Startup credits are free (or deferred) money from cloud and model providers that you spend on their bills while you find users. As of October 2026, the self-serve tiers you can usually get without a VC referral are modest (think AWS Activate Founders starting at $1,000, Google for Startups Cloud Program Start tier up to $2,000, Microsoft for Startups starting at $200 as you verify and use Azure). Bigger piles still mostly need an accelerator or partner Org ID. Apply first for the stack you will actually run on, set billing alerts before the money runs out, and track every program on a one-page sheet so the credit cliff does not surprise you.
What are AI and cloud credits actually good for?
Credits buy runway on infra and model bills. They are not proof that you have a business, and they are a weak reason to pick an architecture. Use them to ship, load-test, and serve early users without the bill wiping your runway. Do not lock yourself into a provider only because it was free this quarter.
They are good for:
Compute and storage you already planned to buy (API calls, databases, object storage, serverless).
Early AI inference on the provider's own models, when the program's fine print says those models are covered.
Buying time while you talk to customers, not while you chase every free tier on the internet.
They are not good for:
Proving demand. A credit approval email is not product-market fit.
Picking your stack. If you would not pay for the service at full price in six months, do not build on it just because credits exist.
Covering everything. Marketplace models, some GPUs, support plans, and registrar fees often sit outside the credit balance. Read the exclusions before you celebrate.
One line on pricing: do not set your own product prices based on a bill you are not paying yet. When you need the cost-per-user math, use a separate guide such as EarlyHunt's piece on how to price an AI product when costs vary per user.
The running example (made-up)
Pipetide is an illustrative, made-up product, and every number about it below is made up too.
The setup: a solo founder builds an AI tool that turns messy CSV exports from boutique agencies into a weekly client report PDF. The stack is mostly Google Cloud plus Gemini for summarization, with a Workers edge on Cloudflare for uploads, and Claude as a backup model via the first-party API. The founder has a custom-domain email, a live landing page, and no VC. They need runway on cloud and model bills for the next few months without spreading applications across five clouds.
Which startup credit programs matter in October 2026?
Amounts and rules change. Everything below was checked on the official pages on 2026-10-14. If a page disagrees with this post later, trust the page.
Program | Who it is for | What you get (as of Oct 2026) | How to apply | Verified |
|---|---|---|---|---|
Pre-Series B startups; Founders for self-funded, Portfolio with an Activate Provider Org ID | Founders: start with $1,000, select startups up to $5,000. Portfolio: up to $200,000 with a Provider Org ID. Invite-only AI tier above Portfolio exists. | Create a Builder ID, choose tier, link a paid AWS account, submit. Hear back in about 5-10 business days. | 2026-10-14 | |
Start tier: early digital-native startups planning to seek venture funding. Scale: VC-funded Pre-Seed/Seed (or Series A within 12 months). | Start: up to $2,000 for one year. Scale: up to $200,000 over two years (Year 1 up to $100,000 at 100%; Year 2 up to $100,000 at 20%). AI-first Scale: up to $350,000 (Year 1 up to $250,000 at 100%). Gemini/Gemma covered; third-party Marketplace models are not. | Apply with a matching business email, public website, and Google Cloud billing account ID. | 2026-10-14 | |
Privately held software startups, not Series C+, under $350,000 lifetime free Azure credits | Up to $200 to start, then milestones up to $150,000 based on verification and sustained Azure workload usage (e.g. $5,000 after business verification; higher tiers need more workloads over ~60 days). | Apply at startups.microsoft.com. Investor Network referral codes open higher benefits. Review often within ~3 business days. | 2026-10-14 | |
Early-stage founders; founded in last 5 years or funded in last 2 (when applications are open) | When open: free year of Claude Team (up to 5 Premium seats) plus a one-time $1,000 first-party API credit; partner Stack offers up to $45,000 in partner value. VC partners may grant up to $100K API credits. Credits do not work on AWS/GCP/Azure. As of 13 Oct 2026 the current cohort is full and new applications are closed. | Subscribe for the next cohort on the program page; apply when applications reopen. | 2026-10-14 | |
Credits for portfolio companies of partner VCs; learning resources open to all | API credits via a unique VC referral code. OpenAI does not publish a public dollar amount on the startups page. Build Hours, repos, and newsletters are open without a partner. | Ask your VC for a referral code; submit with org ID, product description, business email, funding basics. | 2026-10-14 | |
For-profit tech startups under 10 years, live website, domain email, first-time applicants; funded up to Series B | $10k (bootstrapped / under $1M), $100k or $350k with affiliated partner funding. Credits last one year or until used. Caps include R2 up to $10k and Workers AI by tier; AI Gateway and registrar fees are excluded. | Apply on the Cloudflare Startups page. Review often within ~48 hours. | 2026-10-14 | |
AI-native startups, $10M raised or less, company website + matching email; not agencies | 12 months of credits (amount set via partners; generally a $10,000 monthly limit). Core credits exclude GPU Droplets, many GPU products, and third-party frontier models hosted outside DigitalOcean. | Create a DO team with business email and card, then apply (partner path or "Other"). | 2026-10-14 | |
Incorporated startups under 10 years with at least one developer and a working website | Free membership (no equity). Preferred pricing on select NVIDIA hardware/software, training, partner offers, and the ability to request free cloud credits from NVIDIA and partners in the portal. No fixed public credit amount on the page. | Apply anytime; no cohorts or fees. Keep the portal profile updated. | 2026-10-14 |
For Pipetide (made-up), the founder starts with Google Start tier and Cloudflare's $10k bootstrapped tier because that matches the live stack, then watches Anthropic's newsletter for the next Claude Startups cohort instead of inventing a workaround while applications are closed.
What is the honest eligibility reality?
Self-serve tiers exist, but they are smaller. The headline numbers people tweet ($100k, $200k, $350k) usually need an accelerator, VC, or named partner Org ID / referral code. Angel, friends-and-family, crowdfunding, and many grants do not qualify you for Google's Scale tier. AWS Portfolio needs an Activate Provider Org ID. OpenAI API credits for startups come through partner VCs, not a public self-serve form. Cloudflare's $100k and $350k tiers expect funding from an affiliated partner.
Where legitimate codes come from:
Accelerators and incubators that are listed as AWS Activate Providers, Cloudflare affiliated partners, or DigitalOcean partners.
VCs in a provider's network (OpenAI partner network, Microsoft for Startups Investor Network, Anthropic partner VCs).
Bank and formation perks pages that publish partner Org IDs when they have a real relationship (for example some Stripe Atlas or Brex perk directories historically pointed founders at Activate Providers). Only use a code your org actually belongs to. Never buy Org IDs, never invent an affiliation, never open duplicate accounts to re-apply.
If you are bootstrapped, apply for the Founders / Start / $10k-style tiers that say self-funded is allowed. Do not wait six months hoping a fake Portfolio application somehow works.
What should you prep before you apply?
Most rejections are boring: personal Gmail, no live site, billing account on the wrong email, or a product description that sounds like a consultancy. Spend one evening on this checklist.
Custom-domain email that matches the public website (founder@yourdomain.com).
Live website that describes the product in plain language. A parked page with "coming soon" and no product story often fails review.
Company LinkedIn or X / GitHub presence where programs ask for it (Cloudflare lists this explicitly).
Incorporation status you can state honestly. NVIDIA Inception requires incorporation. Microsoft's $5,000 milestone needs business verification as a registered legal entity. If you are not incorporated yet, apply only to programs that allow that, or incorporate first.
One-paragraph product description: who it is for, what job it does, whether AI is core or a bolt-on (Google's AI-first path cares about this).
Honest stage and funding answers. Do not claim institutional VC if you only have friends-and-family cash.
Billing account in good standing with a payment method on file, on the same domain email. Credits that expire unused still need a card for overage.
Org / Builder / billing IDs ready (AWS account, Google Cloud billing ID, OpenAI org ID, Cloudflare account ID).
In what order should you apply?
Start with the provider you will actually run on this quarter. Do not open five clouds "just in case." Every account is another bill to watch and another place to accidentally leave a GPU on.
Primary cloud for the product you are shipping (for Pipetide: Google Start tier).
Edge / CDN / workers if that is already in the architecture (Cloudflare $10k).
Model provider credits once you know your main model path (first-party Anthropic or OpenAI when a real path exists; do not assume Bedrock or Vertex Marketplace models are covered by cloud startup credits).
Secondary cloud only if you have a concrete migration or multi-region plan, not a spreadsheet fantasy.
Membership programs without a fixed credit number (NVIDIA Inception) when GPUs or partner offers matter later.
Skip programs whose exclusions kill your plan. DigitalOcean core credits do not cover GPU Droplets or third-party frontier models hosted outside DigitalOcean. Google program credits do not cover third-party Marketplace models. Claude Startups credits (when available) apply only through Claude Console, not Bedrock, Vertex, or Azure.
How do you track credits so you do not waste them?
Keep one sheet. Columns that matter: program, amount approved, start date, expiry (or "until used"), what it covers, what is excluded, monthly burn, cliff date (30 days before expiry), and a note on alerts set (yes/no).
Here is a made-up Pipetide tracker. Treat the dollars as fiction.
Program | Amount (made-up) | Start | Expiry | Covers | Excluded | Monthly burn (made-up) | Cliff date | Alerts |
|---|---|---|---|---|---|---|---|---|
Google Start tier | $2,000 | 2026-10-20 | 2027-10-20 | GCP + Gemini | Marketplace third-party models | $180 | 2027-09-20 | Yes budget + 50%/90% |
Cloudflare Startups $10k | $10,000 | 2026-10-22 | 2027-10-22 | Workers, R2 (cap), KV | AI Gateway, registrar | $40 | 2027-09-22 | Yes |
Claude Startups (waitlist) | $1,000 API when open | TBD | TBD | First-party Claude API | Bedrock / Vertex / Azure | $0 until approved | TBD | N/A |
Review the sheet on the first Monday of each month. If burn is so low that you will leave money on the table, either ship harder experiments or stop applying for more of the same category.
What happens at the credit cliff?
Credits expire or run out. Then you pay list price on the same architecture. Plan for that before day one.
Set budgets and billing alerts the day credits land, not the week they die. Hard caps where the provider offers them; email alerts at 50% and 90% everywhere else.
Estimate the post-credit monthly bill from the last 30 days of usage at full price. Write that number next to the cliff date.
Do not build a feature you cannot afford at full price. A free GPU month that creates a $2,000 habit is a trap.
Avoid lock-in for the sake of free. If the only reason you are on a platform is credits, schedule a paid-price check before renewal.
When the bill needs to stay tiny after credits (or instead of chasing more programs), the SideHunt guide on keeping a side project's monthly bills under $20 is the cost-control companion. If users can burn your model budget, put a meter on the feature before someone else runs it up, as in this DEV guide to metering an AI feature.
A 14-day plan to get your first credits
Day | Focus | Done when |
|---|---|---|
1-2 | Pick primary stack | One cloud + one model path written down; no five-cloud list |
3 | Domain email + live site | founder@domain works; homepage explains the product |
4 | Billing hygiene | Paid accounts on the right email; cards on file; IDs copied to the sheet |
5 | Write the one-paragraph pitch | Who, job, AI core vs bolt-on, stage, funding honesty |
6-7 | Apply primary cloud | AWS Founders / Google Start / Microsoft application submitted |
8 | Apply edge if needed | Cloudflare (or skip if not in the stack) |
9 | Model path | Anthropic newsletter signup if closed; OpenAI only if you have a real VC code; else pay-as-you-go with a hard spend cap |
10 | Build the tracker | Sheet with columns above; calendar reminders for cliff dates |
11-12 | Alerts | Budgets / alerts at 50% and 90% on every approved account |
13 | Post-credit estimate | One number for "what this costs next month without credits" |
14 | Ship something on the credits | One real workload burning a little credit on purpose |
What traps wipe startup credits?
Credits expiring unused. Mark the confirmation email date. Cloudflare states credits last one year or until consumed, with no extensions. DigitalOcean also says it does not grant extensions.
Multiple accounts and abuse. Duplicate applications and account hopping are a ban path. Cloudflare is explicit that the program is one-time per startup and credits are non-transferable.
GPU and model exclusions. Assume GPUs and Marketplace/third-party models are out until the official page says they are in.
Thinking cloud credits cover OpenAI or Anthropic on that cloud. Often they do not. Google FAQ: third-party models are billed directly and not covered. Anthropic: Claude Startups credits are first-party Console only.
Reapplying blindly. Some programs allow a higher tier later (AWS Founders then Portfolio with a real Org ID). That is different from spamming the same Form again after a rejection without fixing the reason.
Spreading across five clouds. You get five cliffs and five places to miss an alert.
FAQ
Do I need to be incorporated?
It depends on the program. NVIDIA Inception requires incorporation. Microsoft's higher credit milestones need business verification as a registered legal entity. AWS Activate and Google for Startups Cloud Program care more about website, email, stage, and funding type, but you still answer company questions honestly. If you are pre-incorporation, start with programs that allow self-funded early teams and incorporate before you chase tiers that require a legal entity.
Can I get credits without VC backing?
Yes, at smaller tiers. AWS Activate Founders, Google Start tier (up to $2,000), Microsoft for Startups entry credits, and Cloudflare's $10k bootstrapped tier are built for that. The six-figure headlines usually need a partner. OpenAI's published credit path for startups is through partner VCs.
Do cloud credits cover OpenAI or Anthropic models on Bedrock, Vertex, or Azure?
Do not assume they do. Google's startup FAQ says program credits cover Google Cloud services and that third-party models (including Marketplace) are billed directly and not covered. Anthropic states Claude Startups credits apply only through Claude Console, not AWS, Google Cloud, or Azure. Always read the current program exclusions for your exact model path.
What if Anthropic or another program is closed right now?
Believe the page. As of 13 October 2026, Claude Startups said the current cohort is full and new applications are not open. Put a reminder to check again, keep a spend cap on pay-as-you-go, and do not buy "guaranteed approval" services.
Get the applications in, track the cliff, and spend the credits on shipping. When you are ready to show an early AI product to people who hunt new tools, launch your AI product on EarlyHunt.