Airtable is where most of the businesses I work with started, and for good reason. I have built on it for years, alongside Make and portal builders like Noloco and Softr, and I still recommend it to new operations teams. You get a relational database with a spreadsheet face, automations, and a front end a non-technical team can use on day one.
The problem is that Airtable does not tell you when you have outgrown it. The symptoms arrive one at a time and each looks like a small annoyance. Below are the seven signs I check for, with the numbers behind each one, why it matters in business terms, and something cheap to try first. Having moved a large Airtable, Make and Noloco system to Supabase, Vercel and Next.js, I have opinions about when that move is worth it and when it is not.
1. You are closing in on the record cap
Airtable caps records per base at 1,000 on Free, 50,000 on Team and 125,000 on Business, according to Airtable's own plans page. Enterprise Scale goes higher, but the number is not published there and third-party sources disagree on whether it is 250,000 or 500,000.
The business problem is the shape of the next step. On Team at $20 per seat (annual billing) you cross 50,000 records and the only path is Business at $45 per seat. Jet Admin describes this as a 125% per-seat increase for 2.5 times the records. A twelve-seat team goes from $240 to $540 a month in one step, and 125,000 is not far away when every job, shipment or invoice line is a record.
Try first: archive. Move closed records older than a year into a separate base, delete log tables nobody reads, and check whether a junction table is doubling your count. If you are growing 20% or more a year, which techconcepts lists as a migration trigger alongside 50k records, archiving buys you a year rather than a solution.
2. Linked-record lookups take seconds, not milliseconds
Performance degrades well before the hard cap. An Airtable consultant wrote in the community that Airtable starts falling apart with large record sets, with Enterprise clients at around 50,000 records per table already seeing severe slowdowns. In a case study published by Kreante, a logistics business with roughly 80,000 records across six linked tables was waiting 4 to 6 seconds for linked-table lookups every time an automation fired.
Every one of those seconds is paid for by a dispatcher waiting on a grid view, a client in a portal, or Google. One user in the Softr community wrote that loading times were "super bad" and Google was not ranking some pages because of it. Kreante's rule of thumb: grid views over 3 seconds, automations delayed 5 minutes or more, or API 429 errors.
Try first: lookups, rollups and formulas that chain across tables are the usual culprit. Replace a chained rollup with a scheduled automation that writes a plain number into a field. Hide heavy fields from your busiest views. If tables are independent by region or year, split the base.
3. Make scenarios fail because the API throttles you
Airtable's API allows 5 requests per second per base. Exceed that and you receive a 429 status code and must wait 30 seconds before requests succeed again. On Team there is also a cap of 100,000 API calls per workspace per month; after that you are throttled to 2 requests per second until the month resets. Metadata calls count toward the cap.
For an operation that runs on Make or Zapier, this is where things break. A scenario iterating over a few hundred records at the busiest hour hits 5 per second easily, gets a 429, sits for 30 seconds, times out and leaves half the batch unprocessed. That is what happened in the Kreante case above: Make automations timed out because the Airtable API throttled under peak load. Make adds its own ceiling, a 40-minute maximum per scenario run on paid plans according to Make community staff, so long backfills need hand-built queues.
Try first: batch your writes, since the API accepts up to 10 records per request. Replace polling triggers with webhooks, add a short sleep inside iterators, and stagger scenario schedules so they do not all fire on the hour. If you are on Team, Business removes the monthly cap, but the 5 per second per base rule still applies on every plan.
4. Automations pause at the monthly run cap
Automation runs are capped at 25,000 per month on Team and 100,000 on Business, according to consistent third-party summaries from Noloco and eesel.ai; Airtable's plans page separately caps unique automations per base at 100 on Team and 150 on Business. When you hit the run cap, automations pause until the month resets.
The damage is silent. Nobody gets a page when the invoice reminder stops going out on the 24th. Noloco lists automations quietly failing at run caps as one of its own signs that a firm has outgrown Airtable.
Try first: audit which automations fire per record change versus on a schedule. A "when record updated" trigger that touches 500 records a day burns 500 runs; one scheduled automation that finds and processes those 500 records is a single run. Consolidate, and check your run count weekly.
5. The seat bill is growing faster than headcount
Airtable charges per editor: $20 per seat on Team and $45 on Business with annual billing, or $24 and $54 month to month, per the plans page. Noloco points out that contractors, finance reviewers and freelancers each cost a full seat, and portal builders charge again per user on top. Kreante puts the cost inflection point at roughly 5 to 8 seats.
The bill scales with how many people touch the data, not how much value the system creates. A client portal is the worst case: every customer who needs to see a status is a seat, a portal-builder user, or a phone call.
Try first: audit seats quarterly and ask who actually edits. Give occasional contributors a form or a read-only interface instead of a seat. Move to annual billing.
6. Compliance or data residency has become a hard gate
Airtable supports HIPAA only on the Enterprise Scale plan with a signed BAA, per Airtable's own support page and HIPAA Journal. Free, Team and Business cannot hold PHI at all. Even on Enterprise the restrictions are heavy: PHI only in records, no PHI in automated emails, no HIPAA coverage for Airtable AI or marketplace extensions, and no use as a patient portal. On the EU side, a Softr community member described Airtable's EU data locality as coming at a "steeeeep price for startups", while Supabase offers region choice for free.
Unlike the other signs, this is a gate rather than a gradient. One healthcare customer or one EU procurement questionnaire can block a deal regardless of how many records you have. Airtable does hold SOC 2 Type 2 and ISO 27001, which is often enough for a security review, but a BAA or a residency requirement is a different question.
Try first: keep regulated data out of Airtable entirely and reference it by ID from a system that is compliant. Get an Enterprise Scale quote so you know the real number, and ask the customer whether SOC 2 would satisfy them before assuming they need a BAA.
7. You have a renewal date and no plan B
Bending Spoons agreed to buy Airtable in August 2026, closed the deal in September, and laid off a large part of the roughly 800-person staff in mid-October, as reported by the SF Chronicle. Airtable says customers' plans, prices and service offerings will not change. The community is sceptical because of Airtable's own 2023 repricing, when one user wrote that they had "3 weeks to move to another platform".
The sign is not the acquisition itself. It is that if your renewal doubled tomorrow, you would have no answer other than paying it.
Try first: do not panic-migrate. Inventory your critical bases and automations, keep regular exports, note your renewal date, and price one alternative with a single representative base so the number is real when you need it.
Airtable plan limits at a glance
All figures from Airtable's plans page and API limits page as of October 2026, except automation runs, which are third-party figures. Prices are per collaborator per month on annual billing.
| Limit | Free | Team | Business |
|---|---|---|---|
| Records per base | 1,000 | 50,000 | 125,000 |
| Attachment storage per base | 1 GB | 20 GB | 100 GB |
| API calls per workspace per month | 1,000 | 100,000 | Unlimited |
| API rate limit per base | 5 req/s | 5 req/s | 5 req/s |
| Automation runs per month | 100 | 25,000 | 100,000 |
| Price per seat (annual) | $0 | $20 | $45 |
Stay or move?
Stay on Airtable when:
- You are pre product-market fit and the system is still changing weekly.
- You have fewer than about 20,000 records and low growth.
- Non-technical staff need to change fields and schema themselves, without a developer.
- Nobody on the team wants to own a codebase.
- Only one or two signs apply and the cheap fixes have not been tried.
Scope a migration when:
- Three or more of the signs apply at the same time.
- A compliance or residency requirement is blocking revenue.
- Customers or search engines are paying for your load times.
- Your annual Airtable, Make and portal spend is already comparable to a one-off rebuild.
Those stay criteria are close to the ones Kreante publishes. Plenty of businesses never outgrow Airtable, and a migration that delays validating a product is a bad trade.
The middle step: move the data, keep the interface
There is a hybrid option I often recommend as a first step. Move the data to Postgres on Supabase, and keep Airtable or Softr as the editing surface through a two-way sync tool such as Whalesync. Your team keeps the grid they know, your automations and portal read from a database that does not throttle at 5 requests per second, and your exit ramp is half built. There is one well-documented limitation: formula, rollup and lookup fields sync one way only, so logic that lives in those fields stays in Airtable until you rebuild it. Treat the hybrid as a bridge, not a destination.
What a real migration involves
If three or more signs apply, scope it properly rather than exporting CSVs on a Friday. The hard parts are the same in every migration I have read about or done: linked records become foreign keys, attachment URLs expire and must move to real storage, and Airtable automations and formulas do not carry over and must be rebuilt. Kreante's case ran as an eight-week phased project: audit, build the Postgres schema, run both systems in parallel on live records, then cut over the front end. Their advice is never to cut over without a parallel run of at least one full business cycle, and I would not do it any other way. The outcome there was automations dropping from 4.2 seconds to 0.6 seconds on average and the database bill from about $540 to $25 a month. Not every project looks like that, but it is a fair picture of what the move is for.