It's low volumes at first as no large customer orders lots without doing a pilot first to figure out how it best fits their needs.
During their Q1 call, Tesla said they have 2000 orders. So it's safe to assume that they only need a few thousand units per quarter at first (best case scenario) and ramp from there over the years.
They likely reuse a lot of existing parts and given the low volumes, they likely start with few dedicated machines and more human workers - this is really important , start making it without spending much capital on tools as they just can't afford to ramp M3, prepare for MY and new locations in China and EU plus repay some debt all next year. They almost certainly need to take some more debt next year as it is, no matter how well M3 production is going.
A new US location hasn't been mentioned at all and no point starting it in the EU or China so it will likely be split between Fremont and Giga1, just like M3. Doesn't make sense for general assembly at Giga1 unless they plan to do Model Y there too and do almost everything there but that might imply a delay to 2020 for the TST.
A similar thinking can be applied to the Roadster, ofc with the difference that the TST will ramp to higher volumes in time.
Would assume longer term they shift part of M3 prod to non US ;locations and they likely need a new US GF by 2021 as the pickup arrives.