Rafomac News, Tech & Trends That MatterOpenAI has been bleeding billions even as its user base and revenue soared. But in just 90 days, the company made four game-changing moves that could transform the money-losing AI giant into a tech powerhouse.
OpenAI’s money problem: Why growth has meant losses
OpenAI is on track to lose $33 billion this year, despite revenue nearly doubling to a $25 billion run rate. The irony is skin-deep: their best customers—the most active paying users—are in fact the most costly to serve. Sam Altman himself admitted that even $200-a-month subscribers don’t cover their usage costs.
Why? The root of OpenAI’s losses lies in inference costs. Training an AI model is costly but a one-time event—like sending a chef to culinary school. But inference—the repeated running of the model every time a user interacts—is a continuous expense. Every single ChatGPT response requires GPUs firing up to process queries, and with 900 million active users, that adds up to a staggering $14 billion in inference costs alone for 2026.
Tanisha and the paradox of heavy users
Consider Tanisha, a software engineer in Bangalore paying $19.99 monthly for ChatGPT Plus. She writes code, drafts emails, plans trips—uses it so much that her monthly queries cost OpenAI about $30 in compute power. That’s a $10 loss just on one subscriber each month. Multiply by millions and the math reveals why OpenAI can’t profit despite booming revenues.
Simply charging Tanisha $40 instead of $20 sounds like an easy fix, but competition is fierce. Other AI giants like Google and Anthropic are slashing prices aggressively. If OpenAI raises fees, users will quickly switch, and so OpenAI faces a squeeze it can’t escape.
Breaking free from the rent trap: Owning the AI stack
OpenAI has long rented critical parts of its technology stack—chips from Nvidia, cloud infrastructure from Microsoft, energy from power companies. This is the “landlord cost” problem. Taking Jensen Huang’s layered cake analogy, OpenAI only owned the models layer and rented everything else. That drove costs sky-high and left OpenAI vulnerable to price pressures.
To fix this, OpenAI needed its own oven. And it got one: a custom inference-only chip codenamed jalapeno, built in partnership with Broadcom. Early tests suggest jalapeno delivers inference costs roughly 50% lower than Nvidia’s best GPUs, matching performance but costing half as much. This chip promises to cut that $14 billion annual inference bill in half in the coming years, a giant leap toward profitability.
The power trio: GPT-5.6, Codex, and jalapeno
OpenAI’s sprint over the last 90 days has been remarkable. They launched GPT-5.6—code-named Soul, Terra, and Luna—which matches or outperforms competitors like Anthropic’s Claude Fable 5 but costs about one-third as much per task. Soul, for example, costs $1.86 per task compared to Claude’s $3.15, with Luna dropping as low as seven cents.
This sharp drop in cost has shaken up the market, prompting rivals to rethink their pricing strategies to stay competitive. Meanwhile, Codex, OpenAI’s AI that writes code and runs multi-step software tasks autonomously, saw a 500% spike in active users. Codex promises to do more than simple Q&A—it lets anyone build software through natural language prompts, turning users into creators.
When every Codex task costs roughly 100 times more than a simple ChatGPT query, OpenAI’s cost problem deeply intensifies. But with jalapeno slashing inference expenses, OpenAI can deliver this power without bleeding money.
What’s next for OpenAI—and the AI gold rush?
While cheaper AI should save money, history suggests otherwise. The Jevons paradox says that when an input becomes cheaper, it’s used far more extensively. As AI access becomes effectively free, demand and use will explode even further, pushing OpenAI to repeatedly find new ways to cut costs.
This relentless drive to lower AI’s cost barrier could spark a new wave of innovation, empowering billions to build and invent with minimal friction. As the infinite canvas of AI opens up, the only question is what you’ll create when thinking and building become limitless.
