Start like everyone else does: work out your business plan. Understand exactly what it is your business will be doing, what it will involve, who else (personnel and other companies) it will involve. What resources do you need to begin, and what will you need to actually implement your service/product.
What is your revenue stream and what are your outlaying costs? Will you need to borrow any capital to start with? What will you charge for your service/product, and how long will it take you to earn back and pay off your borrowing?
You will also need to consider regulation and legislation: how much of it is there, what applies to you, will you need to register with a professional or national body, do you need certification, do you need permits or licences?
What stage is your offering at: is this just an outline of an idea, of do you have a proof of concept? How much will it cost to prototype or test this concept?
How and to whom will you market this offering? What are it's USP's? Can you back-up your claims for it with data and evidence? Have you done the research to show there is a space for your offering?
Yishan Wong, the former CEO of Reddit, already has, so you might ask him - but I suggest you read all his blog posts on the subject first so that you don't ask anything that he's already covered.
Maybe chat with the folks from Marine Cloud Brightening Project and ask which obstacles they faced?
Maybe consider starting something less ambitious first, but where you have a good handle on what you can achieve, a unique idea and an advantage over those who would want to compete against you and over political interests who would want to destroy you?
For the love of the spark, fucking don't. At least separate yourself from the social ladder of EA and learn the real version of rationality first.
Or: ignore that advice, but at least don't do the actual MCB implementation worldwide that costs a billion a year, talk with the scientists who worked on it and figure out the way that MCB could be done most efficiently. And then get things to the point of having a written plan, like, "hey government, here's exactly how you can do MCB if you want, now you can execute this plan as written if/when you choose". Do a test run over a small area, iterate and improve on the technology. B/c governments or big NGOs are more likely to do it if it's fleshed out, e.g. lower risk from their POV.