For example, this TI PHY chip DP83630 has two MAC addresses that can be used as destination and source MAC addresses for PHY control frames. (page 72 and 94)
Does this means all PHY chips will have the same MAC or am I missing something?
(promoted from comment to answer)
These addresses are used for PHY CONTROL frames only, not for normal data frames. As far as I understand such frames they are not destined for a particular node. I suspect that the actual address is unimportant, and the choice of two addresses is only for debugging and testing (so you can see from which of two sides a frame is sent).
Many PHYs do not have MAC addresses, in particular the cheapest ones as you need to pay for a block of MAC addresses and they leave that up to consumer.
Example: the popular (at least among hobbyists) ENC28J60 does not have a pre-programmed MAC.
MAC addresses have nothing to do with the PHY layer. These are relevant in the MAC layer, which is of course why they are called "MAC" addresses in the first place.
Every ethernet MAC is supposed to have a globally unique 48 bit address. I suppose it would be possible for a manufacturer to create MACs with built-in addresses and then serialize them in production. However, I am not aware of that being done for bare MACs or MAC/PHYs. Some module do this since they have their own processor and EEPROM.
Ususally the MAC has registers that are programmed with the MAC address by the controller during system initialization. That controller typically reads its assigned MAC address from some sort of non-volatile memory that is customized during production.