the localhost:0010 | the lucky and unlucky

Jensen is out. Data centers went 1 for 41 at the ballot box. Who’s lucky in AI now?Voters went against data centers in 40 of 41 local measures. Then Jensen Huang resigned from Nvidia. On Friday the 13th, the five of us pick the lucky and the unlucky: Apple, AMD, the hyperscalers, and maybe humanity itself. Plus real customer numbers on local AI, why device cost barely registers with AI teams, and what we want to see at Microsoft Ignite.Next week: the localhost, live from Ignite in San Francisco.Views are our own and do not represent our employers.

It's the Halloween edition, which means a pirate, a horse, a lederhosen enthusiast, a mountain man and an NYPD captain spent 49 minutes having a genuinely good conversation about edge AI. Everyone showed up in costume and immediately started introducing each other as each other. It goes downhill from there, in the best way.Our guest is Mike McGinley, a Surface device specialist who sells to some of the largest banks in the world. He also got stopped on Fifth Avenue by a casting director who was looking for, and we quote, an older heavier set Irish retired New York City police officer. He is now in a Paramount+ miniseries. He has a SAG card. He was in roughly 50 commercials as a kid, including one for a Pogo Ball. We have the footage.Somewhere in between the costumes we actually did the show. Jacob is back from GitHub Universe where they had a DGX Station with 748 gigabytes of memory pushing 19 million tokens of context. Neil talks about wiring up Boeing airplane parts with Raspberry Pis in college and building a restaurant that set its prices based on whether the Cubs were in town. Frank found a company putting fire-detecting AI on tiny sensors in the forest. Chauncey found a robot duck. And the 128GB RTX Spark SKUs are already sold out.The thread running through all of it: the T-shirt sizes keep getting bigger on both ends. A phone doing transcription on one side, a $100,000 edge server running trillion parameter models on the other, and the same question underneath. Where should the intelligence actually live?Plus a demo of Ride Edge, Frank's app that reads 23 API feeds and tells him where to park his car to find Uber riders. Yes, it runs on the workstation behind him. Yes, he hijacked his own podcast to show it.CHAPTERS0:00 Cold open: from sensors to servers0:50 Halloween edition, and nobody knows who they are5:00 Meet Mike McGinley6:00 The casting director on Fifth Avenue8:00 A SAG card, 50 commercials, and the Pogo Ball10:00 Relationship selling when every dollar gets scrutinized13:00 Beyond cost savings: why banks actually want local15:00 Jacob returns from GitHub Universe16:00 748GB and 19 million tokens of context17:00 The 128GB Sparks are sold out18:00 Sam Altman thinks your DRAM is selfish18:30 Tiny sensors that spot forest fires19:00 Flock cameras, again20:00 Boeing parts, Raspberry Pis, and the restaurant of the future22:00 Why always-on workloads break the token model23:00 DEMO: Ride Edge, 23 API feeds, one workstation26:00 Proving the device pays for itself27:00 Retail branches and shelf restocking28:00 The Pixel deal that paid for the phones in weeks31:00 MicroDuck, and open sourcing a robot33:00 Finding weld spatter on a car body, locally35:00 You can't put a price tag on innovation37:00 Bank of New York builds its own colo39:00 T-shirt sizes get an XXL40:00 From "we won't touch DeepSeek" to "we prefer Qwen"42:00 Neil's offline PowerPoint generator at 35,000 feet43:00 Around the horn: what everyone's working on47:00 Snapdragon X2 vs RTX Spark49:00 YarrABOUT THE SHOWthe localhost is an ongoing podcast about AI that runs on your own hardware. Open weight models, NPUs, workstations, sensors, home labs, and the question underneath all of it: who controls intelligence, you or a data center?The crew: Chauncey Larsen and Jacob Rhoades of the Microsoft Surface Marketing team. Neil Misak of the Microsoft Surface Engineering AI Solution Factory. Robert Henry, Windows specialist, and not a salesman, as we have now been told. And Frank Buchholz, retired tech guy, now vibe coding, Uber driving, and unapologetic Chappell Roan and Phoebe Bridgers fan.This week's guest: Mike McGinley, Surface device specialist, Bergen County's finest, and soon to be on your television.DISCLOSUREAll opinions expressed on this show belong to the individual hosts and do not represent the views or positions of Microsoft or any other organization. We speak for ourselves and nobody else.New episodes weekly.https://thelocalhost.show

Linus Tech Tips asked who on earth buys a $6,000 AI laptop. Frank has a machine with two graphics cards in his office that never turns off. So who's right?Cloud AI keeps getting cheaper. AI hardware keeps getting more expensive. And somehow the workstation is back.This week: 400 people lined up at a Best Buy in San Francisco to buy RTX Spark devices, a company burned six figures in tokens building an ERP platform they had already licensed, and the crew tries to work out what is actually pulling people back toward expensive hardware they spent a decade being told they didn't need.Plus the best argument anyone made for buying one: when you're running on a meter, your behavior changes. You play it safe. Take the meter away and you try the weird ideas. One developer called it the right to fail cheaply.Chauncey is in Mexico. Scott Slater is in New York.CHAPTERS0:00 Cold open: who buys a $6,000 AI laptop?0:42 Around the horn: where everybody's been4:55 Scott Slater joins from New York5:25 What customers are actually asking about AI on device7:09 The Linus Tech Tips problem10:41 You're not buying tokens, you're buying freedom11:50 The cost of lost experimentation12:46 "The right to fail cheaply"13:12 Cloud overspend dwarfs the price of RAM13:54 What happens when every seller can build software16:32 The $100,000 ERP nobody needed17:55 The return of the workstation21:46 So what's actually moving the needle?22:47 Hybrid intelligence: the AI you never notice25:11 will.i.am's 200 students and their Spark devices26:35 Robert's hybrid intelligence blog27:26 Keeping your intelligence: IP protection and the second wave29:25 Why a JPEG is good enough31:03 Models checking models33:15 Nemotron, or "the new Transformers model"34:01 On my device: 23 API feeds and an Uber app36:53 An RTX Spark... in a car?39:09 The $100,000 DGX Station question41:41 Wrap: there's no place like 127.0.0.1ABOUT THE SHOWthe localhost is an ongoing podcast about AI that runs on your own hardware. Open weight models, NPUs, workstations, home labs, and the question underneath all of it: who controls intelligence, you or a data center?The crew: Chauncey Larsen and Jacob Rhoades of the Microsoft Surface Marketing team. Neil Misak of the Microsoft Surface Engineering AI Solution Factory. Robert Henry, Windows AI Solution Sales specialist. And Frank Buchholz, retired tech guy, now vibe coding, Uber driving, and unapologetic Chappell Roan and Phoebe Bridgers fan.This week we're lucky to have Scott Slater stepping in for Chauncey. Scott is a Managing Director in Windows Solution Sales.DISCLOSUREAll opinions expressed on this show belong to the individual hosts and do not represent the views or positions of Microsoft or any other organization. We speak for ourselves and nobody else.New episodes weekly.https://thelocalhost.show

Microsoft said "hybrid intelligence" on a main stage for 90 minutes. So the one host who no longer works there asks the four who do: what actually changed for a business?Introducing FAQ Around and Find Out: Frank Asks Questions. Can a defense contractor who won't send IP to the cloud finally use AI? What is an Execution Container in plain English? Why buy a Surface Laptop Ultra instead of a Mac full of unified memory? Is the NPU dead now that the GPU is back? And why did an analyst upgrade the stock the same week Microsoft made local inference free?In this episode:- Hybrid intelligence: on the device, in your data center, or in the cloud, under one set of rules- Execution Containers (MXC): bowling-alley bumpers for agents, now GA and cross-platform- Hydro Fusion: GitHub Copilot routes coding work to a local model automatically- Copilot on Windows vs GitHub Copilot vs Codex vs the CLI, untangled- Agent 365: identity and audit for agents, because agents are outnumbering employees- Surface Laptop Ultra vs Mac: Apple has the wider memory highway, NVIDIA makes the cars smaller- The NPU question: more important than ever, just less of a headline- "What's the catch?" How free local inference becomes an E7 storyChapters:00:00 Cold open00:50 Intros03:15 Hybrid intelligence comes full circle05:30 FAQ Around and Find Out06:00 Can a defense contractor use AI now?08:06 Execution Containers, explained10:51 MXC goes GA, Hydro Fusion routes local13:12 Copilot vs GitHub Copilot vs Codex17:03 Which announcement would you build on?18:24 Agent 36521:35 Surface Laptop Ultra vs Mac Mini24:29 FP4, speculative decoding, and memory bandwidth28:19 Does a claims adjuster care what it's called?30:30 Who pays $7,000 for a laptop?33:47 Is the NPU still relevant?38:19 Why did the stock go up?41:12 What's the catch? E7 for the win42:36 WrapHosts: Frank Buchholz, Chauncey Larsen, Jacob Rhoades, Robert Henry, and Neil Misak. Chauncey, Jacob, Robert, and Neil work at Microsoft. Everything here is personal opinion, not their employer's.the localhost is a weekly podcast about local AI: running models on your own hardware, with your own data. No cloud required. 127.0.0.1. There's no place like home.Watch on YouTube: https://youtube.com/@thelocalhost-showEverything else: https://thelocalhost.show

Ever nod along to an AI conversation and quietly have no idea what a "runtime" or a "parameter" actually is? This episode is for you.A listener (a Microsoft MVP, no less) told us he loves the show but doesn't understand all of it. Fair. So this week we go back to the studs and build local AI from the silicon up, one layer at a time, with zero jargon left unexplained. Head chefs, movie players, mixing boards, open-book tests. If you can follow a kitchen, you can follow this.What you'll learn:- Why memory, not the chip, decides which AI models your laptop can run- CPU vs GPU vs NPU, explained with a restaurant kitchen- What a model actually is (spoiler: it's just a file)- What "8 billion parameters" really means- Mixture of Experts, and why it fits like a big model but runs like a small one- Context windows, tokens, and why the model "forgets" you- MCP, agents, and why the app matters more than you think- Robert's closet full of old Surface laptops now running as one AI cluster with NVIDIA PAIRChapters00:00 Cold open: the comment that started it00:50 Intros (from a Sprinter van in Iowa)02:26 The stack: five layers, all on your machine03:20 Layer 1: Silicon. Head chef, prep cooks, and the food processor08:50 Layer 2: Runtime. The VLC for AI11:04 NVIDIA PAIR: one address, every machine in the house15:46 Layer 3: The model. It's just a file19:53 What's a parameter? Knobs on a mixing board21:46 Mixture of Experts: pull the L encyclopedia28:58 Layer 4: Context. The open-book test33:06 Layer 5: The app. Agents, 11-hour coding sessions, and the harness38:15 MCP: USB-C for AI42:17 Wrap: the whole stack in five linesAbout the localhostthe localhost is a weekly podcast about local AI: running models on your own hardware, in your own house, with your own data. No cloud required. We're five friends from the Windows and devices world who spend our free time (and a lot of our electricity) figuring out what actually works, and then explaining it in plain English. 127.0.0.1. There's no place like home.The hostsFrank Buchholz: Former CTO of Surface Marketing at Microsoft, now running Cadence 3, a consultancy focused on local and on-device AI. Recording this week from his Sprinter van.Chauncey Larsen: Microsoft. Diehard tech fan, Lego nerd, and the guy whose electricity bill proves it.Robert Henry: Microsoft, Windows Incubation. Owner of a closet full of Surface devices that are now an AI cluster. Will be wearing a different hockey jersey every week until further notice.Jacob Rhoades: Microsoft. Thinks about devices as endpoints for a living and runs more local models than anyone should.Neil Misak: Out this week, somewhere over the Atlantic on his way back from London. Back next episode.Chauncey, Robert, Jacob, and Neil work at Microsoft. Opinions here are their own and not their employer's.Shout out to Lars Berlau at HP Poly for the comment that made this episode happen. Lars, let us know if it landed.If you've been nodding along without following, you're exactly who we want to hear from. Drop a comment and tell us what to explain next.Listen everywhere: https://thelocalhost.show#LocalAI #AI #NPU #NVIDIA #DGXSpark #Surface #CopilotPlusPC #Ollama #LMStudio #MixtureOfExperts #MCP #Podcast

Five guys, five machines, and one question: what actually works?No debate this week. No vote. We just opened up our laptops and showed each other what we're actually running, what broke, and what turned out to be worth the trouble.Jacob figured out the whole local-plus-cloud workflow on a train from Munich to Paris with a spotty hotspot. Robert and Jacob are both splitting AI work across multiple machines with NVIDIA's new Personal AI Router, which means that graphics card in your desk drawer just became useful again. And Frank built an app that tells Uber drivers where to be, running entirely on a server in his office, controlled from his phone.Plus: Mozilla says the gap between the best cloud models and the ones you can download has narrowed to about four months. Neil takes over the Surface AI Factory. Robert finds out about it live on the show.And if you've ever thought "I could never run AI myself," the whole middle of this episode is the answer. There's no syntax anymore. It's just a conversation.CHAPTERS0:00 Cold open: what actually works1:00 Around the horn, and Neil's news2:18 Robert's next chapter3:16 The Rundown: the four month gap11:35 How do you actually start?12:10 Chauncey: okay but what do I DO with it?13:00 Jacob's train from Munich to Paris14:50 Get over the fear factor23:00 NVIDIA PAIR: your old GPU just got useful25:49 A brief and important pop culture quiz26:40 Perplexity Computer, and the bicycle company36:27 The Uber driver app running on The Beast39:00 Wrap: there's no place like 127.0.0.1the localhost is five tech industry veterans arguing about AI that runs on your own hardware. Open weight models, NPUs, home labs, and the question underneath it all: who controls intelligence, you or a data center? New episodes weekly.https://thelocalhost.show

An escape hatch is the thing that lets you leave. You build on something that also works on somebody else's hardware, so your vendor can't lock you in.Open weight models were supposed to be that hatch for AI. Download them, run them yourself, walk away whenever you want.Then NVIDIA bought Hugging Face, the place everyone downloads them from.Frank worked at Silicon Graphics in the '90s and says he's seen this movie before. Five of us argue about it for forty minutes: is the hatch still open, or did the company you'd escape from just buy the exit?Plus what happened when we ran a 180 billion parameter model at home, and why one of us cancelled a $200 a month subscription.(0:00) Cold open: I've seen this movie before(2:18) The Rundown: NVIDIA closes the Hugging Face deal(12:31) More from the week's news(22:43) The Tension: is the escape hatch still open?(27:46) Neil's counter: the ovens, the recipes, and the CUDA moat(29:15) Robert flips it: a door instead of a hatch?(30:10) The vote(33:52) On My Device(39:38) Wrapthe localhost is five tech industry veterans arguing about AI that runs on your own hardware. Open weight models, NPUs, home labs, and the question underneath it all: who controls intelligence, you or a data center? New episodes weekly.thelocalhost.show

Last week we argued the $25 token was dying. This week: that's exactly when the hardware boom starts.Frank takes the hot seat with a paradox. If intelligence keeps getting cheaper, we don't use less of it — we invent more things to do with it. More agents, more inference, more workloads. Which leaves every enterprise with one question they haven't had to ask before: which intelligence should we rent, and which should we own?Citi and Vercel are reportedly running open-weight models more than half the time. In June it was 29%. By late August, 53%. A new family of models out of the UAE ships six versions built for six classes of hardware — phone to server — with no quantization involved. NVIDIA's Hugging Face acquisition is official. And OpenAI shipped Astra.We also had four people instead of five, a guest from London, and for the first time in three episodes, a completely unanimous vote.━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━CHAPTERS00:00 If intelligence gets cheap, what do you buy?00:47 Welcome back — who's here, who's at Disneyland01:28 Laura Osborn joins from London02:14 Neil, and the Bears/Seahawks problem03:07 Chauncey, and the disclosure03:43 The Rundown: NVIDIA and Hugging Face is official04:25 Neil: what GitHub can tell us about this05:48 Citi and Vercel pass 50% open weights07:50 The three-tier model: endpoint, edge, cloud08:56 Six models for six classes of hardware10:22 Purpose-built beats shrinking things down?11:21 What UK and European customers actually ask for14:41 From cloud architect to local AI15:18 OpenAI ships Astra16:26 Chauncey: be careful what you point it at17:51 Neil: code generation at the edge19:33 Baseline: what should people actually care about?20:20 Open weights are no longer philosophical21:27 What's coming out of IFA22:52 The medical school of the future25:00 Why doesn't anyone know this is possible yet?26:53 "We have more agents than human employees"28:23 Tension of the Week: the hardware boom begins30:27 Chauncey: fixed cost beats a variable one32:39 Laura: I spent five years telling everyone to go cloud34:18 Neil: when pennies turn into a million dollars35:00 The brewery test36:06 The data center next door36:34 The vote38:42 On My Device: the 36B gets the job40:07 Chauncey: Copilot CLI, running locally41:44 The Dream Team and the one-person virtual business41:56 Laura: an on-device speaker coach42:58 Neil: Mac vs the Beast, round two━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━WHAT CAME UP• NVIDIA acquiring Hugging Face is confirmed. NVIDIA says it stays open and that NVIDIA hardware won't be required. Neil's read: this looks like Microsoft buying GitHub — the same fears, and possibly the same outcome. His angle is that Hugging Face has been effectively off-limits to a lot of enterprises, and being inside a vendor with a security and compliance apparatus may be what finally gets those models sanctioned.• Citi and Vercel are reported to be running open-weight models for more than half their AI workloads — 29% in June, 53% by late August. Ten weeks.• The MBZUAI Institute of Foundational Models released six models spanning roughly 1B to 375B parameters, each trained for a specific class of hardware rather than quantized down from something larger. Frank ran the 36B on his dual-3090 workstation against Qwen3.8-27B: about 68 tokens/sec against 40, on a comparable qualifying profile.• Neil's three tiers: the endpoint, the cloud, and an emerging middle — departmental edge servers. Local doesn't grant you HIPAA or FERPA compliance by itself, but no round trip to the cloud is a smaller attack surface.• The economics nobody models: an insurance agent filing 50 claims a day, times 10,000 agents. Even at pennies per call, that's over a million dollars a year in workloads that could run at the edge.• Laura, five years a cloud architect, on doing a full 180: the constraint in Europe isn't just sovereignty rules, it's that cloud capacity is at its limits.• And Neil's field test of public sentiment, conducted on a bartender who asked him point blank whether he was anti-AI.The vote was 4–0. First unanimous verdict of the series.━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━THIS WEEK'S CREWFrank Buchholz — host, independentLaura Osborn — guest, joining from LondonNeil MisakChauncey LarsenJacob Rhoades and Robert Henry are off this week. Jacob is celebrating his first anniversary. Robert was on a plane, sending Teams messages anyway.DISCLOSURE: Laura, Neil and Chauncey are employed by Microsoft. Frank is independent. All views expressed are their own, nothing discussed is unannounced or non-public, and none of this is financial advice.━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━localhost is a weekly show about running AI on hardware you own. Independent and unsponsored.New episodes weekly: https://thelocalhost.showThere's no place like 127.0.0.1.#LocalAI #OpenWeights #AI #EdgeAI #AIHardware #Sovereignty #OnDeviceAI #AIPodcast

Last week the conversation was cost vs. control. This week the cost floor fell out. Frontier models are getting dramatically cheaper, the models you can run on your own hardware are getting dramatically better — and Robert makes the case that within 18 months the only thing frontier labs will really be able to sell is speed.The crew gets into NVIDIA's reported interest in Hugging Face, a 170B+ parameter model coaxed onto a single consumer card, what Apple's latest silicon means for local AI, and the question underneath all of it: when intelligence becomes a commodity, where does the money go?Plus: an aerospace CEO in the back of an Uber who can't send his engineering IP to the cloud, and what happened when Robert turned on autopilot and walked away from his desk.the localhost is Frank Buchholz (independent), Jacob Rhoades, Robert Henry, Neil Misak, and Chauncey Larsen (Microsoft, opinions their own). Independent and unsponsored.New episodes weekly: https://thelocalhost.showThere's no place like 127.0.0.1.

Five tech industry veterans, one hot seat, and a vote. In our first episode: Meta gave away a new AI model and promised the big one soon, NVIDIA's CEO used his first ever post on X to defend open weights, and 270 companies signed on. One did not. So Frank takes the hot seat to argue that openness is positional, not principled, and the crew tries to take him apart. Along the way: what open weight actually means (up to the secret sauce), the Flock camera controversy as edge AI's cautionary tale, and a job interview for five AI models on a used $700 graphics card, with a winner nobody was talking about. Plus the stat of the night: one host's router setup runs 86 percent of his AI locally and has already saved him $1,600.We run the models, test the claims, and share what actually happened, including our own mistakes. New episodes weekly. There's no place like 127.0.0.1.Creators & Guests Chauncey Larsen - Host Frank Buchholz - Host Jacob Rhoades - Host Neil Misak - Host Robert Henry - Host (00:00) - cold open: from cost to control (01:05) - meet the hosts (and who writes the paychecks) (04:50) - the rundown: qwen 3.8 lands (07:55) - open weight vs open source, explained (14:45) - flock cameras: edge capture, central control (17:40) - this week's tension: open is a position (21:05) - the debate: manifestos, chess games, and IPOs (29:15) - the vote (31:40) - on my device: five models walk into a job interview (34:45) - what's running on the crew's devices (37:45) - the 86 percent stat (40:20) - wrap: there's no place like 127.0.0.1 Mark Zuckerberg's manifesto, "The Future Is for Everyone": https://www.meta.com/thefutureisforeveryone/ The Open Weights letter coverage and Spark 1.2 status: https://www.orcarouter.ai/blog/meta-muse-spark-1-2-explained Glimmer launch coverage: https://www.cnbc.com/2026/08/10/meta-muse-glimmer-open-weight-ai.html The May "not suitable for open sourcing" quote: https://www.implicator.ai/meta-releases-30b-open-weight-muse-glimmer-and-promises-spark-1-2-weights/ Flock's announced changes: https://www.nbcnews.com/tech/security/flock-safety-police-abuse-oversight-data-retention-rcna592217 Qwen 3.8-27B: https://huggingface.co/Qwen/Qwen3.8-27B The 2.4T open release: https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B Frank's full benchmark, raw traces and all: https://github.com/frankcx1/bakeoff The results post: https://www.linkedin.com/posts/frankjbuchholz_on-monday-meta-gave-away-a-brand-new-ai-share-7493881833236123648-1kd1/ the localhost is five friends talking about AI on your own hardware: Frank Buchholz (independent), Jacob Rhoades, Robert Henry, Neil Misak, and Chauncey Larsen (Microsoft, opinions their own). The show is independent and unsponsored. Find us at https://thelocalhost.show