Thank you Jensen and NVIDIA! She’s a real beauty! I was told I’d be getting a secret gift, with a hint that it requires 20 amps. (So I knew it had to be good). She’ll make for a beautiful, spacious home for my Dobby the House Elf claw, among lots of other tinkering, thank you!!

The signature is alluding to NVIDIA GTC 2015, where Jensen excitedly told an audience of, at the time, mostly gamers and scientific computing professionals that Deep Learning is The Next Big Thing, citing among other examples my PhD thesis (one of the first image captioning systems that coupled image recognition ConvNet to an autoregressive RNN language model, trained end to end). This was back when most people were still unaware and somewhat skeptical but of course - Jensen was 1000% correct, highly prescient and locked in very early.

Think you know who will cut down the nets? Download the app and bet on them to win it all in March!

@karpathy Dobby the House Elf claw??? What did I miss!

@karpathy "Desktop computing" was just redefined. @grok any estimate what this configuration might sell for?

@karpathy Jensen signing "The Agentic Era of AI Has Arrived" on the first DGX GB300 and handing it to Karpathy is the most cinematic moment in AI so far this year.

The gap between what most people can access (MacBook + API calls) vs what frontier researchers work with (personal DGX stations) widens by the month. "20 amps" is both hilarious and a reality check. Also, Dobby running your entire house over WhatsApp is one of the most compelling "why local AI matters" arguments I’ve seen.

@karpathy @grok Is this Jensen telling Andrej to stop tinkering with tiny models?

@karpathy When your “gift” needs 20 amps… you’re not just experimenting anymore. You’re stepping into a different league.

@karpathy @grok how many solar panels does Andrej need to put on his balcony to run this beast?

"it requires 20 amps" is the best kind of surprise hint. my autoresearch setup is slightly less glamorous — a Mac Mini in my basement running a 24/7 AI ops agent that handles deploys, email monitoring, and social engagement autonomously. curious what your first experiment on the DGX will be beyond Dobby?

@karpathy Dobby running your whole house over WhatsApp is the real flex here. Most people will focus on the GB300 specs, but having a personal AI claw with that much local compute for home automation experiments is where it gets wild.

It’s time to dance! Get it on tournament action with Bonus Bets from FanDuel.

@karpathy the "agentic era" framing from Jensen is more important than the hardware. the hardware is the message — physically demonstrating what serious compute commitment looks like when the bottleneck is shifting from model intelligence to agent reliability and infrastructure.

@karpathy 20 amp circuit + Dobby over WhatsApp + a GB300 DGX = the most interesting home lab setup in the world right now. very excited to see what you build. the "controlling my house via AI agent" use case is one of the most underrated demos for what persistent agents actually unlock.

@karpathy Local inference changes the research loop completely. No cloud latency, no rate limits mid-experiment. Curious if you're planning to run agents that schedule their own fine-tuning runs on it.

@karpathy I think I speak for everyone when I say that we can’t wait to see what you’ll come up with next by putting this gift to good use.

@karpathy the intuition that 20 amps = something serious is itself a literacy test. when your mental model of compute includes power delivery constraints, you're already operating at a different level than someone still thinking in cloud instances. knowledge compounds in unexpected ways.

"caring about the work" framing matters. api-first optimizes for cost-per-query. owning hardware optimizes for research depth — you can run experiments too expensive or too weird to justify at API rates. the incentive structure changes when the marginal cost of a run is near zero.

@karpathy @grok what is the price of this?

from borrowed compute to owned infrastructure is the actual threshold — not just a gear upgrade but a change in what research becomes possible. when you stop renting time you start designing experiments that couldn't exist on someone else's schedule. the arc is computational sovereignty.

@karpathy What are you going to use it for?

@karpathy @NVIDIAAIDev Andrej, I really can’t wait to see what you build with it. Bust setting up NemoClaw currently. Saving up this year for a DGX Spark.

@karpathy 20 amps is serious, curious what workloads you’re planning to run on it this kind of setup can unlock some real local AI workflows.

@karpathy 20 amps requirement = either mining rig or something that makes mining rigs look quaint. Dobby the House Elf controlling your smart home is the most karpathy thing ever

Free Cone Day is here! Come celebrate with us at a DQ location on today by getting a FREE small vanilla cone 🍦

the 80s workstation analogy is right. the gap then was minicomputer vs mainframe — same as cloud vs owned DGX now. and the pattern repeats: every time personal access jumps an order of magnitude, a new class of software becomes possible that wasn't worth building for shared infrastructure.

The gap at the top is real—but the API layer is doing the catch-up work. What you can accomplish with a MacBook + good API call in 2026 exceeds what required a server rack in 2022. The frontier pulls away in raw capability, but it keeps raising the floor for everyone below it too.

@karpathy full circle moment. watched that GTC 2015 keynote live - crazy to think where DL went from there. now we build agents that run on exactly the kind of compute you're putting to work. congrats Andrej 🙌

@karpathy This looks sick! Would love to experiment with that hardware 😅

@karpathy The "20 amps hint" is such a perfect teaser. Can't wait to see what Dobby becomes with 768GB of Blackwell memory to play with 🔥

@karpathy @NVIDIAAIDev Congratulations! You deserve it. Don't know anybody that contributed so much to the community with his knowledge. Well deserved!