How to Get $300K in Cloud Credits (Microsoft, AWS, GCP)
- The published ceilings as of August 2026: up to $200,000 from AWS Activate, up to $150,000 from Microsoft for Startups, up to $350,000 from Google for Startups Cloud for AI-first companies. Every top tier is gated behind a partner relationship, not a better application form.
- I secured $300K in cloud credits across Microsoft for Startups and AWS Activate, plus $100K from LiveKit. What moved the number was demonstrated high-volume production usage on the platform, not a pitch deck about future usage.
- Credits are a bridge, not a business model. Model your unit economics at list price and treat the balance as runway, not margin. Then a burn chart never doubles overnight in front of your board.
How much can a startup actually get in cloud credits?
As of August 2026 the published ceilings are up to $200,000 from AWS Activate, up to $150,000 from Microsoft for Startups, and up to $350,000 from Google for Startups Cloud for AI-first companies. Almost nobody receives the top number through a self-serve form. The gap between the headline and the median award is what this post is about.
The self-serve floors are honest and small. AWS Activate Founders, for bootstrapped and self-funded companies, starts at $1,000 with select participants qualifying for up to $5,000, per the AWS Activate Credits page. Google for Startups Cloud publishes a $2,000 pre-funded tier for building an MVP. These amounts are real and arrive fast. They cover a few months of a small staging environment. They will not change your runway.
The ceilings are a different product. AWS Activate Portfolio publishes up to $200,000 but states plainly you need an Organization ID from an Activate Provider: an accelerator, angel investor or venture capital firm. Google publishes $200,000 on the standard early-stage track and up to $350,000 for AI-first startups from seed to Series A. Microsoft advertises up to $150,000 on its Microsoft for Startups landing page. AWS also lists an invite-only tier above Portfolio, described as $200,000-plus in AWS Credits for AI Startups, reached through an account manager rather than an application.
I have been on the receiving end of two of these. I secured $200K through Microsoft for Startups and $100K through AWS Activate, plus $100K in LiveKit credits for a voice workload. Both cloud numbers have since moved. AWS Activate Portfolio now publishes a higher ceiling than the award I received; Microsoft advertises a lower one. That is the first useful lesson here: the published number is a snapshot of a program repriced every few quarters, and what you actually get is negotiated against evidence, not read off a page.
Want to know what those credits are worth against your real infrastructure, not a hypothetical one? Run the shape of your bill through the cloud cost calculator first. Credits on a badly architected bill just delay the reckoning, which is the argument the whole cloud cost optimization practice rests on.
Who is actually eligible for cloud credits?
AWS publishes the hardest gates: pre-Series B, founded within the last 10 years, an AWS account on a paid support plan, and either new to Activate Credits or requesting more than you previously received. Google targets seed to Series A on its early-stage track. Microsoft verifies the business. Eligibility is cheap to satisfy; the tier is the contested part.
Read the AWS list again, because three of those four conditions catch people. Pre-Series B means the window closes permanently at a funding event, so a company planning a Series B in six months should apply this month, not after the round. Founded in the last 10 years quietly excludes the services business that pivoted into software. And the paid-support-plan requirement means the free-tier account you opened to experiment does not qualify until you upgrade it, a small monthly cost people discover only after rejection.
The clause that matters most is the last one: you must be new to Activate Credits, or asking for more than you previously received. That converts the program from a one-shot lottery into a ladder. Taking the $1,000 Founders award today does not disqualify you from a Portfolio award later; it only means the later request has to be larger. Having already used credits at the small tier is helpful evidence when you come back, because you now have a usage history on the platform.
Google's AI track adds a product condition, not a corporate one: you use or plan to use Google's AI services as a foundation of your primary product, alongside qualifying venture funding from seed to Series A and not having already received more than a small amount of Google Cloud credit. Microsoft's public page is the least specific of the three. It advertises the ceiling and routes you into a verification flow rather than publishing a tier table, and it lists separate specialty programs for startups that fit particular partnership profiles.
The practical read across all three: eligibility is a filter, not a competition. Pass the gates and you are in the pool. Everything that follows is about where in the pool you land, and that is decided by evidence and by who introduces you.
| AWS Activate | Microsoft for Startups | Google for Startups Cloud | |
|---|---|---|---|
| Self-serve floor | $1,000, up to $5,000 for select participants | Verification-based entry tier | $2,000 pre-funded MVP tier |
| Published ceiling | Up to $200,000 (Portfolio) | Up to $150,000 | $200,000, up to $350,000 AI-first |
| What gates the ceiling | Org ID from an Activate Provider | Verification, progress and partnership profile | Qualifying VC funding, seed to Series A |
| Stage window | Pre-Series B, founded within 10 years | Not published as a hard stage gate | Seed to Series A on the early-stage track |
| Account prerequisite | AWS account on a paid support plan | Business verification | Google Cloud account |
| Repeat applications | Allowed if requesting more than before | Credits unlock progressively | Limited if you already hold credits |
| Tier above the published ceiling | Invite-only AI startups tier, $200,000+ | Specialty programs, terms not public | Series B+ custom arrangements |
| Named provider network | a16z, Brex, Carta, Greylock, Sequoia, Y Combinator and others | Accelerator and investor partners | Google for Startups accelerator network |
Why is a cloud credit application a pitch, not a form?
Because a human at the provider is deciding how much of a finite budget to allocate to you, and the only thing they can act on is the evidence you give them that your consumption will grow. The form collects facts. The free-text fields are where the decision is actually made, and most applicants write two sentences in them.
Think about the incentive on the other side of the desk. Credits are a customer acquisition cost. The provider is buying the probability that you become a paying account that grows for years, and every award is a bet with a known cost and an uncertain return. An application that says 'we are building an AI product and need infrastructure' gives them nothing to underwrite. One that says 'we currently process 40,000 inference requests a day, our infrastructure spend has grown 18% month on month for six months, and here is the specific architecture we would move onto your platform' gives them a number for the spreadsheet.
The strongest applications I have written and reviewed share four elements. They state current production volume in a unit the provider recognises: requests, minutes, gigabytes, tokens. They state current spend and the trend. They name the specific services they will consume, because that maps to the provider's own product goals. And they name a migration or expansion the credit unblocks, with a date attached.
The fourth element is the one people leave out, and it is the most persuasive. A credit that unblocks a decision is worth more than one that subsidises an existing bill, because it changes where your workload lives permanently. When I applied to Microsoft for Startups, the case was that a specific set of production workloads would move onto Azure and stay there, with the volume figures to prove what that was worth. That is a commercial proposal, not a request for help.
Write it as if you were raising a small round from a strategic investor whose return is measured in your future invoice. Because that is exactly what it is.
- Current production volume in the provider's own unitRequests per day, minutes processed, GB stored, tokens per month. Not user counts.
- Current monthly infrastructure spend, plus the trend over the last six monthsGrowth rate is the single most persuasive number you have
- A named list of the specific services you will consumeMaps directly onto the provider's internal product objectives
- A concrete workload the credit unblocks, with a dateA migration or a launch beats a general subsidy every time
- Evidence you are already running in production somewhereDashboard screenshots, a status page, a customer logo: anything verifiable
- The Org ID from your accelerator, angel or VC if you have oneAWS states plainly that Portfolio requires it
- A named technical owner who will answer the follow-up questionsThere is usually a follow-up; a fast answer keeps you in the queue
- An honest projection of steady-state spend after the credit is exhaustedProviders underwrite the post-credit account, not the credit period
What actually unlocks the top credit tiers?
Demonstrated high-volume production usage. Not a deck, not a projection, not a logo wall. The strongest signal you can send is a real workload already running at scale somewhere, including on a competitor's cloud, because it converts your application from a bet on a plan into a bet on a migration.
This is counterintuitive to founders, who assume credits are for companies that have nothing yet. At the top tiers the opposite is true. Nothing you write persuades a provider like a graph of production traffic. If you serve a hundred thousand inference calls a week, or thirty thousand voice minutes a month, or half a terabyte of new object storage a month, put that number in the first paragraph. The provider's account team can size the annual value immediately, and that number decides whether your file goes to the self-serve pile or to a human.
The second unlock is the Org ID. AWS is unusually direct: the Portfolio tier requires an Organization ID from an Activate Provider, and it names a16z, Brex, Carta, Greylock, Sequoia and Y Combinator among thousands in the network. If you have raised from anyone, ask whether they are an Activate Provider before you apply. The answer is yes far more often than founders expect, and it is the difference between a five-thousand-dollar award and a two-hundred-thousand-dollar one. Accelerators, incubators and even some cap table and banking platforms sit in these networks.
The third unlock is being the kind of workload the provider is currently trying to win. Google's AI-first tier is nearly double its standard tier, conditioned on using its AI stack as the foundation of your product. AWS lists an invite-only credits tier for AI startups above Portfolio. These are not neutral programs; they are strategic budgets aimed at specific product lines, and applications that sit inside those lines get treated differently. If your architecture genuinely fits, say so in the provider's own vocabulary.
The fourth unlock is boring and works: ask for a call. Credit programs are staffed. Once your volume figures sit in front of a startup account manager rather than a queue, you are in a commercial conversation where the number is negotiable and the follow-on support (architecture review, migration assistance, occasional additional credit) is often worth more than the headline award.
Real traffic, real spend, real growth rate. The only evidence that converts a plan into a migration in the provider's model.
highestAWS states Portfolio requires it. Your existing investor, accelerator or even your cap table platform may already be one.
gatingGoogle publishes a higher AI-first ceiling; AWS runs an invite-only AI startups tier. Sitting inside the product line being funded changes the number.
largeTurns a queued form into a negotiation, and usually adds architecture support that outlasts the credit.
compoundingHow do you get cloud credits without a VC or an accelerator?
Take the self-serve floor now, build measurable production usage on it, then come back. AWS explicitly permits repeat applications provided you request more than you previously received, which turns the small award into the first rung of a ladder rather than a consolation prize.
The bootstrapped route is slower but not closed. Start with AWS Activate Founders, designed for exactly this profile and requiring no Org ID. Take Google's pre-funded MVP tier. Complete Microsoft's verification flow. None of these changes your runway individually, but together they buy you six to twelve months of non-production environments for free, and, more importantly, they create the usage history a later application can point at.
In parallel, widen the definition of a partner. The Activate Provider network is much broader than VC firms: AWS names banking and cap table platforms alongside investors. Startup banking accounts, equity management platforms, industry accelerators, university programs, government innovation schemes and some SaaS partner programs all sit in one or more of these networks. It costs an email to ask each of yours whether they can issue an Org ID.
Cloud providers are not the only source of credits. I secured $100K in LiveKit credits alongside the cloud awards, because infrastructure vendors one layer up the stack (real-time media platforms, vector databases, observability tools, model providers, CDNs) run their own startup programs with far less competition and far shorter applications. On a voice AI workload, a media infrastructure credit can be worth more than a compute credit, because that is where the marginal cost lives. The economics of that stack are in how I cut a $200K/year cloud bill by more than 70%.
Finally, be patient about timing the big ask. The pre-Series B window is years wide. There is no advantage to burning your one large AWS application in the month you have the least evidence, and real advantage to spending two quarters generating a usage graph first.
AWS Activate Founders starts at $1,000 with up to $5,000 for select participants; Google publishes a $2,000 pre-funded tier. Take all of them, spend nothing on the application, and come back in two quarters with usage.
AWS states Portfolio requires an Organization ID from an Activate Provider, and the network includes thousands of firms. This one email is worth more than every other action in this list combined.
Demonstrated high-volume usage is the strongest non-partner signal there is. Lead with requests per day, spend, and growth rate, and request a conversation with a startup account manager.
Google publishes up to $350,000 for AI-first startups against $200,000 standard, and AWS lists an invite-only AI startups tier above Portfolio. Applying to the generic track first wastes your one shot.
AWS Activate is explicitly pre-Series B. At this stage the lever is a committed-spend agreement with a discount, not a credit award, and that is a procurement conversation.
Can you stack cloud credits across AWS, Microsoft and Google?
Yes. The programs are independent and nothing stops you holding awards from all three at once. What you cannot do is spend one provider's credit on another provider's bill, so stacking only helps if your architecture can genuinely be split across clouds without the split costing more than it saves.
This is where most stacking advice becomes irresponsible. A second cloud adds real, permanent overhead: a second IAM model, a second set of deployment pipelines, cross-cloud data transfer billed per gigabyte in at least one direction, duplicated observability, and an on-call surface that just doubled. Split a tightly coupled application across two clouds to chase a credit and you will spend the credit on integration and inherit the complexity forever.
Stacking works when the boundaries are already natural. Independent workloads with their own data (a batch analytics pipeline, a media processing job, a model training run, a staging environment, a set of internal tools) sit on a second cloud with almost no coupling cost, because they talk to the rest of your system across an API rather than a shared database. That is how I ran it: GCP to Azure and AWS, each workload placed where its pricing and its credits were best, executed solo with zero downtime.
The one part of stacking that is unambiguously free is the part nobody does. Non-production environments (CI runners, integration test clusters, load-testing rigs, demo environments) have no data gravity, no uptime requirement, and no user impact if they move. Putting the entire non-production estate on a credited second cloud costs a day of pipeline work and removes a whole category of spend from your primary invoice for as long as the credit lasts.
The honest ceiling on stacking is your team's attention, not the credit total. One engineer can hold two clouds in their head. Three is where the cost of context switching starts to show up in incident response times, and no credit balance compensates for that.
- Primary API on cloud A, primary database on cloud B
- Cross-cloud egress on every request, permanently
- Kubernetes cluster split across providers
- Two control planes, two IAM models, one on-call rota
- Shared cache accessed from both clouds
- Latency and per-GB transfer on the hottest path you own
- Observability duplicated per provider
- Two bills, two dashboards, no single trace
- Team of three holding three clouds
- Slower incident response, forever
- Entire non-production estate on the credited cloud
- Zero data gravity, one day of pipeline work
- Batch analytics and model training jobs
- Read once, write results back, no chatty coupling
- Media and transcoding pipelines
- Self-contained, queue in and object storage out
- Cold archive and DR copies
- Written once, read approximately never
- Internal tools and admin surfaces
- No user-facing latency budget to protect
What are the expiry and coverage traps with cloud credits?
Every credit has an expiry date printed on the award, credits generally cannot be converted to cash or transferred, and a meaningful share of your bill is usually not eligible at all. The trap is not the expiry itself. It is building a cost structure during the credited period that you cannot afford at list price on the day it ends.
Start with what credits do not cover, because it surprises people every time. AWS states that all Activate Credits are subject to its Promotional Credit Terms and Conditions, and promotional credits across providers commonly exclude third-party software bought through a marketplace, some support plan fees, taxes, and certain reserved or committed-spend purchases. So the exact line items you would normally optimise first (a committed-use discount, a Savings Plan, a third-party observability tool billed through the marketplace) may sit outside the credit entirely. Read the terms attached to your specific award, not a summary blog post, including this one.
Then look at the expiry structure. Awards carry a defined validity window and unused balance disappears at the end of it. Some programs also stage the award, releasing a portion up front and the rest on verification or a usage milestone. Two consequences follow. First, an unused credit balance is not savings; it is an asset with a hard maturity date, and if you are consuming it more slowly than it decays, move workloads onto it faster. Second, if your award is staged, the second tranche is conditional, so a decision you make in month three can cost you money in month seven.
The trap that actually damages companies is the cliff. Credits mask your true unit economics for the whole period they are active. A team on a large award makes exactly the architectural choices credits make painless (per-request managed services everywhere, no lifecycle policies, generous log retention, oversized non-production environments) and then discovers on expiry day that its cost per customer was never viable. The bill does not gradually rise. It steps.
The defence is a single discipline: track two numbers every month, your credited spend and your list-price spend, and manage the business on the second one. If list-price cost per unit of value is falling while the credit runs, you have used the credit correctly. If it is flat, you have bought time and nothing else.
- Day 0Read the actual award terms
Note the exact expiry date, whether the award is staged, and which categories are excluded. Marketplace purchases, support plans, taxes and committed-spend purchases are common exclusions. Put the expiry in the company calendar, not just your inbox.
- Week 1Instrument list-price spend separately
Set up cost reporting that shows what the bill would be with no credits applied. Every provider can show gross usage cost. This single dashboard is what prevents the cliff.
- Month 1Move the highest-value workloads onto the credit first
An expiring asset should be consumed in order of value. Non-production and batch workloads are the fastest to move; user-facing services move later and more carefully.
- Months 2 to 6Optimise as if the credit did not exist
Run the standard hygiene loop (commitments, per-gigabyte meters, lifecycle policies, log levels) on the uncredited number. This is the phase most credited startups skip entirely.
- Expiry minus 6 monthsModel the post-credit bill and decide
You now know your list-price run rate. Either it is affordable, or you have two quarters to re-architect, or you apply for the next tier with a much better usage story than you had last time.
- Expiry minus 60 daysExecute the plan, do not renegotiate under pressure
Providers can sometimes extend or add credit, but a request made from a position of dependence is a weak one. A request made alongside a growing paid account is a strong one.
How did I actually secure $300K in credits?
By leading with production volume rather than a plan. I secured $200K through Microsoft for Startups and $100K through AWS Activate, plus $100K in LiveKit credits, and in every case the argument was the same: here is a workload already running at scale, here is what it costs today, here is exactly what moves onto your platform and when.
The credits were one lever in a larger piece of work: cutting an equivalent $200K-per-year enterprise cloud bill by more than 70%, through a multi-cloud re-architecture I executed solo with zero downtime. Be precise about how those two facts relate. The steady-state saving came from architecture: replacing per-request managed services with reserved compute, removing per-gigabyte network paths, and placing each workload where its pricing was best. The credits sat on top and bought time. They did not create the saving, and I would not present them as if they had.
The Microsoft application was the largest award and the most commercial conversation. What carried it was specificity: named workloads, current volumes, a migration date, and a projection of the account at steady state after the credit was exhausted. Founders instinctively hide that last part. The provider is underwriting the post-credit account. Telling them what it looks like is not a weakness in your application; it is the substance of it.
The LiveKit credit is the one I would point a voice or media company at first. Vendor credits one layer up the stack are less contested, faster to obtain, and often land directly on your dominant marginal cost. Running voice AI at roughly 2.5 cents a minute in production, the media infrastructure layer is a large share of that number, so a credit there moves unit economics in a way a general compute credit does not.
What I would do differently: instrument list-price spend from day one. I spent longer than I should have with a bill that looked healthy because credits were absorbing it, and reconstructing the true unit economics afterwards took real time. The dashboard that shows what you would be paying without credits is a thirty-minute setup and it is the single most valuable artefact in a credited company.
When do cloud credits become a trap?
When they let you defer the question of whether your product works at list price. A credit that funds a viable business is runway. A credit that funds a business whose unit economics only close because the compute is free is a countdown, and the countdown is invisible until roughly sixty days before it ends.
There is a second, quieter trap: architectural lock-in bought cheaply. Credits are a customer acquisition tool, and they work. Two years of free managed services produce a codebase wired to one provider's proprietary surfaces, and the exit cost at expiry can exceed the entire value of the award. This is not an argument against taking credits. It is an argument for keeping your data in portable formats, your compute in containers, and your provider-specific dependencies behind interfaces you could reimplement. The mechanics of paying that bill later are in what a GCP to AWS migration costs and what breaks.
A third trap is opportunity cost on the application itself. A serious Portfolio-tier application, with the volume data assembled and the follow-up call prepared, is a couple of days of founder time. That is good value against a six-figure award and terrible value against a five-thousand-dollar one. If you are structurally in the self-serve tier this quarter, spend twenty minutes on the self-serve form and put the rest of the time into building the usage that makes next quarter's application work.
The test I apply before advising anyone to chase a large award is simple. Take your current bill, remove every credit, and ask whether the business still works at that number within two quarters of plausible growth. If yes, credits are pure runway and you should take every one you can get. If no, the credit is not the intervention you need. The architecture is, and that is what the AWS cost optimization checklist exists for.
None of this is an argument against applying. Apply to all three. Take the self-serve floors today, work the Org ID route hard, and lead every application with production volume. Just do the arithmetic on the uncredited bill in parallel, because that is the number your business actually runs on.
Cloud credits: common questions
→How much can a startup get in cloud credits?
As published in August 2026: AWS Activate advertises up to $200,000 on the Portfolio tier and up to $5,000 self-serve on Founders; Microsoft for Startups advertises up to $150,000; Google for Startups Cloud publishes $200,000 standard, up to $350,000 for AI-first startups from seed to Series A, and $2,000 for pre-funded companies. The top tiers of all three require a partner relationship or qualifying funding.
→Do you need a VC to get AWS Activate credits?
Not for the Founders tier, which is designed for bootstrapped and self-funded companies and starts at $1,000 with up to $5,000 for select participants. You do need an Organization ID from an Activate Provider for the Portfolio tier, but that network is much wider than venture capital and includes accelerators, angel investors and startup banking and cap table platforms, so it is worth asking every partner you already have.
→Can you use AWS, Microsoft and Google credits at the same time?
Yes, the programs are independent and holding awards from all three is permitted. Each credit only offsets that provider's own bill, so stacking only helps if you have workloads with natural boundaries: non-production estates, batch analytics, media pipelines, archives. Splitting a tightly coupled application across clouds to chase a credit usually costs more in integration and cross-cloud transfer than the credit is worth.
→Do cloud credits expire?
Yes. Every award carries an expiry date and unused balance is forfeited, not refunded or transferred. Awards are also sometimes staged, with later tranches conditional on verification or usage milestones. AWS states that all Activate Credits are subject to its Promotional Credit Terms and Conditions. Read the terms attached to your specific award rather than any summary, because coverage exclusions and validity windows differ between programs and change over time.
→What do cloud credits not cover?
Coverage varies by program and is defined in the promotional credit terms, but common exclusions across providers include third-party software bought through a marketplace, some support plan fees, taxes, and certain reserved or committed-spend purchases. That last one matters more than it sounds: it can mean the discount instruments you would normally buy first are outside the credit entirely.
→Are cloud credits worth applying for if you are pre-revenue?
Take the self-serve tiers, which cost twenty minutes, and skip the large application. The top tiers are unlocked by demonstrated high-volume production usage or a partner Org ID, and a pre-revenue company has neither. Spend the two days you would have spent on a Portfolio application building the usage graph that makes the same application succeed two quarters later.