The Hardware and Software Ownership Renaissance
Tech workers are quietly rebuilding their stacks around things they own instead of things they rent. The receipts back them up.
Written by Dana WhitfieldEdited by Luis Chavez-Mattos·Aug 29, 2026·5 min read
The short answer
De-Googling and rejecting rental software means moving from Gmail and sealed devices toward things you actually own: independent email providers like Proton, repairable hardware like Fairphone, and self-hosted open models like Qwen running on your own machine. The pattern shows up in the numbers: Proton has crossed 100 million accounts,1 Fairphone shipments jumped 42% in 2025 even as the smartphone market stalled,2 and developers running local LLMs are cutting AI bills by more than 70%.3
This is not nostalgia for dial-up. It is a cost and control argument, and it is being made with spreadsheets, not slogans.
Email: the first thing people own again
Gmail still runs the table. It holds roughly 39% of global email market share and more than 1.8 billion users.1 But the migration away from it is not hypothetical anymore. Proton, the Swiss encrypted email provider, passed 100 million accounts and has stayed independent, ad-free, and free of venture capital the entire time.1 Signups to Proton VPN jumped 9,000% above average in the days after the invasion of Ukraine began, and protests in Iran drove a 5,000% spike.1 Those numbers are not privacy tourists. They are people who suddenly needed infrastructure they could trust because the stakes got real.
The pitch from switchers is not abstract civil liberties talk. It is plainer than that: you do not want your inbox, your calendar, and your identity provider to be the same company that also builds your ad-targeting profile. None of this requires abandoning convenience for austerity. It requires treating email like an asset with a bill you understand, not a free service with a bill you cannot see.
Hardware: the phone you can actually open
Fairphone is the clearest data point in the entire ownership story. While the global smartphone market struggled to find its footing, Fairphone's shipments surged 42% in 2025, delivering more than 145,000 units and pushing revenue to €73.3 million, up 35% year over year.2 Seventy-eight percent of Fairphone 6 buyers were first-time customers, meaning this is not a shrinking niche recycling itself. It is new buyers actively choosing repairability.2 The company also cut its net losses in half and turned daily operations cash-positive, which shows the model is not a charity case propped up by goodwill.2
The phone's pitch is mechanical, not ideological: swap the battery yourself, replace the camera module, and keep the phone running with years of promised software support instead of two before the OS updates stop.2 For the Linux crowd, Fairphone's officially unlockable bootloader means installing an alternative OS is a documented, reversible process rather than a warranty-voiding hack.
That is the real shift. Repairability used to mean voiding your warranty and hoping. Now it is a supported feature, and buyers are rewarding it with wallet share.
Compute: running your own model instead of renting one
The most concrete numbers in this whole trend come from AI infrastructure. A solo developer named Sam Hartley documented spending $187.42 in a single month on OpenAI, Anthropic, and Google API calls for a normal workload of 300 to 500 queries a day.3 He bought a used RTX 3060 GPU for $150, installed Ollama, and pulled Qwen 2.5 7B. Within a year his hybrid setup, routing roughly 80% of queries to the local model and reserving cloud APIs for complex reasoning, dropped his monthly AI bill from $187 to about $25.3 The GPU paid for itself in under four months, and his projected first-year total cost, including hardware and electricity, was $594 against a hypothetical $2,244 cloud-only bill.3
None of this replaces frontier models for genuinely hard reasoning. Hartley's own local Qwen setup missed a subtle security flaw in a database schema that GPT-4o caught immediately.3 The honest lesson from serious local-LLM writeups is the same: local models handle most daily work for free, and the harder fraction still needs a rented frontier model. Ownership does not mean total independence. It means not paying rent on the parts you do not need to.
What the ownership math actually looks like
| Category | Rented default | Owned alternative | Reported saving or growth |
|---|---|---|---|
| Gmail, ad-supported, roughly 39% market share | Proton, independent and ad-free | Proton crossed 100 million accounts1 | |
| Hardware | Sealed, glued smartphones | Fairphone, modular and repairable | Shipments up 42% in 20252 |
| AI compute | Pay-per-token cloud APIs | Local Qwen on an owned GPU | Monthly bill cut from $187 to $253 |
The pattern across all three: the rented option is more convenient on day one and more expensive by month six. The owned option asks for setup time or upfront hardware cost and pays it back within a single billing cycle in the AI case, or within a single purchase decision in the phone case.
Why this is happening now, not five years ago
Two things converged. Cloud AI pricing made the per-token math suddenly visible to individual developers in a way SaaS subscriptions never were, because you can literally watch a token counter tick against your credit card. And geopolitical anxiety about surveillance and platform dependence pushed both individuals and whole countries toward alternatives they could verify and control, which is exactly what drove Proton's growth spikes during real-world crises.1
Neither of these forces required an ideological conversion. They required someone doing the arithmetic once and writing it down.
This same instinct, that the software running your business should be an asset on your balance sheet rather than a recurring subtraction from it, is the whole argument behind treating software as something you own, not rent. Whether it is a personal inbox or a company's internal tools, the calculation is the same: what does this actually cost over three years, and who controls it when the vendor changes the terms.
FAQ
What does de-Googling actually mean in practice? It means replacing Google-run defaults, primarily Gmail and Google Drive as shipped, with independently run or self-hosted alternatives such as Proton, in order to reduce dependence on one company's data practices and pricing decisions.
Is switching email providers actually worth the hassle? For most people the switching cost is a few hours of redirecting logins. Proton's growth to 100 million accounts suggests enough people found it worthwhile.1
Do repairable phones like Fairphone cost more than a flagship? Fairphone's pricing sits closer to mid-range than flagship, and 78% of its 2025 buyers were first-time customers rather than repeat enthusiasts paying a premium out of loyalty, which suggests the price is competitive on its own terms.2
Can a local AI model actually replace ChatGPT for daily work? For routine tasks like summarizing, refactoring, and simple code review, yes, at close to zero marginal cost. For complex reasoning or security-sensitive design work, cloud frontier models still catch errors local models miss, so most practitioners run a hybrid setup.3
How fast does local AI hardware pay for itself? In one documented case, a $150 used GPU running Qwen locally paid for itself in under four months against a $187 monthly cloud API bill.3
It means replacing Google-run defaults, primarily Gmail and Google Drive as shipped, with independently run or self-hosted alternatives such as Proton, in order to reduce dependence on one company's data practices and pricing decisions.
For most people the switching cost is a few hours of redirecting logins. Proton's growth to 100 million accounts suggests enough people found it worthwhile.
Fairphone's pricing sits closer to mid-range than flagship, and 78% of its 2025 buyers were first-time customers rather than repeat enthusiasts paying a premium out of loyalty, which suggests the price is competitive on its own terms.
For routine tasks like summarizing, refactoring, and simple code review, yes, at close to zero marginal cost. For complex reasoning or security-sensitive design work, cloud frontier models still catch errors local models miss, so most practitioners run a hybrid setup.
In one documented case, a $150 used GPU running Qwen locally paid for itself in under four months against a $187 monthly cloud API bill.
- 1.There are now over 100 million Proton Accounts — Proton
- 2.The repairable smartphone revolution is finally picking up speed — Android Authority
- 3.I Ditched ChatGPT for Local LLMs and Saved $2,000 in a Year — The Real Numbers — DEV Community
