AI Tool Better Than OpenClaw? + NVIDIA’S $1T Prediction & AI Image Wars
TL;DR
Nvidia says the AI buildout is far from over — At GTC, Jensen Huang said Nvidia has done roughly $500 billion in chip sales over the last year and expects demand through the end of 2027 to reach $1 trillion in purchase orders, which he called conservative.
The next compute boom is agents and inference, not just giant pretraining runs — Matt’s big takeaway from GTC was that demand is shifting from scraping-the-internet pretraining toward post-training, RLHF, and test-time compute, with OpenClaw/Nemotron-style agents positioned as the “web browser” moment for AI assistants.
Nvidia isn’t launching an OpenClaw rival so much as an OpenClaw wrapper — NemoClaw packages OpenClaw with Nvidia add-ons for security, privacy, and easy installation of models like Neotron 120B, aiming to make local agents less intimidating and more enterprise-ready.
Midjourney’s image lead looks gone — In a live hand test, Midjourney V8 still mangled fingers, while Google’s “Nano Banana” and Microsoft’s new MAI Image 2 handled the same prompt cleanly, reinforcing Matt’s point that Midjourney fumbled a once-dominant position.
AI exposure is high, but job destruction hasn’t fully shown up in the data yet — Reviewing Andrej Karpathy’s job-market visualizer, they noted roles like software developers and lawyers look highly exposed to AI, even though Bureau of Labor Statistics data still shows many of those categories growing.
A weird new labor market is emerging: humans paid to generate training data — DoorDash Tasks now includes work like filming everyday actions or speaking in another language to help train AI and robotics systems, which Matt sees as a possible template for how AI wealth gets partially redistributed without full UBI.
The Breakdown
A bittersweet sendoff starts with Nvidia at the center of everything
The episode opens as a last-for-a-while show, but they jump straight into GTC, which Matt describes as the “Super Bowl” or “Burning Man” for AI. His core vibe from being there: if AI is a solar system, Nvidia is the sun — every cloud company, robotics company, and serious AI player seems to orbit it.
Jensen’s $1 trillion chip-demand claim lands hard
The biggest number from GTC was Jensen Huang saying Nvidia has already done about $500 billion in chip sales and expects that to double to $1 trillion by the end of 2027. Matt says the wildest part came in a press Q&A: Jensen clarified that this number reflects purchase orders and letters of intent — companies saying, in effect, “we want the chips as soon as they exist.”
Why Nvidia thinks agents and inference will drive the next wave
Matt explains that AI demand is shifting away from the old pretraining phase toward post-training, reinforcement learning, and especially inference — the “test-time compute” where a model visibly thinks through an answer. He uses Groq as the vivid example: what once took 15 minutes of reasoning could shrink to 10 seconds, which makes the idea of everyone running personal agents in the background feel much more plausible.
OpenClaw becomes the star of GTC, and NemoClaw is Nvidia’s bet on safer adoption
Jensen reportedly spent around 20 minutes of his keynote on OpenClaw, with Nvidia even running a “Build-A-Claw” booth. The hosts stress that NemoClaw is not an OpenClaw competitor but a bundle: OpenClaw plus Nvidia security, privacy, and model integrations like Neotron 120B, aimed at making agents easier to install and less scary for normal users who still flinch at opening a terminal.
Space data centers sound cool, but the heat problem is still unsolved
They also hit Nvidia’s space-computing ambitions, and Matt is careful not to oversell it. The missing piece is exactly what he’s ranted about before: heat dissipation in a vacuum, where GPUs run hot, the sun adds more heat, and there’s no simple way for that heat to flow away — so Jensen gave no timeline because they genuinely don’t have the answer yet.
Midjourney’s comeback gets tested and mostly flunks the hand exam
Shifting to image models, they react to Midjourney V8 and immediately notice weird fingers, even on showcase images. In a live prompt asking for “a group of people in a circle all putting their hands in,” Midjourney produces the usual cursed anatomy — “elephant trunk” fingers and hands melting together — while Nano Banana passes the same test cleanly.
Microsoft’s MAI Image 2 joins the image wars looking surprisingly strong
They then try Microsoft’s new MAI Image 2, which Matt says feels roughly on par with Nano Banana and ranks highly in blind tests. It nails a neon-sign text prompt (“Joe Fear loves tacos in Tokyo”) and handles hands well, though a San Diego Padres baseball test reveals the usual surreal image-model weakness: correct logos, totally broken stadium geometry.
AI may threaten jobs, but the labor data isn’t collapsing yet
Using Andrej Karpathy’s U.S. job-market visualizer, they compare BLS growth data with an “AI exposure” layer and find the contrast fascinating. Software developers, lawyers, and paralegals look highly exposed to AI, but many of those jobs are still growing in actual labor statistics — which leads them to a more nuanced take: exposure is real, but displacement is arriving unevenly and slower than people fear.
What people want from AI — and DoorDash’s glimpse of a stranger labor future
On Anthropic’s survey of 81,000 people, they’re struck that the top hope isn’t “time freedom” but “professional excellence,” while the second-biggest lived outcome is simply “AI hasn’t delivered.” Then they close on DoorDash Tasks, where workers can get paid to film actions or record speech for AI training; Matt sees that as a revealing, slightly dystopian clue about how future AI economies might pay humans not for traditional work, but for data.
One last robot, one last sign-off
They end with a humanoid robot rallying in tennis — not a general-purpose worker yet, but a strong example of robotics becoming physically capable faster than its “brain” is catching up. It’s a fitting closer: weird, exciting, a little uncanny, and a reminder of why this show was fun in the first place before they sign off for hiatus.