🍭 Nvidia Lobbied For Its Own Rivals 🎯

Good morning. Somewhere in Washington a policy aide is reading the phrase "premature restrictions on downloadable AI models" for the ninth time.

They mean China, right?

Let’s dive in 👇

🍭 What’s Cookin’:

• Nvidia rounded up 24 friends to defend models it doesn't make

• Anthropic shipped Opus 5 at half of Fable's sticker price

• A Delhi judge told ANI that training on its articles is fair dealing

Nvidia's
📜 Lobbying For Its Own Rivals

The Bite:

Nvidia published a three-page policy letter titled "Open Weights and American AI Leadership" on July 24.

Twenty-five companies signed on, including Microsoft, Meta, IBM, Dell, Palantir, Hugging Face, Mistral, and the Linux Foundation.

The letter asks Washington to avoid what it calls premature restrictions on downloadable AI models.

It also urges policymakers not to treat distillation, the practice of training one model on another's outputs, as misappropriation.

OpenAI, Anthropic, and Google did not sign.

Chief Executive Jensen Huang promoted the letter in his first-ever post on X.

Snacks:

• The letter runs three pages and never names China, Moonshot AI, or DeepSeek.

• Signatories include Dell, ServiceNow, CrowdStrike, Box, Replit, Perplexity, Andreessen Horowitz, and Y Combinator.

• Every model developer on the list already publishes weights: Meta, Mistral, Black Forest Labs, Arcee AI, Reflection.

• Huang said at CES 2026 that one in four tokens generated today comes from an open model.

• It landed four days after reports the administration was reviving a push to ban Chinese models.

• Treasury Secretary Scott Bessent said this week that officials found US model watermarks in Chinese systems.

Why it Bites:

First, read the signatory list as a market map.

Chipmakers, server vendors, clouds, security firms, venture funds, and five labs that already give their weights away.

Not one company on it sells access to a closed frontier model.

Open weights have to get served somewhere.
They run on enterprise clusters, regional clouds, and racks owned by buyers with no in-house TPU program (Google’s accelerator for machine learning) and no Trainium roadmap (the AWS version).

Every one of those racks is a Nvidia sale.

In comparison, closed frontier models run on a handful of hyperscaler endpoints, and those hyperscalers are busy designing their own silicon.

So we can say the open ecosystem is a distribution channel.
Which is why the distillation paragraph is the tell.

Nothing in a letter about downloadable weights requires a defense of training on another model's outputs, unless you expect Washington to start calling that theft, and unless the labs most exposed to the charge are the ones building the open models you need people to keep downloading.

If you ban Chinese open weights, China keeps shipping.
What stops is the American enterprise running Kimi on a rack in Ohio.

Watch who ends up protected when the rule actually lands.

Steal This Prompt
🎭 Shadow Puppet Theatre

Turn any idea into a dramatic shadow-puppet performance that looks like it escaped from a centuries-old theater.

The prompt is built for cinematic image and video generation with a theatrical silhouette aesthetic.

Use it to:

• Reimagine movie scenes as shadow puppet performances

• Turn fantasy creatures into eerie silhouette art

• Create thumb-stopping social posts with a timeless aesthetic

Workflow:

1. Hit this link: Shadow Puppet Theatre

2. Paste into your AI model

3. Replace the #s with your scene, characters, or story

4. Watch it stage a puppet show your art teacher would've framed

   Try the prompt on Snack Prompt

_ToolBox™
🧰 5 BRAND NEW AI LAUNCHES

💬 Comms

Puts a live agent on a real iMessage line in about 30 seconds, instead of the carrier paperwork and $225 a month it used to take.

📊 Artifacts by Databox

Turns one question to your AI analyst into a shareable report or deck built off live data, not a stale export.

🔐 Rivault

Holds your passport and card details in a zero-knowledge vault, hands them to an agent mid-task on a Face ID unlock, then redacts them when it's done.

🎨 Repaint Socials

Points at your Google Business or Instagram page and builds a real website out of the content you already posted.

🔍 Webhound

Research agents with a depth dial, so you decide whether you want a quick answer or a fully cited dataset.

_{Everything Else}
🧠 You Need to Know

✨ Anthropic Ships Claude Opus 5 At Half Price
→ Opus 5 launched July 24 at $5 and $25 per million tokens, becoming the default model on Claude Max and the strongest option on Claude Pro.

📜 Nvidia Leads 25-Company Letter Against Open Weight Limits
→ Nvidia, Microsoft, Meta, IBM, and Palantir signed a July 24 letter opposing broad restrictions on downloadable model weights, which OpenAI, Anthropic, and Google did not join.

⚖️ Delhi High Court Denies ANI Injunction Against OpenAI
→ Justice Amit Bansal ruled July 24 that training ChatGPT on Asian News International articles falls under the Copyright Act's fair dealing exception, refusing interim relief.

📱 Meta Brings Muse Spark 1.1 To Meta AI
→ Meta switched its consumer chatbot to the 1-million-token Muse Spark 1.1 on July 24 in a limited set of markets, with WhatsApp support planned within weeks.

💽 AMD Puts Helios AI Rack Into Full Production
→ AMD said its 72-GPU Instinct MI455X rack entered full production July 23, with Anthropic, OpenAI, and Meta executives detailing deployments from the keynote stage.