The last Voidcom update shipped in early July. Then nothing for eight weeks: no new builds, no posts, no visible activity. If you’re in the beta and wondered whether the project had died, that’s a reasonable thing to wonder, and you deserve an explanation rather than a changelog that pretends the gap didn’t happen.
Why it went quiet
Two reasons, and neither is dramatic.
The first is money. Voidcom doesn’t earn anything yet. The servers and the build machines cost money every month and nothing comes in, so the project can’t buy its way out of the second reason.
The second is time. I have a job that pays the bills, and this summer it needed the hours Voidcom usually gets. Evenings and weekends were the first thing to go.
Both of these will happen again. Most of the development work is mine at this point, and when I’m out, there isn’t a second engineer who picks up where I left off. I’d rather say that plainly now than leave you to guess at the silence. The project isn’t abandoned. It moves in bursts, and the length of the pauses between bursts depends on things outside it.
What happened this week
The first thing back was housekeeping rather than features, on purpose. After two months away I didn’t trust my memory of the codebase, so I went through all of it systematically: the server, the desktop and mobile apps, the native audio, video and crypto layer, the bot API and the launcher. That sweep turned up 27 serious issues and 157 smaller ones. All of them are fixed.
Every third-party library we build on got updated too, one at a time, so anything that broke could be traced to a single change.
And the build system moved. In April we put our code and builds on Codeberg, the Germany-based nonprofit Git host. The idea was right; in practice it was too unreliable, and a solo developer with a few hours in the evening can’t spend them fighting the infrastructure that’s supposed to carry the work. The pipeline that turns our source code into the installers you download now runs on GitHub Actions, still on build machines we own and operate ourselves. That move matters for the rest of this post.
The hardware market changed under us
In April I wrote about moving our infrastructure to Europe. Cloudflare out; Hetzner, Bunny and Scaleway in; more cost and more work accepted as the price of matching what we say to where the data lives. I meant it, and I’d make the same choice again in a market that looked like April’s.
The market doesn’t look like April’s.
Since autumn 2025 the three companies that make most of the world’s DRAM have shifted production toward high-bandwidth memory for AI accelerators. Ordinary server RAM got scarce, and scarce got expensive. In November, Counterpoint Research forecast that server DDR5 prices could double by the end of 2026. That turned out to be optimistic. OVHcloud now says the price it pays for RAM was six times higher in June 2026 than a year earlier, expects nine times by September and twelve times early next year. SSDs are at 3.2 times. Its founder calls the situation exceptional and expects it to last until 2028.
Hosting prices followed. Hetzner raised cloud prices by 30 to 37 percent on April 1, storage by 30 percent, and its 128 GB RAM add-on went from about 46 euro to 264. It raised prices twice more before June and then split its dedicated range into a standard tier and a cheaper “Limited” tier built from lower-cost parts. OVHcloud’s new 2026 servers go up by as much as 87 percent from September. Scaleway and IONOS made comparable moves.
The US hyperscalers, as far as anyone can tell, mostly haven’t. Amazon raised prices on one GPU product line twice this year and left the rest of the catalogue alone. OVHcloud’s own framing is that it used to be three times cheaper than the big US clouds and will end up about twice as cheap. My read is that the big clouds can absorb the memory cost inside AI revenue, while a mid-size European host buys the same chips at the same inflated price and has nowhere to put it but the invoice.
For a project with no revenue that is the whole story. When we priced out what we need for the next stretch, the EU option costs us in some cases about three times what the non-EU equivalent does. Not 20 percent more. Three times.
And “EU-only” was already leakier than it looked
There’s a second thing. When we picked Bunny as our CDN in April, the argument was that a Slovenian company with EU edge nodes keeps our visitors’ data out of US hands. Bunny’s privacy policy now states that customer usage metrics, account identifiers, billing and transaction information, subscription details and historical spend go into Google BigQuery for revenue reporting and analytics. The data residency is EU, but BigQuery is Google, and the whole argument of our April post was that a US provider can be compelled to hand over data no matter where it physically sits. Their sub-processor list tells the same story in smaller print: Google Workspace for email, plus Slack, MailChannels and OpenAI, all US companies, for support and tooling.
None of this is a scandal; it’s how nearly every company operates. But it means the thing we were trying to buy, a European vendor whose own vendors are all European, isn’t on the shelf. We looked, and we found no way to use only EU services that themselves use only EU services.
What we’re doing about it
So here is the change, stated as plainly as the April post stated the opposite.
From now on Voidcom will also use non-EU services where the European option costs multiples or doesn’t cleanly exist. Cloudflare is the obvious first example. The build system’s move from Codeberg to GitHub Actions was the same kind of call, made for reliability rather than price. Where an EU option is close in price and does the job, it stays, and we’ll keep trying to make that the case wherever we can. Where it isn’t, we take the cheaper, better-supported thing and say so.
This holds until the project pays for itself and the financial picture changes. At that point we can afford to make decisions on principle again, and I’d like to. Right now the choice is between a pure stack and a project that keeps moving, and I’m choosing the one that keeps moving.
Two things don’t change. DMs and voice stay end-to-end encrypted the way the encryption posts describe, and none of that depends on who runs the CDN or the build system. And we told you in March that if anything changes we’d say so publicly before it happens. This post is that. The privacy policy gets updated the day any of these services actually switches over, not after.
What comes next
The roadmap hasn’t changed; the pace has. iOS and Android through the store channels, the second server region, voice-fingerprint verification, the bot SDK. All still on the list in roughly that order, and none of them with a date attached. I’ve stopped giving dates because I’ve missed too many, and a solo developer with a day job shouldn’t be promising quarters.
If you’re in the beta, keep sending the crash reports. They’re still what drives the fix list.