Washington goes full stack

Plus: Nvidia weighs an expansion

December 17, 2025


Welcome back! Merriam-Webster just crowned its 2025 Word of the Year. It’s not “agentic,” “multimodal,” or “alignment.” It’s…“slop.” That includes AI slop, the low-quality, mass-produced content you didn’t ask for, didn’t enjoy, and definitely didn’t bookmark (yet somehow saw five times today).

It’s the perfect word for 2025, IMO. This has certainly been the year that slop flooded our feeds and split platforms into two camps: those trying to fight the slop (Wikipedia, YouTube, Spotify) and those leaning all the way in (Meta, OpenAI, even Disney).

Inside the US Tech Force

Via Chip Somodevilla | Getty Images

The Trump administration is launching the US Tech Force, a program that will recruit roughly 1,000 technology specialists—many from Big Tech—to accelerate the government’s use of AI. The program places technologists across agencies for up to two years, after which they can transition back into industry or stay in government.

What the Tech Force will do:

  • Pull from Big Tech and beyond. Participants may come from companies like Amazon, Apple, Google, Meta, Microsoft, OpenAI, Nvidia, Adobe, Uber, xAI, and Zoom, alongside early-career technologists.

  • Push AI adoption inside government. The mandate includes developing AI-powered tools, modernizing data infrastructure, and building new apps for federal agencies.

  • Operate as a fast-moving unit. The structure echoes the Obama-era US Digital Service but with a stronger emphasis on AI deployment and private-sector partnerships.

The bigger picture: Earlier this year, the administration dismantled much of the government’s in-house tech capacity, folding the US Digital Service into DOGE and cutting staff. Now, Washington is re-entering the tech talent market—and doing so explicitly through AI. 

Nvidia Weighs Expanding H200 Chip Production for China

Nvidia is reportedly considering ramping up production of its H200 AI chips after receiving US approval to sell them in China—and seeing demand surge almost immediately. Chinese companies are already lining up to place large orders as soon as imports are formally cleared.

The backstory: Until recently, Nvidia’s H200 chips were off limits to China under US export controls. That changed last week, when the Department of Commerce approved H200 sales to China under a licensing framework that gives the US government a 25% cut of those sales. 

  • For Nvidia, expanding production would let it tap into demand as China races to develop its own chips. 

  • Chinese firms like Alibaba and ByteDance, both building large AI models, are reportedly exploring major H200 purchases, even as Beijing weighs final import approvals.

Why it matters: Demand for advanced AI chips remains extreme—and China is still one of the largest markets in the world, even under restrictions. If Nvidia scales H200 production without constraining US supply, it reinforces how central the company remains to global AI infrastructure, despite years of export controls meant to curb China’s access. 

Disney’s OpenAI Deal Comes With a One-Year Clock

New info is emerging about Disney’s three-year licensing partnership with OpenAI. Turns out, the deal includes just one year of exclusivity, Disney CEO Bob Iger confirmed. After that, Disney is free to license its characters to other AI companies building video models.

What the deal includes: OpenAI gets legal access to more than 200 characters from Disney, Pixar, Marvel, and Star Wars for use inside Sora, making it the only AI platform currently authorized to generate content with that IP base. For Disney, the partnership offers a low-risk way to test generative video without fully committing its catalog to AI distribution.

The bigger picture: Disney isn’t betting on any single AI company—it’s betting on leverage. By limiting exclusivity to one year, Disney keeps control over its IP while AI platforms compete for access.

Nebius Token Factory—Post-training

Nebius Token Factory just launched Post-training—the missing layer for teams building production-grade AI on open-source models.

You can now fine-tune frontier models like DeepSeek V3, GPT-OSS 20B & 120B, and Qwen3 Coder across multi-node GPU clusters with stability up to 131k context. Models become deeply adapted to your domain, your tone, your structure, your workflows.

Deployment is one click: dedicated endpoints, SLAs, and zero-retention privacy. And for launch, fine-tuning GPT-OSS 20B & 120B (Full FT + LoRA FT) is free until January 9. This is the shift from generic base models to custom production engines.

A firewall for AI prompts

Via PromptGuard

PromptGuard inspects and sanitizes prompts and surrounding context in real time to block prompt injection, redact PII, and prevent data leaks before requests hit your LLM. It works across models with minimal latency and policy controls designed for production environments.

How you can use it:

  • Protect LLM apps from prompt injection and jailbreak attempts.

  • Redact sensitive data before prompts reach the model.

  • Enforce prompt governance and compliance at scale.

  • Monitor prompt behavior with logs and analytics.

Pricing: Free and paid plans available

GTM AI-native operating system

Via GrowthOS

GrowthOS unifies product, web, and CRM data into a single platform that powers real-time segmentation, personalization, and automated customer journeys.

How you can use it:

  • Build real-time segments across web, app, and CRM data.

  • Automate personalized, omnichannel campaigns.

  • Prioritize high-value users with predictive scoring.

Pricing: Paid

Jobs, announcements, and big ideas

  • Databricks is raising $4 billion at a $134 valuation.

  • Marketing is a game. We help you play it smarter. Masters in Marketing is your cheat code for fresh insights, expert advice, and strategies that actually work. Subscribe today.*

  • Microsoft releases TRELLIS 2, a high-fidelity image-to-3D generative model designed for detailed 3D asset creation.

  • DoorDash launches Zesty, a social discovery app that uses AI to help users find and share new restaurants.

  • Meta introduces SAM Audio, a multimodal AI model that separates sounds using text prompts or visual context.

  • Joseph Gordon-Levitt launches a creators-led AI coalition aimed at pushing back against unethical AI practices.

  • Google Labs rolls out CC, an AI productivity agent that delivers personalized Gmail briefings and assistance.

  • OpenAI releases GPT-Image-1.5, a faster and more precise image generation model for ChatGPT users and API access.

*This is sponsored advertising content.

Nano Banana finally has competition. Here’s what changed with GPT-Image-1.5 and why it matters for AI image editing.

That’s a wrap! See you Friday.

—Matt (FutureTools.io)