Learn / Infrastructure
Cheap cold email infrastructure, and the sizing math
The setup that used to cost $1,000 to $2,000 a month now runs about $200 to $500. Cold email tools got so crowded that price became the only thing left to compete on, and the floor fell out. Cheap cold email infrastructure is now available to anyone with a credit card, and most operators are still pricing their sending off numbers from three years ago.
Cheap only works if you size it right, though. Get the ratio of domains to inboxes to sends wrong and you either burn the domains or pay for capacity you never use. This is the written version of the training below: the exact math behind a 10,000 email per day build, lead lists at $2.40 per 1,000 instead of thousands of dollars, and personalization at $130 per 100,000 prospects on your own API key.
The sequencer first, and it is not your CRM
Before any math, you need a tool built for sending: Smartlead, Instantly, or EmailBison, which is what we run. EmailBison is a private sequencer, so our deliverability stays isolated rather than pooled with everyone else on a shared platform, which matters at the client count we send for. The reason this cannot be HubSpot or Salesforce is capacity. You are connecting hundreds, sometimes thousands, of inboxes across hundreds of domains, and you want every reply landing in one unified inbox so a setter can actually work them. Your primary domain NEVER touches a cold send.
Work backwards from the daily number
Pick the send volume first and everything after it is division. For this build the target is 10,000 emails a day, roughly 200,000 a month, split about 60% Microsoft and 40% Google. That split is a cost decision, not a deliverability one. Microsoft inboxes send fewer emails each, but they need dramatically fewer domains, and domains are the line item that compounds across a build this size.
The math for 10,000 emails a day
Two providers, two sets of limits, one target. Read the table down each column and the order count falls out of it:
| The math | Microsoft | |
|---|---|---|
| Share of the 10,000 | 40% | 60% |
| Sends per inbox per day | 15 | 5 |
| Inboxes per order | 100 | 99 |
| Sends per order per day | 1,500 | 495 |
| Orders needed | 3 | 13 |
| Domain ratio | 1 domain per 3 inboxes | 1 domain per 99 inboxes |
| Domains needed | 100 | 13 |
| Daily capacity built | 4,500 | 6,435 |
Add the two sides and you get 113 domains carrying 10,935 sends a day. That number probably looks absurd next to the one mailbox your business runs on, and it is the whole point: capacity comes from spreading a small per inbox limit across a lot of inboxes, never from pushing any single inbox harder. The per inbox and per domain ceilings behind those 15 and 5 figures are in the daily sending limits breakdown.
113 domains, bought in one sitting
Naming 113 domains by hand is where most people quit. Ask ChatGPT for 113 lookalike domains built off your real one, short prefixes or suffixes, three to five letters, shortest prioritized, things like "ll" or "inco" hung on the front or back. Paste the whole list into Porkbun's bulk search, select every domain that comes back available, and check out in one go.
Then the extension question. For deliverability, .com is first, .co is second, .info is third. We buy .info anyway: about $2.57 a year against $10 to $12 for a .com and $8 to $10 for a .co. At one domain that trade looks silly. At 113 domains, .info is a third to a quarter of the bill, and the extra sending capacity that money buys is worth more than the deliverability edge you handed back. The records that actually decide inbox placement (SPF, DKIM, DMARC, and how you ramp) are covered in the infrastructure guide, and Google publishes its own bulk sender requirements if you want the rules from the source.
Mailboxes went from $9 a seat to pennies
This used to be the line that killed the whole idea. You would open Google Workspace or Outlook and buy accounts at $9 to $20 per user, or $8 to $10 on the Microsoft side, and do that a thousand times. Resellers now start around $2.99 a month for 100 Google mailboxes, and the same for 100 Microsoft. We use Zapmail for this build. Mailer and Scaled Mail do the same job, and there are a dozen more competing on exactly this price.
The order is a form. You tell them how many accounts, on which domains, and where to send them, and the mailboxes appear in your account. SPF, DKIM and DMARC are set for you, which used to be a full afternoon of DNS records per domain. Warmup starts on its own and runs 2 weeks, with the accounts sending back and forth to each other so they read as legitimate inboxes by the time real sending begins. That 2 week clock is the one part of this you cannot buy your way past, so order the mailboxes before you finish writing the copy.
The stack got cheap. The math is what still stops people.
Lists at $2.40 per thousand
Most people assume a list of this size means ZoomInfo and a five figure contract, so they never look for the cheap path. Build the search in Apollo with your real filters instead. For this one I built AI companies, decision maker titles (CEO, owner, founder, plus marketing and sales leaders), US and Canada, which returned 144,000 people. Then copy the search URL out of the address bar.
Drop that URL into a scraper like AmpLeads, set the maximum number of leads, check you have credits, name the file, and start it. It runs as little as $2.40 per 1,000 leads, which puts a list big enough to feed this build somewhere between $50 and $100. Export as CSV and you are done. Which filters are worth setting, and which ones quietly wreck a list, is in the Apollo walkthrough. Cheap does not mean good, and a list built around a category instead of a person still will not convert. A category is a bucket, and you cannot write to a bucket.
Personalization on your own API key
Clay is where the sticker shock happens. The $350 a month plan comes with 10,000 credits, which against a list like this is nothing, and 100,000 credits runs $1,500 a month. People look at that, decide personalization is for companies with a budget, and send plain.
You do not have to. Upload the CSV to Clay, then go into OpenAI, set up your own API key with pay as you go credits, and connect it inside Clay. Now the AI columns bill to your key at OpenAI's own token pricing instead of eating Clay credits. Pulling and normalizing a prospect's target industry costs about $0.0013 per row. Run that across 100,000 prospects and the bill is $130.
One setting decides whether that holds: the model. Default to 4o mini or 4.1 nano, test both, and stay off the top tier models for fields like this. You are normalizing an industry string, not writing an essay, and the expensive models cost multiples for work the cheap ones do fine.
Three columns do the work: pull the prospect's target industry, normalize it, then pull the exact service they sell. Here is what that looks like once it hits the copy:
Hey {{first_name}}, noticed you primarily service clients in the {{industry}}. If we could help connect you with these types of individuals looking for {{service}}, would that be worth a chat?
That reads as "clients in the pharmaceutical industry" looking for "fractional CFO services" by the time it sends. Not clever, just specific, which is all personalization has to be. The full column stack, including verification and where enrichment stops being worth it, is in the Clay workflow breakdown.
Load it and launch
Export the CSV out of Clay, upload it to the sequencer, paste in the copy, map the personalization fields to the columns you just built, attach every mailbox you bought from the reseller, and launch. That is the whole stack, and every line item in it costs a fraction of what the same build cost two years ago.
One caveat before you go spend the $200. Cheap infrastructure is not a strategy. Volume only earns its keep as speed of learning on an offer that already works, so if the offer is wrong, all you have built is a much faster way to find that out. That is worth something. Just know that is what you bought.
tl;dr: pick the daily send number, divide it by what each inbox can send, buy the domains and mailboxes that math gives you, scrape the list for $2.40 per thousand, and run personalization on your own API key. Roughly $200 to $500 a month, all in.
FAQ
How much does cheap cold email infrastructure cost now?
A setup that ran $1,000 to $2,000 a month a few years ago runs about $200 to $500 today. The three line items are domains (roughly $2.57 a year each on a .info), mailboxes from resellers starting around $2.99 a month for 100, and the sequencer. Tool competition drove the price down; almost nobody has repriced their sending against it.
How many domains do I need to send 10,000 cold emails a day?
About 113 for a 60% Microsoft, 40% Google build. Google needs one domain per 3 inboxes, so 300 inboxes take 100 domains. Microsoft runs one domain per 99 inboxes, so 13 orders take 13 domains. The domain count, not the mailbox count, is what makes one provider mix cheaper than another.
Are .info domains bad for cold email deliverability?
They are third. A .com is best for deliverability, .co is second, .info is third. But a .info costs about $2.57 a year against $10 to $12 for a .com and $8 to $10 for a .co. At 113 domains that price gap buys you more sending capacity than the deliverability edge you gave up, so at scale we buy .info.
How do you build a cold email list for under $100?
Build the search in Apollo with your real filters, copy the search URL, and paste it into a scraper like AmpLeads instead of exporting natively. Scraping runs as little as $2.40 per 1,000 leads, so a list big enough to feed a 10,000 a day campaign lands between $50 and $100. The same list through an enterprise data tool costs thousands.
Can you use your own OpenAI API key inside Clay?
Yes, and it is the single biggest cost cut in the stack. Clay sells 10,000 credits at $350 a month and 100,000 credits at $1,500. Connect your own OpenAI key instead and a normalized industry field costs about $0.0013 per row, so personalizing 100,000 prospects comes to $130. Stay on the cheap models for those fields.
How long before new cold email inboxes can send?
Two weeks. Resellers start the warmup automatically the moment the mailboxes land in your account, and the accounts spend that window emailing each other so they look like real inboxes instead of accounts opened to spam people. Sending before the warmup finishes is how people burn 113 fresh domains in a week.
If you would rather not assemble 113 domains and 1,500 mailboxes yourself, that is what we do: 3,000+ sales calls booked for clients in the past year, on infrastructure the client owns at the end. Or keep learning free, every training grouped by topic sits on the trainings hub, and the full video for this one is on YouTube.
PS - the cheapest part of this stack is the part everyone skips. Naming and buying 113 domains takes one prompt and one checkout, and it is the only reason the other numbers work.