It's doesn't replace a worker is supplants having an experienced welder with having literally anyone willing to work with it while they hone their own skills.
A lot of the people arguing against onshoring of manufacturing and its viability have exactly one mental model of how factories work, which is 50 year old men in dying industries using archaic manual processes learned through blood, sweat and tears. Any solution that doesn't immediately restore exactly that is simply infeasible to them. When in reality, its like starting up any other business or industry, you work with what you got and what tools are available. If all they have is zoomers and six figure CnC hardware, they'll figure out how to make it work. Besides, a $100k machine in a machine shop is a bargain, not a horrific burden. For context, fitting out a new restaurant can run you anywhere from 500k to a million, easy. People don't really have a good context of the scale of the money that goes into any slightly serious business venture.
American manufacturing doesn't need to be instantly perfect in its planning day one, doesn't instantly have to have a matured labor and skills market day one. It doesn't need to immediately replace everyone, everywhere all at once when it comes to the supply of goods. All it needs to do is provide a tariff-enforced cheaper option that can be scaled up, and the markets can work with that. If they can pivot to just 20% of the supply chain being primarily domestic, then they've got a 20% of their production that's now predictable and immune to trade war and geopolitical bullshit, which they can scale up. Importing the other 80% to meet demand becomes an unfortunate cost to be reduced over time.
The companies that get ahead of this now and become the first movers become the prime candidates for investment, and we'll rapidly see this turn into an AI situation - Investors will see X company making Y% returns beyond market norms that they attribute to onshoring, and will start expecting every business they're investing in to show them their onshoring adoption plans. The ones with good plans and good progress (Or good bullshitting skills, but nothing is perfect) on them will continue to receive investor capital, which can offset much of the initial loss and pain period - AI's burned hundreds of billions of dollars to create a chatbot that still can't handle working a drive-thru, its not hard to imagine investors being much more open wallet around industries with literal centuries of financial history and profits behind them.