People always forget those numbers are worst case scenarios. I mean, you can get luck on the very first guess too.
BTC private key space is 256 bit. Let's say a billion wallets, that's 30 bits, so you need to check 226 bits to hit one wallet.
A H100 does about 1000 TFLOPS at the very most, that's 10^15 or 50 bits per second (generously assuming we can check on key per FLOP).
6B days of that will give you an additional 50 bits (6 = 8 = 3 bits, B = 1000^3 = 30 bits, day = 10^5 seconds = 17 bits).
Now we're talking 100 bits. But as discussed above, you need to check 220 bits to hit a key. There's still quite a gap.
For comparison, the entire Bitcoin network (using 1% of world electricity) does about 1000 EH/s at the moment, that's 10^21 or 70 bits per second (so, roughly equivalent to a million of H100, using the rough overestimating sketch above).
Per year, that's 70+25 = 95 bits. Still far.
So if you had 10,000 H100s running, it'd only take ~1500 years.
You'd have a high probability to find key in under ~1000 years, though.
Even if I'm off by 3 orders of magnitude, it would take a decade and cost billions, and not make financial sense.