DEV Community: Jérôme Corbiau

The Hidden Cost of Cloud Dictation: Why You Pay Three Times

Jérôme Corbiau — Tue, 02 Jun 2026 07:30:33 +0000

The Apparent Price: A "Modest" Subscription

**
Fifteen dollars a month. That is the price displayed by most intelligent voice dictation services. A sum that seems reasonable, almost trivial. Less than a restaurant meal, less than a streaming subscription. A small line in the budget, easily justified by the promised time savings.
But this price is only the tip of a colossal economic iceberg. Beneath the surface, multiple costs accumulate, invisible, insidious, that rarely does anyone take the time to calculate. The real cost of cloud dictation is not monetary. It is structural, temporal, and fundamentally personal.
Let us start with the simple calculation. Fifteen dollars monthly is one hundred and eighty dollars annually. Over ten years — the minimum duration of a professional career — that is one thousand eight hundred dollars. And that does not account for inevitable price increases, upgrades to higher tiers to unlock essential features, overage fees during busy months. An active professional user can easily reach three to five thousand dollars over a decade, just for voice dictation.
But the monetary cost is only the first level.

The Second Cost: Your Biometric Data

**
Your voice is a sonic fingerprint. It carries your accent, your cadence, your timbre, your emotional state, your speech habits, your professional vocabulary, your verbal tics, your hesitations when you lie. It is a biometric portrait richer and more intimate than most data you voluntarily share on social media.
When you dictate via a cloud service, this fingerprint does not simply transit to a server to be transcribed. It is stored. Analyzed. Aggregated with millions of other voices to train increasingly powerful models. Models that will be used to identify emotions, detect diseases, authenticate identities, predict behaviors. Models that will be sold to third parties, integrated into surveillance systems, used for purposes you never approved.
The terms of service you accepted by clicking "I have read and agree" — because no one really reads them — explicitly authorize this use. They mention improving services, developing new features, collaborating with partners. Formulations sufficiently vague to cover practically any use, sufficiently precise to deprive you of any legal recourse.
And even if you trust the current company, what guarantees its future behavior? Acquisitions, bankruptcies, strategic direction changes constantly transform promises into commercial opportunities. Your data collected today under a "respectful" policy may tomorrow be exploited under a totally different policy, and you will have no way to recover it.
The Third Cost: Your Structural Dependency
The third payment is the most pernicious, because it is made in the currency of autonomy. Every day spent using a cloud service reinforces your dependency. Your habits align with its features. Your vocabulary adapts to its strengths and weaknesses. Your entire workflow organizes itself around its permanent presence, its indispensable connection, its algorithmic benevolence.
This dependency extends beyond the tool itself. It embeds itself in your collaborations — your colleagues use the same service, creating interdependency. It anchors itself in your documents — your formats, your templates, your processes are optimized for this specific tool. It takes root in your skills — you learn to dictate in a way compatible with its language model, not in a natural way.
And then one day, the service changes. An essential feature disappears behind a paywall. An update breaks your workflow. A promising new competitor attracts investment, and your tool's development slows. You find yourself trapped in an ecosystem that no longer serves you, but from which you cannot extract yourself without a prohibitive migration cost.
This is the economic model of dependency. Companies do not sell tools. They sell habits. And habits, once established, are the most reliable recurring revenue there is.
The Alternative: Pay Once, Own Forever
Faced with this triple billing — subscription, data, autonomy — local AI proposes a radically different model. A model where you pay for your hardware once, where your data stays on your machine, where your autonomy is preserved.
Let us do the reverse calculation. An RTX 3060 graphics card, capable of comfortably running 7 to 12 billion parameter models, costs about three hundred dollars. Add a decent processor, RAM, an SSD — you have a complete machine for one thousand to fifteen hundred dollars. This machine does not serve only dictation. It runs all your software, your games, your office work, your creation. Local dictation is just one use among many.
Even counting only the GPU cost amortized over dictation, the calculation is eloquent. Three hundred dollars divided by one hundred and eighty dollars of annual subscription equals twenty months. Less than two years to amortize the hardware investment. And after those twenty months, you pay nothing more. Zero subscription. Zero hidden fees. Zero surprise price hikes.
More important still: your data never leaves. Your voice is processed in RAM, immediately destroyed after transcription. No storage. No analysis. No resale. Your biometric fingerprint stays where it should be — with you.
And your autonomy? Total. You can work without internet. You can customize your models. You can add your professional vocabulary, your shortcuts, your preferred transformation modes. You can modify the source code if you have the skills, or delegate this task to a developer. You are tied to no company roadmap, no board decision, no strategic pivot.
And there is a dimension that cloud services charge for separately — and expensively — that local AI makes native and free: text-to-speech. Engines like Kokoro or Nivoj allow your local agent to respond to you vocally, in real time, without any connection. This is not a gadget. It is what transforms a dictation tool into a true interactive dialogue: you speak, the agent processes, it responds aloud. A complete conversational loop, entirely offline, entirely under your control. Cloud services that offer this functionality generally reserve it for their premium subscriptions. Here, it is included by default.
Pareto's Law of Technical Independence
Skeptics invoke the superiority of cloud models. And they are right, in absolute terms. Hundred-billion-parameter giants surpass local models on the most complex tasks — complete software architecture, cutting-edge scientific research, advanced abstract reasoning.
But let us apply Pareto's law. Eighty percent of your daily use — writing emails, rephrasing texts, translation, boilerplate code generation, note-taking, shell commands — is perfectly mastered by a local 7 to 12 billion parameter model. Fast, precise, sufficient.
The remaining twenty percent — the truly complex challenges — can be delegated occasionally to cloud tools, used with discernment. The important thing is not to do everything locally. It is to no longer do everything in the cloud by default. It is to reserve external tools for cases where they bring real, irreducible value.
This hybrid approach — local by default, cloud by exception — is economically unbeatable. It preserves your privacy. It maintains your autonomy. And it costs you three to five times less over the long term.
The Real Luxury: Mastery of Your Tools
In a world obsessed with convenience, choosing mastery is an act of resistance. It is refusing the passive consumption model for the active competence model. It is preferring to understand how your tool works to simply using it. It is accepting a slight initial overhead — in learning time, in configuration — for lasting benefits.
PerkySue embodies this philosophy. A voice dictation tool that runs entirely on your machine. Whisper for recognition, llama.cpp for transformation, direct injection at your cursor — and Kokoro/Nivoj for text-to-speech, so your agent can respond aloud and engage in a true interactive dialogue. No account. No cloud. No subscription for the core. Just your voice, your machine, your productivity.
The cost? An install.bat file that detects your GPU, downloads dependencies, configures the environment. Fifteen minutes for a setup that will last years. The core is free — and that includes text-to-speech: the interactive dialogue capability with your local agent is part of the base pack, with no subscription. Advanced features — Pro modes, deep customization — are $9.90 per month, but the essentials, dictation + agent voice response, remain free and permanent.
Compare with the fifteen dollars monthly of a cloud service, multiplied by ten years, regularly increased, conditioned on the longevity of a company. The calculation is no longer economic. It becomes existential.
Conclusion: The Price of Freedom
The next time you evaluate a dictation tool, do not settle for the displayed price. Calculate the total cost of ownership over ten years. Question what happens to your data. Assess your ability to do without the service tomorrow if necessary.
The real luxury of the 21st century is not having access to all the world's services. It is being able to do without them. It is owning your tools, your data, your skills. It is paying once for something that stays, rather than renting perpetually something that can disappear.
Local AI is not a step backward. It is a choice for the future. A choice of mastery, sovereignty, intentionality. A choice that begins with a simple question: what if my voice stayed mine?

About the Author
Jérôme Corbiau is the creator of PerkySue, a local voice dictation tool with AI that works entirely offline, with no remote server or data transmitted. He is also co-founder and software architect of My App Zone SRL (Brussels), and creator of the Cloud Neareo platform — an award-winning CMS notably by Microsoft and the Public Service of Wallonia, deployed in museums and heritage sites. His work aims at a constant objective: putting technology at the service of the user, rather than the reverse.

Local AI: Your Last Line of Defense Against Digital Outages

Jérôme Corbiau — Tue, 02 Jun 2026 07:27:49 +0000

When the Cloud Goes Down, Your Productivity Dies
Imagine an ordinary morning. You open your computer, ready to dictate that urgent report. You press the shortcut for your favorite dictation tool. And then, nothing. An endless loading screen. A cryptic error message. Your subscription expired without warning. The server is under maintenance. Or worse: the company hosting your service has gone bankrupt, been acquired, or simply decided to quadruple its prices overnight.
This scene is not science fiction. It is the daily reality of millions of users who have entrusted their productivity to cloud services. In 2023, Google abruptly shut down its cloud gaming service Stadia, leaving thousands of customers with no recourse. In 2024, pricing policy changes at AI providers multiplied costs by ten for some developers. These companies hold absolute power: the ON/OFF switch for your work tools.
But beyond simple service interruption, there is a more insidious threat. What happens when a government decides to block access to an AI model? When a company censors certain types of content for political or commercial reasons? When your data is used to train models that will be used to influence you, manipulate you, sell to you?
Dependency as an Economic Model
Cloud computing has created an economy of dependency. The tech giants understood that the real value did not lie in selling software, but in the perpetual rental of a service. You no longer own your tools. You rent them. And like any good landlord, the renter can change the terms, raise the rent, or simply close up shop.
This dependency extends far beyond simple productivity tools. Our voice assistants, our navigation systems, our translation platforms, our creation tools — everything relies on infrastructures controlled by a handful of American and Chinese companies. We have traded our autonomy for convenience, without ever reading the fine print of the contract.
And the fine print is terrifying. It authorizes the use of your data to "improve services." It allows sharing with "partners." It retains your information "as long as necessary" — a deliberately vague formulation that means, in practice, forever. Your voice, your writings, your work habits, your very thoughts, become the raw material of an economic model that excludes you from the benefits while making you dependent on the finished products.
Local AI: An Act of Citizen Resilience
Faced with this centralization of technological power, an alternative is emerging. Local artificial intelligence — models that run entirely on your machine, with no internet connection, no account, no subscription. This is not simply a matter of privacy or cost. It is a matter of sovereignty.
Local AI is an act of resistance. It is the refusal to become a passive consumer, a free data producer for opaque economic models. It is the assertion that your productivity, your creativity, your thought, should not depend on the goodwill of a publicly traded company.
Take a concrete example with PerkySue, a voice dictation tool I developed that works entirely offline. Whisper for speech recognition, llama.cpp for AI transformation, all injected directly at your cursor. No remote server. No data transmitted. No account to create. If the internet cuts out, if the provider goes bankrupt, if prices explode — your tool keeps working. Because it lives on your machine, not in a data center on the other side of the world.
This resilience extends to all dimensions of use. You can work on a plane, in a rural area with no connection, in a country with strict internet censorship. You can process sensitive information — medical, legal, financial data — without ever exposing it to third parties. You can customize your tool, add your professional vocabulary, adapt the models to your specific needs, without asking anyone's permission.
Pareto's Law Applied to Autonomy
Skeptics will object that local models are less powerful than their cloud equivalents. This is true, in absolute terms. A 7 to 12 billion parameter model will not compete with 100+ billion giants for the most complex tasks — complete software architecture, cutting-edge scientific research, advanced multi-step reasoning.
But Pareto's law applies here with singular force. Eighty percent of our daily uses — writing emails, rephrasing texts, translation, boilerplate code generation, structured note-taking, shell commands — are perfectly mastered by these "modest" models. The remaining 20%, the ultra-complex tasks, can be delegated occasionally to cloud tools, used with discernment and intention.
The important thing is not to do everything locally. It is to no longer depend entirely on the cloud for tasks that we are perfectly capable of accomplishing ourselves. It is to reserve external tools for cases where they bring real, irreducible value, rather than using them by default because they are "simpler."
This hybrid approach — local by default, cloud by exception — transforms our relationship with technology. We become actors again, not mere consumers. We consciously choose when to outsource, rather than suffering systematic externalization. We retain control of our primary tools, while benefiting from external resources when they are truly necessary.
Toward Collective Technological Sovereignty
The issue goes beyond the individual. When entire communities adopt local and open-source tools, they build collective resilience. A city whose administrations use free software cannot be paralyzed by the bankruptcy of an American publisher. A country whose researchers master their own AI models is not vulnerable to technological sanctions or changes in trade policy.
Open source is the foundation of this resilience. Accessible, modifiable, auditable source code guarantees that the tool will survive its creator. That a community of developers will be able to maintain it, improve it, adapt it. That no company can unilaterally withdraw the service, change the terms, or lock features.
PerkySue, under Apache 2.0 license, embodies this philosophy. Every line of code is visible. Every user can verify what the tool does, how it processes data, where they go — or rather, that they go nowhere. A competent developer can modify the code, add features, fix bugs, create a fork adapted to their specific needs. The tool belongs to its community, not to a company.
The Choice of Intentionality
Adopting local AI is not a rejection of technology. It is a choice of intentionality. It is consciously deciding which tools deserve our trust, which data we are willing to share, which dependencies we accept to create.
It is also a question of real cost. A $15 monthly subscription seems modest. Multiplied by ten years, that is $1,800. Multiplied by all the cloud services we use — storage, office software, creation, communication — the total bill amounts to thousands of dollars per user, and $1,000+ billion globally for cloud companies. Petabytes of personal data exchanged for a convenience that could be provided locally.
Local AI offers us a third way. Neither technophobic rejection, nor passive submission to digital giants. A conscious, controlled, sovereign use of technology. Tools that serve us, without exploiting us. Tools that persist, without depending on a server. Tools that remain ours, even when the rest of the digital world collapses.
The next time you press a shortcut to dictate text, ask yourself: where does my voice go? Who controls it? What happens if the service disappears? If the answers worry you, perhaps it is time to take back control.

Jérôme Corbiau is the creator of PerkySue, a local voice dictation tool with AI that works entirely offline, with no remote server or data transmitted. He is also co-founder and software architect of My App Zone SRL (Brussels), and creator of the Cloud Neareo platform — an award-winning CMS notably by Microsoft and the Public Service of Wallonia, deployed in museums and heritage sites. His work aims at a constant objective: putting technology at the service of the user, rather than the reverse.