Join the decentralized compute network serving AI and gaming workloads globally. Backed by @avax, @borderless_cap, @cv_labs & @oviohq
- Tweets 7,148
- Following 796
- Followers 204,154
- Likes 9,996
Pinned Tweet
$YOM is live across major exchanges.
Pick your platform
Trading links in the replies 🫡
1,776
3,223
1,319
258,799
From this month, Xbox Cloud Gaming is no longer unlimited. Game Pass Ultimate gets 15 hours a month. Premium gets 10. Essential gets 5. After that you buy more hours.
The reason matters more than the news. Reporting suggests Microsoft starts losing money once a user goes past roughly 15 hours.
Think about who that is. The most resourced company in gaming, with its own global data centre footprint, and the economics still break at 15 hours a month per person.
That's not a Microsoft failure. It's what happens when every frame has to travel from a distant campus to a player and back. The GPU was never the expensive part. The distance was.
We've run 60,000 sessions on GPUs sitting near the person playing. Different cost base, because the frames don't have far to go.
1
2
8
975
Tomorrow, people in 10 US states vote on 41 different measures to limit data centre development. Ohio alone has 21 of them. Gallup says 70% of Americans don't want one built near where they live, and that holds across both parties.
Here's what nobody seems to say out loud: the demand for compute isn't the problem. Nobody is protesting against AI existing. They're protesting against a 50-acre building with its own power station turning up next to their house.
8
6
21
2,134
So the question isn't how to make people accept data centres. It's whether you need one.
Within a few miles of every voter casting one of those ballots, there is already hardware capable of doing this work. In homes, in offices, and in server rooms that are half empty. Already built, already connected, already paid for. Most of it switched off right now.
That's the network we've spent four years putting together.
2
251
The @BBC asked whether the @Apple and @Meta open-weight model ends the data centre era.
The numbers in that discussion answer it. 816 billion parameters. A terabyte of memory to store the weights. Your phone has 16GB. A good laptop has 64. Running it at home means roughly £40,000 of hardware.
On-device AI is real, but not at that scale, and not for consumers.
The debate keeps offering two options: a hyperscale campus or the device in your hand. Nobody names the third. The GPU doesn't need to be in your house. It needs to be near you.
Close enough that latency stops mattering. Not so far that you're back in a data centre on another continent.
That hardware already exists. It's in homes, offices and telco sites, and most of it is switched off.
14
14
29
1,962
Here's the part people miss about distributed compute.
Every node someone plugs in makes the network better for somebody who lives nearby. No new data centre. No grid application. No five-year planning fight.
Just hardware on sites that already exist, run by people who can earn rewards for keeping it online. Thousands of small decisions instead of one giant one.
That's how you cover a map.
11
11
33
2,461
YOM retweeted
Watched @BBC 's AI Decoded on whether Apple and Meta's open-weight move ends the data centre era.
The most useful bit is ten minutes in, and it's a number.
1
1
4
331
60,000 sessions on YOM.
Every one of them started on a GPU somewhere near the person playing. Not a data centre on the other side of the world. A machine down the road.
That's the whole idea, and now it's got a number.
8
10
37
3,243
YOM retweeted
ALPHA LAUNCH DAY IS HERE!
Walter.gg Alpha is officially LIVE to the public.
Stream GTA V and jump straight into our first FREE GAME FRIDAY title on any device for $0.99 for your first 3 months!
1
5
10
458
Back from San Francisco with the Gaming Founders Circle, where YOM is one of five companies in the inaugural cohort run by @discord, @Techleapnl and the Dutch Games Association. @DGA_online
A session with Discord's leadership. An investor's dinner hosted by Techleap and Prince Constantijn van Oranje-Nassau. A fireside at Databricks HQ on how startups get large technology companies to bet on them.
The answer from that room: you don't start with the partnership conversation. You start by having solved something they can't solve easily themselves, with numbers to show it.
Dutch Games Awards next month.
12
11
38
2,024
We didn't build this for voice AI. We built it for games.
43 nodes, 16 countries, users in 113. 30 ms median latency worldwide. 14.5 ms if you're near a node.
Game streaming punishes lag harder than almost anything. Clear that bar, and you've cleared it for real-time AI too. Same rails, different workload.
13
11
39
2,382
Here's what nobody building voice AI wants to hear: your model being fast doesn't make it feel fast.
Light takes time. Networks take time. If the GPU is three countries away, you've already lost the conversation before the model does anything.
We put GPUs on towers, in buildings, and at the edge. Close enough that the round trip disappears.
15
13
37
2,691
Block's new voice AI, Dessa, learned entirely from speech. No text. They say it runs faster than real time.
So the model problem is basically solved. What's left is physics. A reply that's instant in San Francisco is laggy in Lagos, and no amount of model speed fixes that.
The compute has to move closer. That's the whole thesis.
14
13
40
2,841
YOM retweeted
Block's Dessa learned from speech alone, no text, and they report inference running faster than real time.
Here's what that unlocks and what it doesn't. A model that fast makes duplex voice possible, where both people can talk at once.
2
1
6
249
Weekly roundup. Four things worth your attention.
Inference cost decides whether an AI business survives. Every conversation burns GPU time. As usage scales, that compute bill becomes the business model, not a line item on it.
Game streaming demand is climbing while AI demand pushes GPU supply tighter. A growing market running on hardware that keeps getting more expensive. The answer isn't more warehouses. It's putting GPUs on sites that already exist, close to players.
Europe is hitting a hard limit. Data centres now take 23% of Ireland's electricity. Spain is drafting caps on water and power for large AI sites. The compute is needed. Concentrating all of it in single regions is what's breaking.
Where the network stands: 43 active nodes across 16 countries, users served in 113 countries, 30ms median latency globally and 14.5ms within 50km of a node.
A new licensing structure is planned for Q1 2027. Listing licenses double, and every YOM NANO ships with two.
Gaming was the proving ground. Real-time AI runs on the same rails.
Gaming first. AI next.
20
13
46
3,408
This is why we built YOM Nano and our distributed edge network. Every node deployed to YOM adds compute capacity where the infrastructure already exists without requiring new data centers, new grid connections, or regulatory fights.
• 43 active nodes | 16 node countries | 113 user countries served
• 30 ms global median latency (14.5 ms within 50 km of a node)
• Operator economics: Node rewards scale dynamically based on specs, uptime, and live session demand.
Gaming was our proving ground for the hardest real-time latency challenge in compute. Real-time AI and voice inference are scaling on the exact same rails.
18
13
38
2,983
YOM retweeted
Alpha Launch is just one week away (Oct 30)!
What game are you testing first on the cloud?
Drop your title below and join Walter now!
2
3
6
304
23% of Ireland’s total electricity now goes to data centres, and Spain is already drafting strict caps on water and power for large AI sites.
Mainstream markets desperately need compute, but stacking massive centralised facilities in single regions is becoming environmentally and economically unsustainable.
17
12
34
2,783
Bloomberg Intelligence forecasts game streaming will hit $38B by 2035 as traditional consoles fade.
The catch?
Surging AI demand is pushing GPU prices higher at the exact same time.
You have a booming consumer market running on increasingly expensive hardware.
That’s why we don’t build centralised warehouses; we deploy GPUs on existing sites close to players.
19
18
39
2,932
The world's largest AI companion app just filed for Chapter 11. The stated cause?
Inference cost per user. Every single conversation burns GPU time. At scale, that compute bill dictates whether an AI business survives or dies.
Compute cost isn't an operational detail; it's the entire business model.
24
20
42
2,844