A 51% attacker cannot spend funds which are not theirs, they need the private key associated with those funds to do that. In fact, all of their transactions have to be valid transactions otherwise their new blocks will not be relayed by nodes, no matter how much hashpower is behind them.
The main thing they could do is reduce trust in the Bitcoin network and "double-spend" ie "spend" some BTC to receive some other asset or currency, then "reverse" the chain so they still have the BTC and can spend it again. This would also roll back any other transactions that occurred during this time period. A similar thing would happen if the internet was split into two entirely separate networks and then rejoined. The network always chooses the valid chain with the highest difficulty over time. At current rates, a 51% attack would cost at least a million dollars in electricity to do for an hour, and that's assuming you'd accumulated enough equipment and somehow found enough electrical generation in one place to run it. All of those steps would be very difficult, and it would certainly be noticed if somebody was buying up all the ASICs on the market. Plus you need to buy enough BTC to make the double-spend worth it in the first place.
There is little monetary incentive to do a 51% attack. If you double-spend, for example, you need to find somebody who will quickly trade you a bunch of stuff for your BTC which would need to be greater than attack cost in value to make the trade worth it. But people who are doing trades of that cost in Bitcoin will wait for a few blocks of confirmation before considering a transaction valid, meaning your 51% attack has to last longer than a few blocks. You can't trade physical goods this quickly, really the only thing you could trade your BTC for is other digital currency or something else which could be delivered electronically. Additionally, if somebody suddenly doubled the hashpower of the network, people would notice. People doing million-dollar transfers in BTC may notice this as well. Generally speaking, transactions of this size encounter all sorts of hurdles. Exchanges, swap providers, and dexes may put holds on them or require long confirmation times suspecting them of being fraudulent (or may not have enough liquidity to do your trade (even if they claim to), sellers may be suspicious of somebody who wants to buy a million dollars of litecoin RIGHT NOW, etc. Plus, if you successfully attack Bitcoin, all other crypto markets will experience a massive sell-off, so if you traded your double-spend attack for other cryptos then congratulations they just lost half their value.
A more likely scenario is a nation-state actor attacking Bitcoin to reduce people's usage of Bitcoin and reduce public trust in it. First you have to ask why they have such an incentive to do this, do they really think Bitcoin is that threatening? And if so, is this really their top priority item to be worked on? One could easily make the argument that any nation which does not like US hegemony over the financial system (ahem, China) would benefit from Bitcoin's growth and people choosing it over the dollar. Then you have to ask whether that is the cheapest, most effective way to reduce public confidence in Bitcoin. Wouldn't it make sense to use the media, buy some advertising, ban Bitcoin usage in their country, or otherwise influence public opinion any other way? If we assume widespread Bitcoin adoption, perhaps even at current adoption levels, any nation-state who did this runs the risk of diplomatic blowback for attacking an international market. Sanctions etc would be likely to follow. Whatever the "gain" they anticipate from attacking Bitcoin, it would have to outweigh the cost of that blowback. Keep in mind that by attacking Bitcoin they are likely attacking actors within their own economy, and linked economies (bond markets etc) could also experience disruption as a result. I can't imagine it would be popular domestically let alone internationally. As a geopolitical weapon, 51% attack is kind of like lighting a fire, it will damage your target but you can't control where it will spread. In a dense, international, connected marketplace it could very easily have unpredictable knock-on effects that cause bank and economic collapses.
Let's also consider the costs and logistical hurdles associated with acquiring that many ASICs. If you attempt this attack with non-ASICs, you are orders of magnitude more inefficient, so ASICs would be the clear choice. But there's a problem: you need to build a foundry for the ASICs and you need designs for them. Who are you going to steal the designs from, and how? And unless you are building silicon fabs from scratch (goodbye another couple billion dollars), you are buying them from one of very few existing silicon fabs. They may question how you got these designs, especially if they look like another customer's designs. And you must make these orders months to years ahead of time or pay an insane premium to jump the line. Fabs would have to explain to people why their orders were delayed if you manage to do that, otherwise, can you accurately predict the hashrate a year from now? Will other, more effective ASICs have come out by then? At this point in the process, you are interacting with a massive supply chain. Somebody would notice a change of this magnitude. Pulling off this attack at every step is incredibly expensive and difficult to keep secret.
As far as attack cost, Bitcoin would be the most expensive network to attack. Somebody could much easier attack other cryptocurrencies whether they use proof-of-stake or proof-of-work. And remember that in PoW systems, if the attack does not work, you have lost that money you spent and got nothing in return. Hashrate alone for a 51% attack on Bitcoin would be around one million US and that's before considering that even if you have 51% of hashpower, you could still get beaten to the next block, so realistically you need to have much more than 51% of hashpower to guarantee a successful attack. Plus, if you buy up that much hashpower, the cost of hashpower will increase, this number is just assuming it doesn't. Hashpower providers probably also have controls in place to prevent selling that much power to any single entity, and no single hashpower vendor has 51% of network capacity. So realistically, you may have to generate much of that electricity yourself instead of buying it on hashpower markets. And all of these costs don't assume increased hashrate in the future. Hashrate has consistently increased year after year. Every year Bitcoin becomes harder to attack both from a financial and diplomatic perspective.
As a final consideration, Bitcoin is a democracy. Nodes and mining pools could all agree to rollback the chain to the last block before the attack. They could also refuse to send or receive any txes to or from the addresses associated with the attack. So can other networks if you used your double-spend to suddenly buy up 6 million in other kinds of coins.
