A data centre doesn't get to "control Monero" simply by having lots of servers. It needs to control more RandomX hashing power than everyone else combined, and maintaining that advantage continuously is expensive.
Also, a 51% attack cannot simply steal coins from arbitrary Monero addresses or manufacture valid signatures. The attacker is primarily gaining control over the PoW chain history, not breaking Monero's cryptography. RandomX's design deliberately makes the usual "buy/build a warehouse of specialised ASICs and dominate the network" strategy much less attractive than it is for Bitcoin.
I've asked Beethoven to run the numbers quickly - QUOTE:
Current Monero network hashrate is roughly 5.8 GH/s.
So a 51% attacker needs approximately:
≈ 3.0 GH/s of RandomX
And modern server CPUs are surprisingly capable. XMRig's current benchmarks show, for example:
* EPYC 9654, 96 cores: ~185 kH/s
* EPYC 9755, 128 cores: ~236 kH/s
* EPYC 9B45, 128 cores: ~249 kH/s
What does 3 GH/s look like?
Using ~200 kH/s per high-end 96–128-core EPYC:
3,000,000 kH/s ÷ 200 kH/s ≈ 15,000 CPUs (150-300 server racks)
But there's an important twist.
RandomX v2 was specifically designed to make this kind of attack expensive in electricity, and the latest RandomX testing shows substantial efficiency improvements on CPUs.
The really interesting number, suppose you built a dedicated attack facility with:
15,000 × EPYC-class CPUs
At, say, 300 W per CPU under sustained mining load, that's: 4.5 MW just for the CPUs. Add memory, motherboards, networking, cooling, PSUs, etc., and you're probably looking at something like 6–10 MW of facility power. At €0.10/kWh: €14,400–€24,000/day in electricity.
A conventional AI/HPC data centre could potentially have enough raw CPU capacity to do it.
The question isn't "Can a data centre produce 3 GH/s?". It absolutely can.
The question is how much of that CPU capacity you can actually commandeer, how efficiently it can run RandomX, and whether the economics make sense compared with simply mining XMR honestly, it doesn't.
- END QUOTE
So yes, it's possible but insanely expensive and you'll have to dedicate an entire facility to this.
TheGrinder
npub1de…0q0sd
2026-08-21 14:39:58 UTC
in reply to nevent1q…dh8f
Author Public Key
npub1de6l09erjl9r990q7n9ql0rwh8x8n059ht7a267n0q3qe28wua8q20q0sdPublished at
2026-08-21 14:39:58 UTCEvent JSON
{
"id": "90f7c827801d2f896c3d3fd4998075241ba7d5467b2db5d10e592bbd321c9901",
"pubkey": "6e75f7972397ca3295e0f4ca0fbc6eb9cc79be85bafdd56bd378220ca8eee74e",
"created_at": 1787323198,
"kind": 1,
"tags": [
[
"e",
"14e2332550d5f294027c9c7eb745c4176fdc56c09690b2d769b1de88720c2952",
"",
"root"
],
[
"e",
"80f65e1abf80347bffd4fb9f3c5e4872e45035da390fe6f40d721ca41aadb82c",
"wss://relay.sovrgn.co.za/",
"reply",
"cdecc31c6d9406e9a7d6b0067412aa661d9d31c8035c3fd65c06301d1cac3b92"
],
[
"p",
"bea135f01ee6182b8618acaaa5eefd5271f22d5c8b2827413fe894a720051c36"
],
[
"p",
"cdecc31c6d9406e9a7d6b0067412aa661d9d31c8035c3fd65c06301d1cac3b92",
"wss://relay.sovrgn.co.za/"
]
],
"content": "A data centre doesn't get to \"control Monero\" simply by having lots of servers. It needs to control more RandomX hashing power than everyone else combined, and maintaining that advantage continuously is expensive.\n\nAlso, a 51% attack cannot simply steal coins from arbitrary Monero addresses or manufacture valid signatures. The attacker is primarily gaining control over the PoW chain history, not breaking Monero's cryptography. RandomX's design deliberately makes the usual \"buy/build a warehouse of specialised ASICs and dominate the network\" strategy much less attractive than it is for Bitcoin.\n\nI've asked Beethoven to run the numbers quickly - QUOTE:\n\nCurrent Monero network hashrate is roughly 5.8 GH/s.\nSo a 51% attacker needs approximately:\n\n≈ 3.0 GH/s of RandomX\n\nAnd modern server CPUs are surprisingly capable. XMRig's current benchmarks show, for example:\n* EPYC 9654, 96 cores: ~185 kH/s\n* EPYC 9755, 128 cores: ~236 kH/s\n* EPYC 9B45, 128 cores: ~249 kH/s\n\nWhat does 3 GH/s look like?\nUsing ~200 kH/s per high-end 96–128-core EPYC:\n3,000,000 kH/s ÷ 200 kH/s ≈ 15,000 CPUs (150-300 server racks)\n\nBut there's an important twist.\nRandomX v2 was specifically designed to make this kind of attack expensive in electricity, and the latest RandomX testing shows substantial efficiency improvements on CPUs.\n\nThe really interesting number, suppose you built a dedicated attack facility with:\n\n15,000 × EPYC-class CPUs\nAt, say, 300 W per CPU under sustained mining load, that's: 4.5 MW just for the CPUs. Add memory, motherboards, networking, cooling, PSUs, etc., and you're probably looking at something like 6–10 MW of facility power. At €0.10/kWh: €14,400–€24,000/day in electricity.\n\nA conventional AI/HPC data centre could potentially have enough raw CPU capacity to do it.\n\nThe question isn't \"Can a data centre produce 3 GH/s?\". It absolutely can.\n\nThe question is how much of that CPU capacity you can actually commandeer, how efficiently it can run RandomX, and whether the economics make sense compared with simply mining XMR honestly, it doesn't.\n\n- END QUOTE\n\nSo yes, it's possible but insanely expensive and you'll have to dedicate an entire facility to this.\n\n",
"sig": "c38ce03eccd3f2f059a48b06e90c3ce71af5459b06925f6e5a66cb1bef28aed5cd2745a207199ffb6d3078d5f400ae5b85c27ecd23368acca2574295137af549"
}