The AI Companies Are Still Spending More Than They Earn, and Introductory Prices Are Starting to End. Use the Cheap Window to Grow Yourself, Not Just Your Output.
The AI Companies Are Still Spending More Than They Earn, and Introductory Prices Are Starting to End. Use the Cheap Window to Grow Yourself, Not Just Your Output.
Why today's AI prices probably aren't permanent, how I'm using them to learn things I couldn't have learned this fast before, and a simple 30-day plan if you haven't started.
Here's a fact that doesn't get enough attention: the AI tools most of us now use every day are, for the companies building them, still very expensive to provide.
Leaked audited figures reported in June put OpenAI's 2025 operating loss at about US$20.9 billion on revenue of about US$13.1 billion. Its net loss was larger again because of a one-off accounting entry. Anthropic reportedly told investors this month that it made an adjusted profit last quarter, on a measure that leaves out some costs and hasn't been published. And the frontier labs' planned spending on computing power runs far ahead of what they earn today. OpenAI alone has reportedly pledged around US$600 billion on infrastructure through 2030.
Businesses that spend more than they earn have to improve the economics somehow: price, efficiency, scale or product mix. Price is only one of those levers, but we're starting to see it pulled in places.
|
2×
Google's published API price for Gemini 3.8 Flash from 1 January 2027, when its introductory rate ends: US$0.75 → US$1.50 per million input tokens
|
~US$20.9bn
OpenAI's reported 2025 operating loss, on revenue of about US$13.1bn (leaked audited figures, reported June 2026)
|
The introductory period is a real thing
Google's own Gemini API pricing page now shows end dates on several current prices, including two headline models. Gemini 4 Argon doubles on 1 December, as we noted in the 17 October news wrap. Gemini 3.8 Flash, one of its cheaper workhorse models, also doubles from 1 January 2027. That covers input, output and caching. Consumer subscriptions haven't moved yet, and this isn't a prediction that your monthly plan will double next year. Some prices will keep falling as competition bites.
But the direction of travel is worth noticing. Launch prices are being set to win users, with an end date attached. For the companies to cover their costs, some of today's generosity will eventually go. That's speculation on my part, not a forecast. The point is that nobody promised us today's prices forever.
For a finance team, that has a practical side, which we've covered in the arithmetic of cost per accepted output. Three quick checks are worth doing before your next budget or reforecast:
Know which workflows sit on introductory pricing. If an automation or integration your team relies on runs on a model with a published end date, put that date in your budget calendar the same way you would a lease review.
Run the budget line at double. For AI subscriptions and usage-based costs, model a scenario where unit prices double — as a stress test, not a forecast. If the business case only works at today's price, you want to know that now.
Don't build on a single model you can't swap. Where you can, keep prompts, rules and checks in your own documents rather than buried inside one vendor's setup. Then a price change becomes a decision, not a crisis.
But I want to spend the rest of this post on the personal side, because I think it matters more.
What I'm doing with the window
I'm a CPA with more than twenty years in finance, and I'm not a programmer. In the last year, AI has let me learn and build things that would have taken me years, or that I'd simply never have attempted. In the spirit of honesty, here's what that actually looks like:
A daily review of my own share portfolio. I'm not asking AI what to buy. I use it to work through why something moved, test my reasoning, and learn the parts of investment analysis I never studied. It's a tutor that never gets tired of my questions.
A game. Yes, really. I'm building one, mostly to learn how software fits together, and partly because it's fun. Every problem I solve there teaches me something I later use at work.
Tools for my own work. I'm testing and building small tools that make my day faster: checks, reports, automations. Some will become part of how we work at PFL. Most are early stages and not ready to talk about in detail yet.
Soft skills. I've worked with AI to plan how to strengthen the softer skills I know I'm weaker at, and built a simple app to keep myself studying regularly.
This blog. I write the first drafts, and they're rough. Sometimes the language is something I'd never put in print. AI helps me turn them into something clean and readable. The ideas, the opinions and the final call are mine. The polish is a collaboration.
The common thread isn't productivity. It's that I'm learning faster than at any time in my career, and the cost of that learning is, for now, a monthly subscription.
Output versus growth
Most workplace AI use I see is about output: write this email faster, summarise this document, draft this commentary. That's useful, and it's what the tools are sold on.
But output gains are easy for everyone to copy. When everyone in your field writes emails twice as fast, nobody is ahead. Growth is different. If you use these tools to understand a new area, build something you couldn't build before, or close a skill gap, that stays with you. It doesn't disappear if prices rise or a tool changes. A finance officer who uses this year to learn how to build a simple automation, or to really understand the funding rules they process every day, will still have that skill when the subscription costs twice as much.
As I wrote on Tuesday, the people who do well with AI use it to do more of their own thinking, not less. The cheap window is the best time to build that habit.
A 30-day plan if you haven't started
If you only use AI to tidy emails, here's a simple way to try the other kind of use:
1. Pick one thing. Something you've always meant to learn or build: a skill gap from your last review, a spreadsheet you rebuild every month, a personal project. One thing, not five.
2. Give it 20 minutes a day. Ask AI to set up a learning plan with you, then work through it. Ask it to quiz you. Ask it to explain your mistakes.
3. Make something by day 30. A working tool, a summary you could teach a colleague, or a small project that runs. Learning sticks when it produces something.
4. Check what it tells you. AI is a confident tutor and sometimes a wrong one. Test anything important against a real source, the same way you would at work.
Then tell someone what you built. You'll be surprised how many colleagues want to know how you did it.
So, how are you using it?
I've shared mine. I'd like to hear yours: the useful, the odd and the ones that didn't work. The tools are cheap now and getting better quickly. That combination may not last, so I'd make the most of it while it does.
Is your finance team using AI to grow, or only to go faster?
Speed gains are easy to copy. Capability lasts. PFL provides senior-level outsourced finance, management reporting, and AI automation for Australian NFP, NDIS, and SME organisations, including helping small finance teams build AI skills and tools that stay useful whatever happens to the price list.
Talk to PFL →
Comments
Post a Comment