch represents a real AI and this launch represents a real turnaround for them. This launch is turnaround for them. This launch is turnaround for them. This launch is absolutely validating where I was absolutely validating where I was absolutely validating where I was telling them that they ought to be going telling them that they ought to be going telling them that they ought to be going because people are hungry for this local because people are hungry for this local because people are hungry for this local compute and Apple Silicon is good with compute and Apple Silicon is good with compute and Apple Silicon is good with AI. The company that seems to have spent AI. The company that seems to have spent AI. The company that seems to have spent most of the AI boom so far looking very most of the AI boom so far looking very most of the AI boom so far looking very slow and behind now has a complete slow and behind now has a complete slow and behind now has a complete desktop line built around the idea that desktop line built around the idea that desktop line built around the idea that a meaningful number of people are going a meaningful number of people are going a meaningful number of people are going to pay a meaningful amount of money to to pay a meaningful amount of money to to pay a meaningful amount of money to own the intelligence they can use. The own the intelligence they can use. The own the intelligence they can use. The immediate objection here is very immediate objection here is very immediate objection here is very obvious. You can pay a lot of money, obvious. You can pay a lot of money, obvious. You can pay a lot of money, like enough money for a small car, and like enough money for a small car, and like enough money for a small car, and you get 512 gigs, and that will still you get 512 gigs, and that will still you get 512 gigs, and that will still not be enough to hold every enormous not be enough to hold every enormous not be enough to hold every enormous Frontier model that we could imagine we Frontier model that we could imagine we Frontier model that we could imagine we want to run, even if it was open want to run, even if it was open want to run, even if it was open weighted, which of course not all of weighted, which of course not all of weighted, which of course not all of them are. A data center might give an them are. A data center might give an them are. A data center might give an agent all the compute it needs, right?

agent all the compute it needs, right? agent all the compute it needs, right? All the context, copies of itself, it All the context, copies of itself, it All the context, copies of itself, it updates for you. A newer model uh that updates for you. A newer model uh that updates for you. A newer model uh that updates all the time when the next updates all the time when the next updates all the time when the next release drops. All of the people who are release drops. All of the people who are release drops. All of the people who are making those memory math videos talking making those memory math videos talking making those memory math videos talking about the limits of the memory that about the limits of the memory that about the limits of the memory that Apple has released are right about that. Apple has released are right about that. Apple has released are right about that. But I think that they're asking a But I think that they're asking a But I think that they're asking a different question from the one Apple is different question from the one Apple is different question from the one Apple is asking to win. Apple doesn't need the asking to win. Apple doesn't need the asking to win. Apple doesn't need the Mac to hold the most powerful model in Mac to hold the most powerful model in Mac to hold the most powerful model in the universe forever. It just needs the universe forever. It just needs the universe forever. It just needs local models to be good enough for a local models to be good enough for a local models to be good enough for a huge share of the work that people huge share of the work that people huge share of the work that people actually do. and it needs enough actually do. and it needs enough actually do. and it needs enough customers to prefer paying once for the customers to prefer paying once for the customers to prefer paying once for the computer and then paying for electricity computer and then paying for electricity computer and then paying for electricity instead of paying an unknown amount for instead of paying an unknown amount for instead of paying an unknown amount for tokens every single month. That is a tokens every single month. That is a tokens every single month. That is a much more interesting bet to me because much more interesting bet to me because much more interesting bet to me because it places Apple against two trends that it places Apple against two trends that it places Apple against two trends that are moving in opposite directions. We're are moving in opposite directions. We're are moving in opposite directions. We're going to talk about both of them. First, going to talk about both of them. First, going to talk about both of them. First, frontier agents are consuming more frontier agents are consuming more frontier agents are consuming more compute and moving into the cloud. At compute and moving into the cloud. At compute and moving into the cloud. At the same time, useful intelligence is the same time, useful intelligence is the same time, useful intelligence is getting smaller, cheaper, and easier to getting smaller, cheaper, and easier to getting smaller, cheaper, and easier to run locally because we continue to see run locally because we continue to see run locally because we continue to see this steady progression of intelligence this steady progression of intelligence this steady progression of intelligence down the cost curve. Smarter tokens keep down the cost curve. Smarter tokens keep down the cost curve. Smarter tokens keep getting cheaper and it's happening getting cheaper and it's happening getting cheaper and it's happening really fast. What will decide this really fast. What will decide this really fast. What will decide this market is how much valuable work falls market is how much valuable work falls market is how much valuable work falls on each side of the local compute chasm on each side of the local compute chasm on each side of the local compute chasm and who makes it easy to do that work.

and who makes it easy to do that work. and who makes it easy to do that work. But before we go there, I want you to But before we go there, I want you to But before we go there, I want you to look at the launch first as a product look at the launch first as a product look at the launch first as a product line. Let's not let's not get too stuck line. Let's not let's not get too stuck line. Let's not let's not get too stuck on the giant memory numbers right now. on the giant memory numbers right now. on the giant memory numbers right now. The base M6 Mac Mini is the little The base M6 Mac Mini is the little The base M6 Mac Mini is the little always on machine. The thing you're always on machine. The thing you're always on machine. The thing you're going to see stacks of that run going to see stacks of that run going to see stacks of that run openclaw, right? It can sit on the desk. openclaw, right? It can sit on the desk. openclaw, right? It can sit on the desk. It can run ordinary applications. It can It can run ordinary applications. It can It can run ordinary applications. It can keep a modest local model like Gemma keep a modest local model like Gemma keep a modest local model like Gemma going. It can keep an Asian alive. Your going. It can keep an Asian alive. Your going. It can keep an Asian alive. Your Hermes can live there. Now, the M5 Pro Hermes can live there. Now, the M5 Pro Hermes can live there. Now, the M5 Pro Mini doubles that top memory to 64 gigs Mini doubles that top memory to 64 gigs Mini doubles that top memory to 64 gigs and nearly doubles the bandwidth as and nearly doubles the bandwidth as and nearly doubles the bandwidth as well. In other words, you're going to well. In other words, you're going to well. In other words, you're going to have more working memory and it's going have more working memory and it's going have more working memory and it's going to run faster. It's a super claw, right? to run faster. It's a super claw, right? to run faster. It's a super claw, right? The M5 Max Studio bigger, doubles memory The M5 Max Studio bigger, doubles memory The M5 Max Studio bigger, doubles memory again, and M5 Ultra offers 256 before again, and M5 Ultra offers 256 before again, and M5 Ultra offers 256 before doubling again at the very top of the doubling again at the very top of the doubling again at the very top of the range to 512. Ultimately, across this range to 512. Ultimately, across this range to 512. Ultimately, across this entire announce line, what Apple is entire announce line, what Apple is entire announce line, what Apple is doing is they're giving buyers a ladder doing is they're giving buyers a ladder doing is they're giving buyers a ladder from an ordinary local assistant, like I from an ordinary local assistant, like I from an ordinary local assistant, like I said, an ordinary local open claw to said, an ordinary local open claw to said, an ordinary local open claw to something that could run a fairly big something that could run a fairly big something that could run a fairly big model in a studio environment locally model in a studio environment locally model in a studio environment locally across long contexts and have several across long contexts and have several across long contexts and have several models working at once. I know people models working at once. I know people models working at once. I know people who are excited for these because they who are excited for these because they who are excited for these because they aren't just stuck with one model aren't just stuck with one model aren't just stuck with one model anymore. They can have a core coding anymore. They can have a core coding anymore. They can have a core coding model, a separate reviewer agent, and an model, a separate reviewer agent, and an model, a separate reviewer agent, and an open claw all running on the same open claw all running on the same open claw all running on the same machine. That's possible now, right? The machine. That's possible now, right? The machine. That's possible now, right? The Mac Studio with M5 Max starts at 2500.

Mac Studio with M5 Max starts at 2500. Mac Studio with M5 Max starts at 2500. M5 Ultra starts at 5500 before the buyer M5 Ultra starts at 5500 before the buyer M5 Ultra starts at 5500 before the buyer adds even more expensive memory and adds even more expensive memory and adds even more expensive memory and storage options. That's where it gets storage options. That's where it gets storage options. That's where it gets really pricey. It gets into car really pricey. It gets into car really pricey. It gets into car territory. Apple has not priced local AI territory. Apple has not priced local AI territory. Apple has not priced local AI cheaply, and Apple wouldn't, right? They cheaply, and Apple wouldn't, right? They cheaply, and Apple wouldn't, right? They never do. They always have protected never do. They always have protected never do. They always have protected margins and priced up. Now, there is a margins and priced up. Now, there is a margins and priced up. Now, there is a strange wrinkle in the chipsets and I strange wrinkle in the chipsets and I strange wrinkle in the chipsets and I think it tells us something about where think it tells us something about where think it tells us something about where Apple's head is at on AI. Apple decided Apple's head is at on AI. Apple decided Apple's head is at on AI. Apple decided to put their new M6 generation at the to put their new M6 generation at the to put their new M6 generation at the very bottom of the line. While the very bottom of the line. While the very bottom of the line. While the powerful mini and studio are still on powerful mini and studio are still on powerful mini and studio are still on the M5 chipsets on M5 Pro, M5 Max, and the M5 chipsets on M5 Pro, M5 Max, and the M5 chipsets on M5 Pro, M5 Max, and M5 Ultra, there is no M6 Pro Max or M5 Ultra, there is no M6 Pro Max or M5 Ultra, there is no M6 Pro Max or Ultra yet. And this doesn't weaken their Ultra yet. And this doesn't weaken their Ultra yet. And this doesn't weaken their strategy. Instead, it tells me about strategy. Instead, it tells me about strategy. Instead, it tells me about Apple's urgency. And the thing that Apple's urgency. And the thing that Apple's urgency. And the thing that Apple would want to remind everyone is Apple would want to remind everyone is Apple would want to remind everyone is that this is not just the headline that this is not just the headline that this is not just the headline numbers. This is a Mac. This is a Mac numbers. This is a Mac. This is a Mac numbers. This is a Mac. This is a Mac and that means an operating system and and that means an operating system and and that means an operating system and applications and browser and the whole applications and browser and the whole applications and browser and the whole walt garden experience. It is also a Mac walt garden experience. It is also a Mac walt garden experience. It is also a Mac and we're going to remember this in a and we're going to remember this in a and we're going to remember this in a couple weeks because it connects to couple weeks because it connects to couple weeks because it connects to iPhone and iPad and watch and the rest iPhone and iPad and watch and the rest iPhone and iPad and watch and the rest of the Apple ecosystem. You just wait of the Apple ecosystem. You just wait of the Apple ecosystem. You just wait until the Apple phone refresh in a until the Apple phone refresh in a until the Apple phone refresh in a couple of weeks. Now, Apple's not alone couple of weeks. Now, Apple's not alone couple of weeks. Now, Apple's not alone here. Anyone worth their salt on models here. Anyone worth their salt on models here. Anyone worth their salt on models is going to talk to you about Nvidia's is going to talk to you about Nvidia's is going to talk to you about Nvidia's DGX Spark. It is a great complete DGX Spark. It is a great complete DGX Spark. It is a great complete computer as well. It is more directly computer as well. It is more directly computer as well. It is more directly optimized for the person who just wants optimized for the person who just wants optimized for the person who just wants a local AI development appliance cuz a local AI development appliance cuz a local AI development appliance cuz that's all it's designed to do. It's that's all it's designed to do. It's that's all it's designed to do. It's connected to uh the CUDA firmware connected to uh the CUDA firmware connected to uh the CUDA firmware ecosystem. It has the NVIDIA data center ecosystem. It has the NVIDIA data center ecosystem. It has the NVIDIA data center stack inside. But I find that most

stack inside. But I find that most stack inside. But I find that most people don't want to just buy a separate people don't want to just buy a separate people don't want to just buy a separate AI appliance and a computer. They want AI appliance and a computer. They want AI appliance and a computer. They want the computer to be the AI machine. That the computer to be the AI machine. That the computer to be the AI machine. That is Apple's bet here. Apple can sell the is Apple's bet here. Apple can sell the is Apple's bet here. Apple can sell the box that everybody already understands box that everybody already understands box that everybody already understands with local inference for your LLMs as with local inference for your LLMs as with local inference for your LLMs as another big reason to buy memory and another big reason to buy memory and another big reason to buy memory and move up the product line. The barrier move up the product line. The barrier move up the product line. The barrier stopping people from setting up that stopping people from setting up that stopping people from setting up that capability is partly technical and capability is partly technical and capability is partly technical and partly compute. Apple with this launch partly compute. Apple with this launch partly compute. Apple with this launch is solving the compute side in numbers is solving the compute side in numbers is solving the compute side in numbers that everybody has been told to that everybody has been told to that everybody has been told to understand for decades. They're talking understand for decades. They're talking understand for decades. They're talking about gigabytes and people understand about gigabytes and people understand about gigabytes and people understand that they can pick the class of work and that they can pick the class of work and that they can pick the class of work and you can expect a certain number of you can expect a certain number of you can expect a certain number of agents to run. Uh and then you just buy agents to run. Uh and then you just buy agents to run. Uh and then you just buy the machine that works for you, right? the machine that works for you, right? the machine that works for you, right? And and I would say like if you're if And and I would say like if you're if And and I would say like if you're if you're talking to me and I'm looking at you're talking to me and I'm looking at you're talking to me and I'm looking at what I'm buying, I'm going for the 128. what I'm buying, I'm going for the 128. what I'm buying, I'm going for the 128. The 128 seems to be at the point where I The 128 seems to be at the point where I The 128 seems to be at the point where I can run two or three agents. I can run a can run two or three agents. I can run a can run two or three agents. I can run a uh significantly sized local model uh significantly sized local model uh significantly sized local model that's my main agent. I can run two or that's my main agent. I can run two or that's my main agent. I can run two or three on the side. It's going to be three on the side. It's going to be three on the side. It's going to be fine. 512 I could do, but it feels like fine. 512 I could do, but it feels like fine. 512 I could do, but it feels like it's a lot of muscle and I would feel it's a lot of muscle and I would feel it's a lot of muscle and I would feel like I would want to be maximizing the like I would want to be maximizing the like I would want to be maximizing the value of that all the time and I use value of that all the time and I use value of that all the time and I use enough Frontier agents that maybe I enough Frontier agents that maybe I enough Frontier agents that maybe I don't do that. So, everyone's going to don't do that. So, everyone's going to don't do that. So, everyone's going to be different, but that's an example of be different, but that's an example of be different, but that's an example of how people are going to think about the how people are going to think about the how people are going to think about the economics of buying those machines. But economics of buying those machines. But economics of buying those machines. But there's a second issue, the technical there's a second issue, the technical there's a second issue, the technical setup issue, and that is where Apple has setup issue, and that is where Apple has setup issue, and that is where Apple has a question mark. One of the challenges a question mark. One of the challenges a question mark. One of the challenges of setting up local compute and local of setting up local compute and local of setting up local compute and local models is that there is not currently a

models is that there is not currently a models is that there is not currently a super seamless way to route to the model super seamless way to route to the model super seamless way to route to the model that you need to have whether it's on that you need to have whether it's on that you need to have whether it's on your laptop or in the cloud. There there your laptop or in the cloud. There there your laptop or in the cloud. There there just isn't. Uh and and there's no just isn't. Uh and and there's no just isn't. Uh and and there's no incentive for the Frontier models to do incentive for the Frontier models to do incentive for the Frontier models to do that at the moment. Uh there's incentive that at the moment. Uh there's incentive that at the moment. Uh there's incentive for Apple to make that happen because it for Apple to make that happen because it for Apple to make that happen because it sells more of their silicon, but Apple sells more of their silicon, but Apple sells more of their silicon, but Apple isn't a model company. Ironically, the isn't a model company. Ironically, the isn't a model company. Ironically, the company probably best positioned to do company probably best positioned to do company probably best positioned to do that right now is HuggingFace, which was that right now is HuggingFace, which was that right now is HuggingFace, which was bought by Jensen Huang just a couple of bought by Jensen Huang just a couple of bought by Jensen Huang just a couple of days after Apple announced their lineup. days after Apple announced their lineup. days after Apple announced their lineup. Yes, those things lined up and I don't Yes, those things lined up and I don't Yes, those things lined up and I don't think it was an accident. Jensen has think it was an accident. Jensen has think it was an accident. Jensen has been looking to facilitate the open been looking to facilitate the open been looking to facilitate the open source ecosystem with Nvidia for a source ecosystem with Nvidia for a source ecosystem with Nvidia for a while. Uh the DGX Spark is a part of while. Uh the DGX Spark is a part of while. Uh the DGX Spark is a part of that, but so are their line of their own that, but so are their line of their own that, but so are their line of their own open source models with Neotron. Uh and open source models with Neotron. Uh and open source models with Neotron. Uh and they're good models, right? They are they're good models, right? They are they're good models, right? They are looking to facilitate more open source looking to facilitate more open source looking to facilitate more open source by purchasing hugging face. Hugging face by purchasing hugging face. Hugging face by purchasing hugging face. Hugging face is the place people with all these Macs is the place people with all these Macs is the place people with all these Macs go to get their models. Ironically, go to get their models. Ironically, go to get their models. Ironically, HuggingFace is best positioned to solve HuggingFace is best positioned to solve HuggingFace is best positioned to solve the local install problem for Apple.

the local install problem for Apple. the local install problem for Apple. That is one of the most interesting open That is one of the most interesting open That is one of the most interesting open questions now is how HuggingFace or questions now is how HuggingFace or questions now is how HuggingFace or someone else chooses to address the someone else chooses to address the someone else chooses to address the local model issue. But local models and local model issue. But local models and local model issue. But local models and local compute aren't the only bet out local compute aren't the only bet out local compute aren't the only bet out there. The economics are super different there. The economics are super different there. The economics are super different when it comes to cloud. And I want to when it comes to cloud. And I want to when it comes to cloud. And I want to talk about cloud as a separate bet talk about cloud as a separate bet talk about cloud as a separate bet because cloud is also a way to address because cloud is also a way to address because cloud is also a way to address exactly some of the weaknesses that exactly some of the weaknesses that exactly some of the weaknesses that we've been talking about here. The the we've been talking about here. The the we've been talking about here. The the simplest way to talk about the benefit simplest way to talk about the benefit simplest way to talk about the benefit of cloud is that cloud is for people of cloud is that cloud is for people of cloud is that cloud is for people that don't want to think about the that don't want to think about the that don't want to think about the model. They just want to turn on their model. They just want to turn on their model. They just want to turn on their application and it just needs to work application and it just needs to work application and it just needs to work regardless of their computer. Most regardless of their computer. Most regardless of their computer. Most people, the bet goes, will want to use people, the bet goes, will want to use people, the bet goes, will want to use the best agent they can get. They will the best agent they can get. They will the best agent they can get. They will be loyal to a particular lab like OpenAI be loyal to a particular lab like OpenAI be loyal to a particular lab like OpenAI or Anthropic or XAR or whoever earns or Anthropic or XAR or whoever earns or Anthropic or XAR or whoever earns their relationship. They will care about their relationship. They will care about their relationship. They will care about that work and they will not care about that work and they will not care about that work and they will not care about the computer that performed it or even the computer that performed it or even the computer that performed it or even the individual model. They just care the individual model. They just care the individual model. They just care that it got done. I would argue that that it got done. I would argue that that it got done. I would argue that there is room for multiple winners in there is room for multiple winners in there is room for multiple winners in this market. My hunch is that if you this market. My hunch is that if you this market. My hunch is that if you were still looking at a significant were still looking at a significant were still looking at a significant technical barrier to install local technical barrier to install local technical barrier to install local models, if we don't have a simple models, if we don't have a simple models, if we don't have a simple Applelike experience for getting local Applelike experience for getting local Applelike experience for getting local compute onto the laptop, then you are compute onto the laptop, then you are compute onto the laptop, then you are still looking at a 5 to 10% of the still looking at a 5 to 10% of the still looking at a 5 to 10% of the market that will invest in Apple silicon market that will invest in Apple silicon market that will invest in Apple silicon in order to get local models because in order to get local models because in order to get local models because they're not afraid of the technical they're not afraid of the technical they're not afraid of the technical complexity. and like 90% of the market complexity. and like 90% of the market complexity. and like 90% of the market is going to be like whatever I can get is going to be like whatever I can get is going to be like whatever I can get from from a lab from open AI from from from a lab from open AI from from from a lab from open AI from anthropic is going to be good enough but anthropic is going to be good enough but anthropic is going to be good enough but that doesn't mean Apple loses. The Mac

that doesn't mean Apple loses. The Mac that doesn't mean Apple loses. The Mac generated more than 10 billion in generated more than 10 billion in generated more than 10 billion in Apple's last reported quarter and Apple's last reported quarter and Apple's last reported quarter and Apple's product gross margin was roughly Apple's product gross margin was roughly Apple's product gross margin was roughly 40%. Apple doesn't need to win the 40%. Apple doesn't need to win the 40%. Apple doesn't need to win the global GPU buildout to grow that slice. global GPU buildout to grow that slice. global GPU buildout to grow that slice. It just needs a nice profitable minority It just needs a nice profitable minority It just needs a nice profitable minority of technical proumers who decide that of technical proumers who decide that of technical proumers who decide that local memory and privacy and fixed cost local memory and privacy and fixed cost local memory and privacy and fixed cost and control are worth purchasing. There and control are worth purchasing. There and control are worth purchasing. There are lots of people who are going to pony are lots of people who are going to pony are lots of people who are going to pony up for that. And this is where up for that. And this is where up for that. And this is where complexifier appears because the cloud complexifier appears because the cloud complexifier appears because the cloud case is also something that this proumer case is also something that this proumer case is also something that this proumer class has to take seriously. we who are class has to take seriously. we who are class has to take seriously. we who are super nerds because open AI appears to super nerds because open AI appears to super nerds because open AI appears to believe and other labs appear to believe believe and other labs appear to believe believe and other labs appear to believe something very different about the something very different about the something very different about the future of personal computing. The cloud future of personal computing. The cloud future of personal computing. The cloud bet is not just a convenience play. It bet is not just a convenience play. It bet is not just a convenience play. It is a bet that says intelligence is not is a bet that says intelligence is not is a bet that says intelligence is not solved in the way I just described. In solved in the way I just described. In solved in the way I just described. In other words, there's a bet that says, other words, there's a bet that says, other words, there's a bet that says, you know, wave around Jven's paradox, you know, wave around Jven's paradox, you know, wave around Jven's paradox, whatever you want to say. There's whatever you want to say. There's whatever you want to say. There's another level of agent capability around another level of agent capability around another level of agent capability around the corner that will be too useful and the corner that will be too useful and the corner that will be too useful and too computationally expensive to ignore.

too computationally expensive to ignore. too computationally expensive to ignore. So that even someone with a fancy local So that even someone with a fancy local So that even someone with a fancy local machine will still want to rent that machine will still want to rent that machine will still want to rent that agent on a server in the valley agent on a server in the valley agent on a server in the valley somewhere because it can do work that somewhere because it can do work that somewhere because it can do work that your machine can't. I think the early your machine can't. I think the early your machine can't. I think the early product shape for this is already here. product shape for this is already here. product shape for this is already here. Grockbot gives the user a persistent Grockbot gives the user a persistent Grockbot gives the user a persistent computer in the cloud literally in the computer in the cloud literally in the computer in the cloud literally in the valley. It has a browser and a file valley. It has a browser and a file valley. It has a browser and a file system and a terminal and application system and a terminal and application system and a terminal and application login and persistence so that it keeps login and persistence so that it keeps login and persistence so that it keeps working after the laptop closes. Several working after the laptop closes. Several working after the laptop closes. Several bots can use that same environment. I bots can use that same environment. I bots can use that same environment. I have a couple of dozen going and they have a couple of dozen going and they have a couple of dozen going and they can hand work to one another. The user can hand work to one another. The user can hand work to one another. The user can sign into services on the cloud can sign into services on the cloud can sign into services on the cloud computer and then the agent can return computer and then the agent can return computer and then the agent can return to those sessions later. Chad GPT work to those sessions later. Chad GPT work to those sessions later. Chad GPT work also has its own cloud browser now as also has its own cloud browser now as also has its own cloud browser now as well. It can ask the user to sign in well. It can ask the user to sign in well. It can ask the user to sign in through a secure form and keep the through a secure form and keep the through a secure form and keep the authenticated session with a token and authenticated session with a token and authenticated session with a token and continue working after the phone or the continue working after the phone or the continue working after the phone or the laptop closes and come back when it laptop closes and come back when it laptop closes and come back when it needs another decision. So the model needs another decision. So the model needs another decision. So the model doesn't need the user's computer to stay doesn't need the user's computer to stay doesn't need the user's computer to stay awake because the computer doing the awake because the computer doing the awake because the computer doing the work belongs to, you know, XAI or open work belongs to, you know, XAI or open work belongs to, you know, XAI or open AI. This is a lot closer to the future AI. This is a lot closer to the future AI. This is a lot closer to the future where you rent a personal supercomput where you rent a personal supercomput where you rent a personal supercomput than it is to subscribing to a chatbot.

than it is to subscribing to a chatbot. than it is to subscribing to a chatbot. It's it's why I think that the labs are It's it's why I think that the labs are It's it's why I think that the labs are very bullish on their revenue. They see very bullish on their revenue. They see very bullish on their revenue. They see an unsolved future for intelligence that an unsolved future for intelligence that an unsolved future for intelligence that is that useful that they think a is that useful that they think a is that useful that they think a significant number of people who are in significant number of people who are in significant number of people who are in that same proumer class will pay for. that same proumer class will pay for. that same proumer class will pay for. And if you do pay for that, if that's And if you do pay for that, if that's And if you do pay for that, if that's what we get by the fall, then the lab what we get by the fall, then the lab what we get by the fall, then the lab can upgrade the model, can add more can upgrade the model, can add more can upgrade the model, can add more compute for you, can run more agents in compute for you, can run more agents in compute for you, can run more agents in parallel for you, can maintain the parallel for you, can maintain the parallel for you, can maintain the entire environment for you without entire environment for you without entire environment for you without shipping another device. There's a lot shipping another device. There's a lot shipping another device. There's a lot of convenience that comes with that, of convenience that comes with that, of convenience that comes with that, right? If the next generation of agents right? If the next generation of agents right? If the next generation of agents ends up being able to operate hundreds ends up being able to operate hundreds ends up being able to operate hundreds of tools, keep enormous live context of tools, keep enormous live context of tools, keep enormous live context alive across dozens of agents, try 20 or alive across dozens of agents, try 20 or alive across dozens of agents, try 20 or 30 approaches at once, work for days or 30 approaches at once, work for days or 30 approaches at once, work for days or weeks at a time, the cloud has an weeks at a time, the cloud has an weeks at a time, the cloud has an advantage that no Mac Mini is going to advantage that no Mac Mini is going to advantage that no Mac Mini is going to erase there. There's also a business erase there. There's also a business erase there. There's also a business model advantage, right? The cloud model advantage, right? The cloud model advantage, right? The cloud provider can charge by the meter for the provider can charge by the meter for the provider can charge by the meter for the intelligence that's being used. If intelligence that's being used. If intelligence that's being used. If agents become a labor layer for a agents become a labor layer for a agents become a labor layer for a company or an individual, a recurring company or an individual, a recurring company or an individual, a recurring bill tied to that work may end up being bill tied to that work may end up being bill tied to that work may end up being far larger than the margin on the far larger than the margin on the far larger than the margin on the computer sold every few years. Now, as computer sold every few years. Now, as computer sold every few years. Now, as usual, Jensen and Nvidia benefit either usual, Jensen and Nvidia benefit either usual, Jensen and Nvidia benefit either way, right? They benefit from the way, right? They benefit from the way, right? They benefit from the compute paradigm evolving regardless of compute paradigm evolving regardless of compute paradigm evolving regardless of where the winning agent comes from.

where the winning agent comes from. where the winning agent comes from. Whether it comes from OpenAI, whether it Whether it comes from OpenAI, whether it Whether it comes from OpenAI, whether it comes from XAI, whether it comes from comes from XAI, whether it comes from comes from XAI, whether it comes from Anthropic, whether it comes from someone Anthropic, whether it comes from someone Anthropic, whether it comes from someone we haven't met yet. They even benefit we haven't met yet. They even benefit we haven't met yet. They even benefit thanks to the hugging face acquisition. thanks to the hugging face acquisition. thanks to the hugging face acquisition. if you're not even using their silicon. if you're not even using their silicon. if you're not even using their silicon. The hard question for Apple is not The hard question for Apple is not The hard question for Apple is not really whether models get larger. Then really whether models get larger. Then really whether models get larger. Then Apple can always expand and they will Apple can always expand and they will Apple can always expand and they will expand. They'll build a better Mac, a expand. They'll build a better Mac, a expand. They'll build a better Mac, a larger Mac, a better chip, and models larger Mac, a better chip, and models larger Mac, a better chip, and models will become cheaper and smarter to run will become cheaper and smarter to run will become cheaper and smarter to run over time. The danger is that the cloud over time. The danger is that the cloud over time. The danger is that the cloud agent becomes the permanent home for the agent becomes the permanent home for the agent becomes the permanent home for the user's files, their memory, their user's files, their memory, their user's files, their memory, their browser session, everything that helps browser session, everything that helps browser session, everything that helps them to get work done. Now, at that them to get work done. Now, at that them to get work done. Now, at that point, all you have for for your Mac is point, all you have for for your Mac is point, all you have for for your Mac is it's an excellent terminal. It's an it's an excellent terminal. It's an it's an excellent terminal. It's an accident window onto your agent and the accident window onto your agent and the accident window onto your agent and the agent provider is renting you the entire agent provider is renting you the entire agent provider is renting you the entire daily computing experience. Apple daily computing experience. Apple daily computing experience. Apple doesn't want that. Apple doesn't want doesn't want that. Apple doesn't want doesn't want that. Apple doesn't want you to want that. Apple wants you to you to want that. Apple wants you to you to want that. Apple wants you to care about owning your own compute, care about owning your own compute, care about owning your own compute, owning your own AI. But the larger point owning your own AI. But the larger point owning your own AI. But the larger point here is that you as an individual are here is that you as an individual are here is that you as an individual are going to need to decide which future you going to need to decide which future you going to need to decide which future you believe in. Do you believe in the cloud believe in. Do you believe in the cloud believe in. Do you believe in the cloud future with frontier agents where you're future with frontier agents where you're future with frontier agents where you're going to be able to rent that future, going to be able to rent that future, going to be able to rent that future, you're going to rent the intelligence, you're going to rent the intelligence, you're going to rent the intelligence, you're going to rent the work, you're going to rent the work, you're going to rent the work, ultimately a lot of your own memory and ultimately a lot of your own memory and ultimately a lot of your own memory and files and everything else is going to files and everything else is going to files and everything else is going to live with that frontier lab or do you live with that frontier lab or do you live with that frontier lab or do you believe in the future where a lot of it believe in the future where a lot of it believe in the future where a lot of it is going to be local and you're going to is going to be local and you're going to is going to be local and you're going to have a souped-up computer and you're have a souped-up computer and you're have a souped-up computer and you're going to have your own GPU and your own going to have your own GPU and your own going to have your own GPU and your own local models that do a lot and you'll go local models that do a lot and you'll go local models that do a lot and you'll go to the frontier labs when you need

to the frontier labs when you need to the frontier labs when you need frontier intelligence. That is the frontier intelligence. That is the frontier intelligence. That is the question a lot of serious AI workers are question a lot of serious AI workers are question a lot of serious AI workers are going to face in the next 90 to 120 going to face in the next 90 to 120 going to face in the next 90 to 120 days. And the answer is going to days. And the answer is going to days. And the answer is going to determine how Apple's launch does. It's determine how Apple's launch does. It's determine how Apple's launch does. It's going to determine how we think about going to determine how we think about going to determine how we think about OpenAI's next agents that they launch OpenAI's next agents that they launch OpenAI's next agents that they launch Claude's next agents that Anthropic Claude's next agents that Anthropic Claude's next agents that Anthropic launches the new version of Grockbot launches the new version of Grockbot launches the new version of Grockbot whenever that comes out. These are real whenever that comes out. These are real whenever that comes out. These are real questions and everyone is competing for questions and everyone is competing for questions and everyone is competing for the proumer class because the proumer the proumer class because the proumer the proumer class because the proumer class is where AI is going. they are class is where AI is going. they are class is where AI is going. they are spending and investing in their careers. spending and investing in their careers. spending and investing in their careers. We're we're and I put myself in that We're we're and I put myself in that We're we're and I put myself in that category. We are investing in our category. We are investing in our category. We are investing in our futures and everyone wants to know from futures and everyone wants to know from futures and everyone wants to know from us where are we going to invest? Are we us where are we going to invest? Are we us where are we going to invest? Are we going to invest in local AI? Are we going to invest in local AI? Are we going to invest in local AI? Are we going to invest in these Macs? From what going to invest in these Macs? From what going to invest in these Macs? From what I can tell talking to a bunch of people, I can tell talking to a bunch of people, I can tell talking to a bunch of people, there's going to be a lot of investing there's going to be a lot of investing there's going to be a lot of investing in Macs. I have not found one person in Macs. I have not found one person in Macs. I have not found one person that I would put in the proumer class that I would put in the proumer class that I would put in the proumer class who is not drooling over these Macs. And who is not drooling over these Macs. And who is not drooling over these Macs. And I include myself in that. These are I include myself in that. These are I include myself in that. These are really cool computers. They can do a really cool computers. They can do a really cool computers. They can do a lot. I do think that we are going to lot. I do think that we are going to lot. I do think that we are going to have 80 or 90% of our intelligence work have 80 or 90% of our intelligence work have 80 or 90% of our intelligence work that we can do locally and these Macs that we can do locally and these Macs that we can do locally and these Macs make that possible and easy. And this is make that possible and easy. And this is make that possible and easy. And this is absolutely Apple leaning in to that absolutely Apple leaning in to that absolutely Apple leaning in to that local compute vision that I talked about local compute vision that I talked about local compute vision that I talked about in the spring. I'm so excited that in the spring. I'm so excited that in the spring. I'm so excited that they're doing that. I think it's really they're doing that. I think it's really they're doing that. I think it's really cool. I don't think we can sleep on the cool. I don't think we can sleep on the cool. I don't think we can sleep on the possibility that the labs are going to possibility that the labs are going to possibility that the labs are going to release something extraordinarily release something extraordinarily release something extraordinarily useful. And so my bet is we are going to useful. And so my bet is we are going to useful. And so my bet is we are going to end up in a bothand scenario where we

end up in a bothand scenario where we end up in a bothand scenario where we are investing in local compute. We are are investing in local compute. We are are investing in local compute. We are investing in frontier lab capabilities investing in frontier lab capabilities investing in frontier lab capabilities where we need it and there is a missing where we need it and there is a missing where we need it and there is a missing middle where somebody can make a lot of middle where somebody can make a lot of middle where somebody can make a lot of money figuring out how to route between money figuring out how to route between money figuring out how to route between them and make it easy to install a bunch them and make it easy to install a bunch them and make it easy to install a bunch of local models on your machine because of local models on your machine because of local models on your machine because it's not easy enough right now. It is it's not easy enough right now. It is it's not easy enough right now. It is just not easy enough right now. And just not easy enough right now. And just not easy enough right now. And ironically, Jensen owns the company best ironically, Jensen owns the company best ironically, Jensen owns the company best positioned to do it. I'm not sure he's positioned to do it. I'm not sure he's positioned to do it. I'm not sure he's incentivized to do that, but he does incentivized to do that, but he does incentivized to do that, but he does love open source, so he might do it in love open source, so he might do it in love open source, so he might do it in that future. Nvidia may end up capturing that future. Nvidia may end up capturing that future. Nvidia may end up capturing all the expensive work upstream. Apple all the expensive work upstream. Apple all the expensive work upstream. Apple is selling the complete computer and the is selling the complete computer and the is selling the complete computer and the memory where the model finally runs and memory where the model finally runs and memory where the model finally runs and the labs are getting 10 to 20% premier the labs are getting 10 to 20% premier the labs are getting 10 to 20% premier work and you are getting premier prices work and you are getting premier prices work and you are getting premier prices for that premier model. That could be for that premier model. That could be for that premier model. That could be where proumers are going. I think that where proumers are going. I think that where proumers are going. I think that the jury is out. I think that we're the jury is out. I think that we're the jury is out. I think that we're going to be looking at stuff like open going to be looking at stuff like open going to be looking at stuff like open router to see where people are spending router to see where people are spending router to see where people are spending on tokens. More and more of open router on tokens. More and more of open router on tokens. More and more of open router is going to open source tokens by the is going to open source tokens by the is going to open source tokens by the way. So I would not be surprised to see way. So I would not be surprised to see way. So I would not be surprised to see a future by December where we are at 80% a future by December where we are at 80% a future by December where we are at 80% of open router tokens on open- source of open router tokens on open- source of open router tokens on open- source open weights models. And that is a open weights models. And that is a open weights models. And that is a future that strongly supports these future that strongly supports these future that strongly supports these Macs. So we're going to see we're going Macs. So we're going to see we're going Macs. So we're going to see we're going to find out. I would like you to tell me to find out. I would like you to tell me to find out. I would like you to tell me in the comments, are you going to buy in the comments, are you going to buy in the comments, are you going to buy one of these Macs? Do you think it's one of these Macs? Do you think it's one of these Macs? Do you think it's worth it? Do you think the install worth it? Do you think the install worth it? Do you think the install experience for local models is something experience for local models is something experience for local models is something that we can get past and make easier? Do that we can get past and make easier? Do that we can get past and make easier? Do you have ideas for how you would solve you have ideas for how you would solve you have ideas for how you would solve the routing problem for tasks? It's the routing problem for tasks? It's the routing problem for tasks? It's actually a very hard problem. How do we

actually a very hard problem. How do we actually a very hard problem. How do we solve the routing problem for tasks so solve the routing problem for tasks so solve the routing problem for tasks so that we can have intelligence on the that we can have intelligence on the that we can have intelligence on the machine as much as we possibly can, but machine as much as we possibly can, but machine as much as we possibly can, but route to frontier models where needed? route to frontier models where needed? route to frontier models where needed? These are real questions. And the These are real questions. And the These are real questions. And the exciting thing to me is that we all are exciting thing to me is that we all are exciting thing to me is that we all are going to figure out how that answer going to figure out how that answer going to figure out how that answer works with the money we choose to spend works with the money we choose to spend works with the money we choose to spend or not spend. Where are you putting your or not spend. Where are you putting your or not spend. Where are you putting your dollars down? That's the question I have dollars down? That's the question I have dollars down? That's the question I have for you. I'll see you next time.