BrokeIt - Daily AI News · All episodes: ↗

Engineered Scarcity, Meeting AI, AMD's Inference Push

2026-11-14 · 7 min

Listen · Apple Podcasts Listen · Spotify

Stories covered

Transcript

Intro

Ivy: They were giving startups money... on the condition they use it to buy back their own GPUs.

Marcus: [beat] Just let that sink in. I'm Marcus.

Ivy: And I'm Ivy.

Marcus: And you're listening to Broke It, with your daily AI news for Saturday, November 14th, 2026.

Ivy: We've got three big stories today, and they hit close to home for anyone building in this space.

Marcus: First up: AMD is taking aim at llama.cpp with a new, experimental way to run local models.

Ivy: Then, we'll talk about the newest member of your team: an AI that sits in on your video calls.

Marcus: And finally, we'll get into that bombshell opener: explosive allegations that Nvidia manufactured its own scarcity crisis.

Testing hrx-loom: AMD's New Stack for Custom Inference Kernels

Marcus: Alright, let's start with squeezing every last drop of performance out of our local LLMs.

Ivy: Something you are professionally obsessed with.

Marcus: Exactly! So AMD has a new experimental stack, hrx-loom, and it's promising to generate inference kernels even faster than the gold standard, llama.cpp.

Ivy: Okay, 'experimental' is the key word there. And 'kernels'—we're talking about the super low-level code that does the actual math for the AI, right?

Marcus: That's right. This isn't a whole new framework. It's a tool for building these hyper-optimized little engines for a specific model, on specific hardware. The dev got insane speeds on AMD's new Strix Halo platform.

Ivy: Right, 'specific hardware.' And by 'new' you mean 'hardware that almost no one has yet'.

Marcus: Well, yeah, but that's how progress happens! You build for the future. This could be huge for on-device AI performance on next-gen laptops.

Ivy: So is llama.cpp in trouble?

Marcus: Not yet. Llama.cpp runs everywhere, on everything. For now, this is a niche tool for power users. But it proves AMD is getting serious about competing on software and tools, not just silicon.

Ivy: [dry] So for most people, this means... nothing yet. Unless you're buying a top-of-the-line AMD laptop next year and love compiling custom code, this is just a neat experiment for now.

Marcus: A very fast curiosity, Ivy. A very fast one.

The AI Has Entered the Meeting

Ivy: We'll see. Okay, next up: we're moving from optimizing your machine... to optimizing your meetings.

Marcus: Right! Anthropic is putting Claude directly into Google Meet. It joins the call just like a person, its little icon pops up in the grid, and it gets to work.

Ivy: It 'joins like a person'. That's not creepy at all. What does it do?

Marcus: It takes notes, live! It can summarize discussions, track action items, and even help build documents or whiteboards based on the conversation, all in real-time.

Ivy: So it's listening to everything. Every side comment, every hesitation. Does this change how people talk? Do you feel like you can't speak freely because the corporate AI is taking permanent notes?

Marcus: I think the productivity gain outweighs the awkwardness. Imagine never having to write meeting minutes again. Ever.

Ivy: But it's more than a transcriber, right? This is about redefining 'collaboration.' When the AI is a 'participant,' it's not just a tool anymore. It's part of the social dynamic.

Marcus: An actor that can instantly tell you what was decided five minutes ago when you zoned out. I'm calling that a win.

Ivy: Until your boss asks it to summarize who was most and least engaged. The takeaway here is, you need to be really aware of when these AI agents are listening and what your company's policy is. The social rules are being written in real-time.

Nvidia Allegedly Engineered Scarcity by Funding AI Startups to Buy Its GPUs

Marcus: [sighs] Always the dystopian angle. Okay, let's talk about that bombshell you dropped at the top of the show.

Ivy: Right. A report from TechPowerUp alleges that Nvidia has been propping up its own sales figures in a... let's call it a creative way.

Marcus: This sounds wild. So Nvidia's venture capital arm invests in all these hot AI startups...

Ivy: Right. But the allegation is that a significant portion of that investment came with the explicit or implicit condition that the startups had to spend it on Nvidia GPUs.

Marcus: Wait—so Nvidia gives a startup 100 million dollars...

Ivy: ...and the startup immediately gives 50 million of it back to Nvidia in exchange for a mountain of H100s. The money just makes a round trip.

Marcus: Whoa.

Ivy: Exactly. So Nvidia gets to book that $50 million as revenue. It makes their sales look stronger, and it creates massive demand for their chips, which drives up the price and creates the 'scarcity' everyone's been complaining about.

Marcus: So they're not just selling shovels in a gold rush, they're lending prospectors money to buy their shovels, making it look like there's an even bigger gold rush than there is, and making the shovels more expensive for everyone else.

Ivy: That's a perfect analogy. And it's borderline market manipulation. You're creating artificial demand to justify your prices and squeeze out the competition. This isn't just clever business; it feels fundamentally dishonest.

Marcus: So, if you're a builder or researcher who's been tearing your hair out trying to get a GPU for the last two years... this might be why. If this is true, the scheme ate up a massive amount of supply that should have been on the open market.

Ivy: And it means regulators are probably going to be taking a very, very close look at Nvidia's books. This story is just getting started.

Marcus: Alright, quick recap. AMD's new hrx-loom tool promises faster local AI, but you'll need next-gen hardware to see it.

Ivy: AI is coming to your meetings, forcing a trade-off between productivity and privacy.

Marcus: And Nvidia is under fire for allegedly funding startups just to have them buy back its own GPUs, creating a cycle of artificial demand.

Ivy: Before we go, I saw an AI recipe generator making the rounds that suggested a... 'deconstructed pumpkin spice latte salad'.

Marcus: [laughs] Does that come with an espresso bean vinaigrette? What's even in that? Lettuce and... foam?

Ivy: I didn't dare look at the details. Some things are better left un-generated.

Marcus: Fair enough. That is our show! Thanks for listening to Broke It.

Ivy: We'll be back tomorrow. In the meantime, don't accept any venture capital that comes with a shopping list.

This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.