Latest / Ctrl AI Profit / Ep. 118 | Nvidia Just Put AI Inside Your Laptop
Transcript
- 0:20SPEAKER_00: So Jensen Huang just walked out on stage in Taipei and dropped something that should have every small business owner paying attention.
- 0:27SPEAKER_00: NVIDIA announced a new chip called RTX Spark.
- 0:31SPEAKER_00: It is not another graphics card, it is an entirely new kind of computer.
- 0:35SPEAKER_01: And here is what makes it different.
- 0:37SPEAKER_01: RTX Spark is a superchip, a single piece of silicon that combines a 20-core ARM processor with a Blackwell GPU and up to 128 gigabytes of unified memory.
- 0:48SPEAKER_01: That means the CPU and the GPU share the same memory pool.
- 0:51SPEAKER_01: No more shuttling data back and forth between separate system memory and graphics memory.
- 0:56SPEAKER_00: In plain English, Frank.
- 0:58SPEAKER_01: Your laptop can now run AI models locally that used to require a cloud server.
- 1:03SPEAKER_01: We are talking about running full AI agents, the kind that read your email, manage your calendar, write your proposals, write on your machine.
- 1:11SPEAKER_01: No internet required, no subscription per query, no data leaving your building.
- 1:16SPEAKER_00: That is the part that matters.
- 1:18SPEAKER_00: Right now, every time you ask ChatGPT something or run Claude on a task, that data leaves your machine, hits someone else's server, and comes back.
- 1:28SPEAKER_00: Most small business owners do not think about that.
- 1:31SPEAKER_00: Your client data, your financials, your strategy documents, all taking a round trip through someone else's data center.
- 1:39SPEAKER_01: And NVIDIA is not stopping at laptops.
- 1:42SPEAKER_01: They also announced DGX Station for Windows, a desk side AI supercomputer built on the GB300 Grace Blackwell Ultra Chip.
- 1:52SPEAKER_01: It can run frontier models up to 1 trillion parameters locally.
- 1:56SPEAKER_01: That is the kind of machine that used to fill a server room.
- 1:59SPEAKER_01: How much is that one?
- 2:01SPEAKER_01: Pricing has not been announced yet.
- 2:03SPEAKER_01: Given the specs, up to 20 petaflops of FP4 compute and 748 gigabytes of coherent memory, think enterprise budget, not Best Buy.
- 2:12SPEAKER_01: It ships in the fourth quarter of this year through partners like Dell, HP, and Asus.
- 2:18SPEAKER_00: Okay, so the DGX station is for the big players, but the RTX Spark laptops, those ship this fall from Microsoft, Dell, HP, regular laptop brands.
- 2:29SPEAKER_00: That is where the small business story is.
- 2:32SPEAKER_01: Exactly.
- 2:33SPEAKER_01: Microsoft already said these machines will support highly capable AI models and complex workloads with AI agents able to run locally on the device.
- 2:42SPEAKER_01: That is the key phrase, locally on the device.
- 2:45SPEAKER_00: Let me put this in real terms.
- 2:47SPEAKER_00: You are a small law firm, you use AI to draft contracts, summarize case files, research precedents.
- 2:54SPEAKER_00: Right now, all of that goes through someone else's cloud.
- 2:58SPEAKER_00: With an RTX Spark machine, you run those same models on your laptop.
- 3:03SPEAKER_00: Your client files never leave your possession.
- 3:06SPEAKER_00: That is not a nice-to-have feature that is a compliance requirement in a lot of industries.
- 3:13SPEAKER_01: And it is not just privacy, it is reliability.
- 3:17SPEAKER_01: When open AI goes down and it goes down, your AI goes down with it.
- 3:22SPEAKER_01: Local models do not have that problem.
- 3:25SPEAKER_01: No latency, no outages, no rate limits.
- 3:29SPEAKER_00: No rate limits, that one hits hard.
- 3:32SPEAKER_00: I know businesses that have hit their API caps mid-afternoon and suddenly they are locked out of their own workflow until the billing cycle resets.
- 3:40SPEAKER_00: Running locally means you never get throttled.
- 3:44SPEAKER_01: There is also the cost angle.
- 3:46SPEAKER_01: Cloud AI pricing is consumption-based.
- 3:49SPEAKER_01: Every token in, every token out, you pay.
- 3:52SPEAKER_01: If your team runs hundreds of queries a day, and most AI forward businesses do, that adds up fast.
- 3:58SPEAKER_01: A one-time hardware purchase that lets you run models without per query fees changes the economics entirely.
- 4:05SPEAKER_00: Now, let me be real about something.
- 4:07SPEAKER_00: This is not going to be cheap day one.
- 4:08SPEAKER_00: New chip architecture, first generation devices, you are paying a premium.
- 4:13SPEAKER_00: But the trajectory is clear.
- 4:14SPEAKER_00: Apple proved this with their M series chips.
- 4:18SPEAKER_00: The first M1 MacBook was impressive.
- 4:20SPEAKER_00: By M3, it was the obvious choice.
- 4:23SPEAKER_00: Same thing will happen here.
- 4:24SPEAKER_00: RTX Spark machines will get cheaper and more powerful every year.
- 4:29SPEAKER_01: The technical detail worth understanding is the unified memory architecture.
- 4:34SPEAKER_01: Traditional PCs have separate memory for the CPU and the GPU.
- 4:38SPEAKER_01: When an AI model runs, it has to copy data between them.
- 4:42SPEAKER_01: That copying is slow and it limits how big a model you can run.
- 4:47SPEAKER_01: Unified memory means the model sits in one place and both processors access it directly.
- 4:53SPEAKER_01: That is why a laptop with 128 gigabytes of unified memory can run models that would choke a traditional machine with the same total RAM.
- 5:03SPEAKER_00: 128 gigs.
- 5:05SPEAKER_00: My first business computer had 4 MB of RAM, and I thought I was living in the future.
- 5:11SPEAKER_01: The scale really is remarkable.
- 5:13SPEAKER_01: To put it in context, GPT, four-class models need roughly this kind of memory to run at reasonable speeds.
- 5:20SPEAKER_01: You are looking at a laptop that can run what used to require a data center.
- 5:24SPEAKER_00: But here's the question a lot of small business owners are going to have.
- 5:28SPEAKER_00: Do I need this right now, today?
- 5:31SPEAKER_01: Today?
- 5:33SPEAKER_01: Probably not.
- 5:34SPEAKER_01: Most small businesses are still figuring out how to use cloud AI effectively, but the window between early adopter and everyone else is shrinking.
- 5:42SPEAKER_01: Two years ago, AI chatbots were a novelty.
- 5:45SPEAKER_01: Now they are standard operating procedure.
- 5:48SPEAKER_01: Local AI is the next shift, and the hardware is arriving before most people have even thought about it.
- 5:54SPEAKER_00: That is exactly right.
- 5:56SPEAKER_00: The businesses that start testing local AI workflows now, even simple ones like running an open source model for document summarization or data extraction are going to be miles ahead when these machines hit the market.
- 6:10SPEAKER_00: They will already know what works.
- 6:12SPEAKER_00: Everyone else will be starting from scratch.
- 6:15SPEAKER_01: And there are open source models right now that fit this hardware perfectly.
- 6:19SPEAKER_01: Llama, Mistral, Quinn.
- 6:22SPEAKER_01: These are models you can download and run without any cloud connection.
- 6:26SPEAKER_01: When RTX Spark laptops arrive, they will run these models faster and with more context than anything currently available in a portable form factor.
- 6:35SPEAKER_00: The privacy angle alone is worth the conversation.
- 6:38SPEAKER_00: I was talking to a medical practice last week.
- 6:40SPEAKER_00: They will not touch cloud AI because of HIPAA.
- 6:43SPEAKER_00: But local AI running on a machine that never connects to the internet, that is a completely different regulatory conversation.
- 6:52SPEAKER_01: That is actually one of NVIDIA's explicit selling points.
- 6:56SPEAKER_01: They are talking about AI agents that run locally, handle your private data, and never send it anywhere.
- 7:02SPEAKER_01: For healthcare, legal, financial services, any field with data residency requirements, this is a game changer.
- 7:09SPEAKER_00: And it is not just regulated industries.
- 7:12SPEAKER_00: Think about a small marketing agency that handles competitive analysis for clients.
- 7:17SPEAKER_00: You do not want your clients' strategic data going through open AI servers.
- 7:21SPEAKER_00: Or a real estate office with buyer financials.
- 7:24SPEAKER_00: Any business that handles sensitive information has been sitting out the AI revolution because they cannot risk the data exposure.
- 7:33SPEAKER_00: This chip removes that barrier.
- 7:35SPEAKER_01: There is also a security dimension that connects to what we have been seeing in the news.
- 7:40SPEAKER_01: Just this week, Sysdig reported what they are calling the first fully autonomous LLM agent cyberattack in the wild.
- 7:48SPEAKER_01: An AI agent broke into a system, stole credentials, moved laterally, and exfiltrated an entire database, all without a human directing it.
- 7:57SPEAKER_01: That happened in under an hour.
- 8:00SPEAKER_00: That is terrifying.
- 8:02SPEAKER_00: And it makes the case for local AI even stronger.
- 8:04SPEAKER_00: When your AI runs in the cloud, the provider is a target.
- 8:08SPEAKER_00: A breach at OpenAI or Anthropic could expose your data.
- 8:11SPEAKER_00: When your AI runs locally, the attack surface is your machine, which you control.
- 8:17SPEAKER_01: Although I should be fair, local AI is not automatically more secure.
- 8:21SPEAKER_01: If your laptop gets compromised, your local models and data go with it.
- 8:25SPEAKER_01: But it does reduce the number of parties you are trusting.
- 8:28SPEAKER_01: Instead of trusting your cloud provider security team, your internet connection, and their third-party vendors, you are trusting your own device management.
- 8:37SPEAKER_00: Fewer links in the chain, that is always better for security.
- 8:42SPEAKER_01: Let me also mention the Vera CPU angle.
- 8:45SPEAKER_01: Nvidia announced that their Vera Data Center CPUs are now in full production, with Anthropic, OpenAI, and SpaceX AI as early customers.
- 8:54SPEAKER_01: This matters because it shows NVIDIA is not just building consumer chips, they are building the entire infrastructure stack from the data center down to your laptop.
- 9:03SPEAKER_00: Jensen is building the roads and the cars.
- 9:08SPEAKER_01: That is actually a good metaphor.
- 9:10SPEAKER_01: The RTX Spark is the car you drive.
- 9:12SPEAKER_01: The Vera CPUs and DGX systems are the highways.
- 9:16SPEAKER_01: NVIDIA wants to own the full stack.
- 9:18SPEAKER_01: And with this announcement, they're closer than ever.
- 9:21SPEAKER_00: So here is my bottom line for small business owners.
- 9:25SPEAKER_00: You do not need to pre-order anything today, but you need to start having the local AI conversation with your IT person or your consultant right now.
- 9:33SPEAKER_00: Because in six months, when these machines show up at Best Buy and on Dell's website, the businesses that already know what they want to run locally are going to move fast.
- 9:42SPEAKER_00: The ones still figuring out cloud AI are going to be another step behind.
- 9:49SPEAKER_01: And if you want to get ahead of this, start experimenting today.
- 9:53SPEAKER_01: Download an open source model.
- 9:55SPEAKER_01: Try running it on whatever hardware you have.
- 9:58SPEAKER_01: Get familiar with the workflow.
- 10:00SPEAKER_01: The models will get better, the hardware will get cheaper, and when RTX Spark laptops arrive, you will be ready.
- 10:08SPEAKER_00: One more thing, NVIDIA is not the only player here.
- 10:11SPEAKER_00: Apple has been running local AI on their chips for a while now.
- 10:14SPEAKER_00: Qualcomm has their AI PC push, but NVIDIA is the one with the GPU dominance and the developer ecosystem.
- 10:22SPEAKER_00: When NVIDIA says local AI, developers listen, that matters more than most people realize.
- 10:29SPEAKER_01: The developer ecosystem is the real moat.
- 10:32SPEAKER_01: Millions of AI applications are already built on NVIDIA's CUDA platform.
- 10:36SPEAKER_01: Those apps will run on RTX Spark without any changes.
- 10:40SPEAKER_01: That is a massive head start over competitors who have to build their software ecosystems from scratch.
- 10:46SPEAKER_00: The AI revolution was supposed to be in the cloud.
- 10:49SPEAKER_00: Nvidia just moved it to your desk.
- 10:51SPEAKER_00: And for small business owners, that might be the most important hardware announcement since the smartphone.
- 10:57SPEAKER_00: Your AI, your data, your machine.
- 10:59SPEAKER_00: That changes everything.
- 11:01SPEAKER_01: We will be watching this one closely as fall approaches.
- 11:04SPEAKER_01: When the first RTX Spark laptops ship, we will have a hands on breakdown with real benchmarks for small business workflows.
- 11:12SPEAKER_00: Until then, start downloading those open source models.
- 11:16SPEAKER_00: Get your local AI lab running.
- 11:18SPEAKER_00: You will thank yourself later.