You want AI that respects your privacy. You want it there when you need it, gone when you don’t. You want it free, or at least cheap. Here’s the problem: that combination doesn’t exist. It can’t. And Kagi — the indie search engine that actually tried to build it — just hit the wall.
Last week, Kagi announced something quietly devastating buried in a changelog. Their translation feature and news tools, both powered by AI, had become unexpected hits. Users loved them. Usage exploded. And then Kagi had to pull the plug on free translations because the compute costs were eating them alive.
If an AI tool is free, you are the product. If it’s not free, you still might be. The only honest AI is the one you pay for — and pay enough that it hurts.
Let’s be clear about what happened here. Kagi isn’t some ad-funded surveillance machine. They’re a user-funded, privacy-first search company that charges for their product. They built AI translation as a free perk. People used it. A lot. The underlying compute costs — the GPUs, the inference, the infrastructure that makes modern AI actually work — scaled faster than any subscription model could absorb. So they cut it.
This isn’t a failure of engineering. It’s a failure of a dream. The dream that says: we can have powerful, private, opt-in AI without sacrificing anything. We can’t.
Think about what Google does. Google gives you free translation, free search, free AI summaries — and harvests every interaction to feed their ad targeting engine. The compute costs are astronomical, but they’re subsidized by your data. Your behavior. Your identity, sliced and sold. That’s the trade. It’s always been the trade.
Every free AI feature you’ve ever used was paid for by something you didn’t agree to sell.
Kagi refused that trade. They said: we won’t track you, we won’t sell you, we’ll just charge you directly. And when they tried to extend that philosophy to AI-powered tools, the math broke. Because AI compute doesn’t scale like traditional software. Every translation, every summary, every query hits a GPU that costs real money per token. There’s no economies-of-scale fairy that makes inference free. The more people use it, the more it costs — linearly, sometimes worse.
So Kagi did the honest thing. They temporarily removed free translations. They kept access to original-language articles. They didn’t sugarcoat it in corporate PR speak. They said: this surprised us, the costs spiked, we had to act.
When was the last time a tech company told you the truth about why a feature disappeared?
Here’s what this means for you. If you care about privacy — really care, not just in a performative I-use-DuckDuckGo-once-a-week way — you need to accept that the future of private AI is paywalled. Strictly. Aggressively. Not as a freemium tease, but as a hard gate. Because the economics don’t bend for good intentions.
Generosity in AI isn’t a strategy. It’s a countdown timer.
The companies offering you free AI right now are doing one of three things: burning venture capital they’ll never recover, harvesting your data to compensate, or planning to rug-pull you the moment usage scales. There is no fourth option. Kagi just proved that even the most well-intentioned player hits this wall.
The next time you see a free AI tool that promises privacy and power, ask yourself one question: who’s paying for the GPUs? If the answer isn’t you, it’s someone who expects something in return. And in the AI economy, what they expect is almost always you.
Kagi didn’t fail. They ran the experiment we all needed. They showed us that the dream of free, private, powerful AI is a contradiction in terms. The rest of the industry is still pretending otherwise — until their investors stop subsidizing the fantasy.
The most valuable thing Kagi gave us wasn’t a translation tool. It was the proof that free AI was never free at all.
FAQ
Q: Isn't Kagi just badly managed? Other companies offer free AI fine.
A: No. Other companies offer 'free' AI by selling your data or burning VC money. Kagi refuses to sell user data and is user-funded, so when compute costs spike, there's no hidden subsidy to absorb it. They're not badly managed — they're honestly managed, which means they hit economic reality faster.
Q: So I just have to pay for everything AI-related now?
A: If you want it private, yes. The compute cost per AI interaction is real and non-negotiable. Subscription models that charge enough to cover actual usage are the only sustainable path for privacy-respecting AI. Free tiers will always come with a hidden cost — your data.
Q: Isn't this just Kagi being greedy by removing a free feature?
A: Greedy would be keeping the feature free while quietly harvesting user data to cover costs — which is what most tech companies do. Kagi chose transparency and cut the feature instead. That's the opposite of greedy. It's the rare case of a company choosing honesty over growth theater.