Expensive AI is coming


Expensive AI is coming

The world changed rapidly in November 2022 when ChatGPT launched as a free research tool. It reached 1 million users in five days and 57 million users by the end of its first month. This astounding growth came through a traditional Silicon Valley approach: the freemium model. The free tier gave people access to an unprecedented experience while placing certain limits on usage. In February 2023, users could remove those limits by paying a monthly fee. The goal was straightforward: grow the user base in hopes that some percentage would become paying customers and fund further growth.

Since then, we've grown accustomed to generous free tiers of the latest AI models from all companies. Anthropic offers a free tier of Claude, X allows free usage of Grok, and Google allows anyone to use Gemini for free. Products built on top of AI models also offer free access to powerful functionality, including Jules, Stitch, AI Studio, and NotebookLM from Google, v0 from Vercel, Canva AI, Firefly by Adobe, GitHub Copilot, Replit, Perplexity, and Figma Make. Every company with an AI product is fighting for market share, so free options are plentiful. However, that is changing.

We are now past the honeymoon phase of AI. Three years in, companies still aren't making money on AI.

In 2025, companies were still taking losses on their AI offerings to capture market share. Instead of raising prices on AI directly, tech companies looked for other ways to save money, most notably through mass layoffs and increasing prices on other services. Three years into the AI revolution, companies are tired of losing money and are seeking ways to recoup the massive investments they've made and cover ongoing expenses.

With neither OpenAI nor Anthropic profitable and Nvidia and AMD planning to increase GPU prices, 2026 could be the breaking point for companies no longer willing to wait for profitability from winning the AI race. What can you do to prepare.

  1. If you're depending on free AI services, expect them to be dramatically reduced or eliminated in 2026.
  2. If you're depending on AI APIs, expect costs to at least double in the short term and potentially increase 5-10x in the long term.
  3. If you're using paid AI services, expect prices to increase and offerings to be reduced in an effort to push you toward more expensive plans.

The AI free ride won't last forever, and signs point to it significantly slowing or disappearing in 2026. Make sure you're prepared.

Key Takeaways

  • The era of free AI is ending. After three years of companies bleeding money to capture market share, 2026 is shaping up to be the year when free tiers get slashed and API costs skyrocket.
  • Every major AI company is losing money. From GitHub losing $20 per user to OpenAI losing money on $200/month subscriptions, the math simply doesn't work, and companies are done waiting for profitability.
  • Prepare for dramatic price increases across the board. Whether you're using free services, APIs, or paid plans, expect costs to double at minimum and potentially increase 5-10x as companies try to recoup their massive investments.

New Post: How GitHub could secure npm

2025 saw a rise in sophisticated open source supply chain attacks on npm. GitHub's response has been muted. Here's what they could do to more effectively secure npm.


Stuff I've Enjoyed this Month

📝 Using service auth for Gitea access via Git Client + Cloudflare Tunnel by Chatri Ngambenchawong
I've been trying to figure out how to access my Gitea server (running on my NAS) securely away from home. This article walks through setting up a Cloudflare Tunnel to allow external access to local resources while ensuring the Git CLI works as expected.

📖 Limitless by Jim Kwik
This book is like a hacking guide for your brain. The author teaches you how to easily remember large amounts of information, speed read, and more.

📝 How to compile JavaScript to C with Static Hermes by Devon Govett
Walks through how Facebook's Static Hermes compiles JavaScript libraries into C so they can be used in Rust.

🎬 Real warp drive doesn't need negative energy by NASA Space News
Reality keeps catching up with science fiction. Warp drive was long thought to require negative energy, but recently, scientists discovered that it doesn't. This is an important step (but small) step towards making warp drive a reality.

📖 Under the Bridge by Rebecca Godfrey
What happens when a novelist decides to investigate the murder of teenage girl by other teenagers? A gripping, real-life story written in a way that kept me reading for hours on end.

📝 How to write a great agents.md: Lessons from over 2,500 repositories by Matt Nigh
We're at the point where every project should have at least one agents.md file to help AI work in the codebase. This blog post highlights the most effective agents.md patterns.

🛌 Memory Foam Pillow by Royal Therapy
I was staying at my parents' house for the holidays and found the guest room pillow to give me neck pain. Searching for a quick fix, I came across this pillow and it's fantastic. A bit softer than I would like but very supportive and without the typical awful memory foam smell.


What I'm Working On

🏠 Real Estate: One of my property managers unexpectedly handed in his notice. Fortunately, I had another person available to take over those properties. Follow my Instagram for real estate photos.

🍞 Bredbox: Bredbox now supports highlights on saved articles, mimicking the functionality from Pocket. At this point, Bredbox is basically feature complete with the Pocket web application for your saved URLs. I'm now spending some time improving the article extraction logic and will then focus on publishing a REST API for you to have complete control over your saves.

💻 ESLint:

  • We are continuing to publish v10.0.0 prereleases. Make sure to try it while there's still time to incorporate feedback: `npm install eslint@next`
  • It was a bit of a slow month as month of the team took time off for the holidays.

Coaching for Tech Leads and Staff+ Engineers

If you're a tech lead or staff+ engineer, chances are your manager has no experience in your role. While they may be well-intentioned, they may not know how to help you grow in such a challenging position.

That's where working with me can help. As a former tech lead and principal engineer at multiple companies, I learned firsthand the ins and outs of these roles. I work with my clients one-on-one to develop their leadership, communication, organization, and problem-solving skills to succeed in the upper ranks of the individual contributor track.

Human Who Codes Newsletter

A once-per-month newsletter discussing topics important to senior-level software engineers, with a particular focus on frontend technology and leadership.

Read more from Human Who Codes Newsletter

Static Analysis and AI If you’re like me, the way you write code has changed significantly in the past year. December 2024 was the first time I decided to use AI to write code, and today it generates most of the code I work on. It also writes the tests and documentation. My job is now focused on reviewing code, and that can get tiring. How can I be confident that the code meets my standards, remains secure, and scales well? The answer is static analysis. Static analysis tools inspect code...

Working with Legacy Code The term “legacy code” often strikes fear into the hearts of software engineers. While there’s no single definition, IBM describes it as “code that still serves its purpose but was developed using now outdated technologies.” Any successful tech company has some amount of legacy code because that code worked well enough to attract and keep customers. When something is working, especially if it’s difficult to change, the logical choice is often to let it keep running...

npm Security For the past few weeks, an ongoing supply chain attack on npm packages has shaken the JavaScript community. It’s not that npm packages haven’t been compromised before, but this time, the targets were surprising. Attackers went after popular packages maintained by well-known authors, creating the potential for widespread damage. In late August, the popular monorepo tool Nx was compromised, leading to several malicious package versions. These included a postinstall script that...