May 4, 2026

3 Minute Read or Listen ->

🔱 Sign up for a free demo today!

See how the latest AI and digital advocacy tools can transform your work.

THIS WEEK'S EDITION

AI Subsidies Coming To A Screeching Halt

Remember when an Uber ride across town cost $5? Then those same rides became $25, surge pricing kicked in, and you started checking the fare before tapping "request." That same arc is now hitting AI. The subsidies are ending, prices are going up, and public affairs teams that built workflows during the all-you-can-eat era are about to find out what they actually cost. In today's short read, we'll show you what's coming and how to prepare.

GitHub Just Pulled The First Big Plug

On April 20, GitHub — Microsoft's coding platform used by tens of millions of developers — stopped letting new customers sign up for its paid Copilot plans entirely. A week later, the company announced that on June 1, 2026, every Copilot subscription moves from "unlimited" to a token-metered system tied to actual usage. GitHub's product VP put the math plainly: a single user's heavy session now routinely racks up more cost than the user pays for the entire month.

GitHub isn't an outlier. Cursor ended its unlimited fast-requests tier last summer. Anthropic added weekly limits to Claude Code. OpenAI, by Sam Altman's own admission, has been losing money on Pro subscribers. Per Ars Technica's reporting on leaked GitHub documents, Copilot's costs had nearly doubled week-over-week since January.

If You Don't Adapt, Your AI Bill Is About To Skyrocket

The heavy AI users — anyone running agentic workflows, multi-step research, long context windows, or top-tier models — are about to pay real money for what they got nearly free during the subsidy era. Light users (basic chat, occasional drafts) may be largely unaffected for a bit longer. But for advocacy teams that have built AI into research, drafting, ad production, and outreach, the bill is coming.

Two forces broke the flat-rate model. First, agentic AI: a single autonomous AI session can now burn through more compute in an hour than a flat-rate plan was designed to cover in a month. Second, while per-token costs keep falling, power users are consuming exponentially more tokens — and the light users were quietly subsidizing them. There is simply not enough compute on the planet to meet all of this soaring demand. The math on subsidies finally caught up.

But there are ways you can adapt now to keep a lid on the price hikes.

How To Contain The Coming Price Hikes

🔱 Right-size your models. Not every task needs the most powerful model on the planet. The flagship models from OpenAI, Anthropic, and Google are often 10-20x more expensive than their smaller siblings — but for most routine work (drafting, summarization, classification), the cheaper models are more than capable. Most teams default to the top tier for everything because that was free during the subsidy era. That habit just got expensive.

🔱 Find vendors building on open-source models. A growing field of open-source models are now competitive with closed-source flagships for many tasks — and free for vendors to deploy, if they know how. Smart software vendors are quietly switching their backends from using OpenAI and Anthropic for every task, to fine-tuned open-source models running on their own infrastructure, that deploy the frontier models only when needed. You won't notice a quality difference. You will notice that unlike other vendors, they're not raising your prices.

🔱 Ask about local deployment. For sensitive workflows — opposition research, internal strategy memos, client communications — there's now a real alternative: AI that runs entirely on your vendor's own hardware. No data leaving the building. No per-token meter running. The cloud-vs-local quality gap has narrowed dramatically. 

Two Ways to Operationalize This Email

🔱  Check your current AI workflows and test lower tier models instead of using frontier models for every task.

🔱  Find AI vendors who are ahead of the curve on using open source and local models to contain AI costs. Test them on the difference between this approach and a purely frontier-model based approach.

🔱 If you're enjoying this content, please consider forwarding this email to a colleague or friend. 

🔱 If you're not already a subscriber, please sign up here to stay up to date on the latest developments in political technology.

🔱 If you have suggestions for future newsletters or any other feedback, we'd love to hear from you.