The main assumption here is that the network would actually be used. If it isn't used, then the tokens would have no value. We all knew that from the start which is why everyone is pining from mainnet. If it's all just speculation then it's just your typical hypebeast muh partnership shitcoin. While it can still potentially make LINK moon, it's harder to quantify because we're still waiting for those partnerships to be confirmed.
First of all, as stated by Sergey several times already, LINK will be used to pay node operators. I don't want to bother arguing "Why not just ETH/BTC/fiat?" because it doesn't matter at all. Besides the technicals of the token, the mere fact that Sergey said so is enough. While there may be solutions around this, i.e. the one suggested by the artificiallawyer article, in the end it will still be using LINK, that is, even if they'll be paying in fiat, LINK will still be consumed. It's just a possible workaround to make things convenient.
With that in mind, it means that as long as the network is being used, LINK will be bought and used by the users of the network. What we should look out for is the volume of the network when mainnet is finally launched. Obviously, it won't be immediate, since people would have to have their nodes up first, but that's besides the point. The higher the volume of requests, the higher the demand for LINK. That's one way for LINK to increase in value, not just "because it needs this high amount as collateral, which means LINK needs to go to this price".

Besides demand from companies, I would also like to include that with the high volume of requests comes the fees that node operators would profit from. What does that imply? Again, this whole explanation comes from the assumption that the network is actually being used. It means that node operators are profiting from their nodes, and judging from this part in the whitepaper talking about reputation metrics, specifically "Amount of penalty payments" (page 18 under 5.2 Reputation System), it means that the amount of LINK you hold will affect your reputation. If being a node operator turns out to be massively profitable, it makes sense to assume that people would be scrambling to become a node operator, and in the process buying LINK to increase their reputation. I think this is the part where all the talk about collateral comes in, where only whales with massive stacks of LINK would be the only ones taking in high value contracts (high value as in the amount of fiat being handled in the contract and high value as in the amount of fees it will pay).
I would like to note that there are 350M tokens in reserve for node operators so the second point will not be as relevant as the first but at least it's safe to assume (definitely could be wrong) that the tokens won't simply be dumped because they need to be staked. Maybe there would be a smart contract that would require the LINK to be staked at least a year before it can be sold or some shit. I don't know.
Also, there is the other 300M in the hands of Sergey. If we assume that this is where the high-value clients (SWIFT, Salesforce, Accord) will get their LINK, they'd have reason to pump the price. While this sounds deluded, imagining them going on Binance to pump fucking LINK for some reason, you need to keep in mind that while LINK is used as payment when using the network, the fee per request is still measured in fiat. If they are sold 100M LINK each valued at $1 per token, it means they'd have $100M LINK worth to be used as payments for the network. If per request, it costs $1 each, they can use the network 100M times before they need to buy LINK again. I think their accounting departments can easily work out that using $1M to pump 10x an illiquid market (at the beginning) to increase the value of their LINK stack by $900M would be profitable. Not to mention hoarding as much LINK as they can before other companies do the same thing I mentioned. This seems extremely deluded but if you put yourself in the shoes of SWIFT and the like, it just seems simple game theory at that point.

Again, all of this assumes that the network is actually being used, that companies would actually benefit from utilising smart contracts. I avoided as much as possible including numbers because the point of this thread is to only talk about how LINK would increase in value, not how much. To summarize, 1. Speculation/hype from partnerships, 2. Demand from the users of the network, 3. Demand from would-be node operators. If you feel restless waiting for it to moon I think the most productive thing you can do >>FOR LINK<< (because there are definitely other shit you can do for yourself) besides making LINK content is studying how to run a node. Check out how to use AWS, or building your own VPS to run a node (Thomas says in Telegram that it's easily achievable).