Credits expire: how founders waste a $100,000 grant
Twelve-month windows, no extensions, clocks that start at acceptance rather than first use. The expiry rules matter more than the amounts, and nobody reads them.
The most common way to lose a large credit is not rejection. It is winning it at the wrong time.
Founders treat an approved credit as money in the bank. It is closer to a voucher with a printed date on it, and the dates are shorter and stricter than the marketing pages suggest.
The windows, as published
| Program | Window | The clause that costs people |
|---|---|---|
| Cloudflare | 12 months | Expires at 12 months or when consumed, and extensions are not available |
| DigitalOcean | 12 months credits, 15 months support | Core credits exclude GPU Droplets and dedicated inference |
| Neon | 12 months | Counts from acceptance, not from first use |
| MongoDB | 12 + 12 months | 12 months to activate the code, then 12 to spend; extensions case by case |
| ElevenLabs | 12 months | Converts to the free plan at 12 months or when 10,000 credits remain |
| Notion | 1, 3 or 6 months | Length is decided at review, not by you |
| Google Cloud (Scale) | 2 years | Year two covers 20% of usage, not a second lump sum |
Two patterns are worth pulling out of that table.
Pattern one: the clock starts before you are ready
Neon's twelve months run from acceptance. If you apply in January because a checklist told you to, and your product does not have real database load until August, you have spent seven of your twelve months on an empty database.
Cloudflare is the harsher version of the same thing, because there is no extension mechanism at all. The credit is gone at month twelve whether you used it or not.
MongoDB is the design to wish for: twelve months to activate, then twelve months to spend, with extensions available case by case if you ask within three months of expiry. That structure lets you win the program early and start the real clock when you are ready.
The lesson is not "apply late". It is that for consumption-based credits, the application date should be chosen against your own roadmap, not against a listicle. Ask one question before every application: can I plausibly consume this within twelve months of today? If the answer is no, and the program is not tiered by funding, waiting is free.
Pattern two: coverage is not a lump sum
Google Cloud's Scale tier covers 100% of your Google Cloud and Firebase usage up to $100,000 in year one, then 20% of usage up to another $100,000 in year two.
Read the second half carefully. To consume the whole second-year allowance you would need to spend $500,000 in that year, because you are only being covered for a fifth of it. The published "up to $200,000 over two years" is accurate and also unreachable for almost every company that receives it.
This is not a criticism of the program — the first-year coverage is genuinely excellent and among the best offers in the market. It is a warning against putting the headline number in a financial model.
Pattern three: the carve-outs
DigitalOcean's core credits explicitly exclude GPU Droplets, dedicated inference and third-party AI model hosting; those need a separate GPU credit application. Anthropic's credits work only on the first-party Claude API through Claude Console, not on AWS Bedrock or Google Cloud Vertex AI.
For an AI-native team these carve-outs can hollow out the value entirely. A general cloud credit is not GPU budget, and a model credit tied to a first-party API cannot be routed through your existing cloud commitment. Check the exclusion list before you architect a workload around a credit you have not yet read the terms of.
What to do on the day a credit is approved
Five minutes of work that regularly saves five figures:
- Write the expiry date in a shared calendar, not a note. Include the activation date if the clock starts separately.
- Set a reminder at 60% of the window. For a twelve-month credit that is month seven — early enough to change how you use it, late enough to know your real consumption rate.
- Record the carve-outs next to the amount. "$100,000, excludes GPU Droplets" is the useful form. "$100,000" is the form that misleads you six months later.
- Check whether extensions exist. MongoDB: ask within three months of expiry. Cloudflare: they do not exist, so do not plan on one.
- Note what happens at the end. ElevenLabs converts to the free plan. Cloud programs revert to standard billing, which means the month after expiry is the month your infrastructure bill appears for the first time — at full price, on production traffic.
That last point deserves its own line, because it is where the real damage happens. A credit window ending is not a neutral event. It is the moment your unpriced infrastructure becomes a real cost, usually at the scale you grew to because it was free.
The uncomfortable conclusion
Credits are best understood as a subsidy on a specific twelve-month period of your company's life, not as an asset. Their value is set by three things in this order: how much you can consume inside the window, what the carve-outs exclude, and what the bill looks like on the first day after.
The amount comes fourth. Which is why the database lists the window and the exclusions next to every number, and why the sequencing playbook treats "when can I actually spend this" as a first-class question rather than an afterthought.