$1M is the upper-tier award from the AWS Generative AI Accelerator — granted to roughly 5% of accepted startups, which corresponds to the standout cohort companies, typically Series-A AI-native startups with named institutional backing and demonstrated commercial trajectory. The median accelerator outcome is $300K; the floor is $200K; the $1M tier is the cohort outlier award. This page covers the acceptance funnel, the eligibility signals AWS reviewers look for at the $1M tier, the tranche structure for a million-dollar award, and the stacking ceiling at $1.15M when the standard $150K credit stack layers on top.
The "$1M AWS credits" figure originates from the Generative AI Accelerator marketing language ("up to $1M in credits"). The ceiling is real. It is also the highest single award AWS issues under any credit program — there is no $2M or $5M tier, and the standard partner-filed stack tops out at $150K. Understanding what $1M actually means at the program level explains who reaches it and who does not.
The Generative AI Accelerator was introduced in 2024 as the AWS-funded counterweight to OpenAI direct, GCP Vertex AI, and Azure OpenAI Service capturing AI-native startups during their inference-platform decision window. The economic logic at the $1M tier is specifically platform-defining: AWS calculates that a $1M credit award to a Series-A AI startup with substantial commercial trajectory translates into roughly $5M–$15M in Bedrock revenue over the credit-funded period and the multi-year runway that follows, plus the case-study and reference-customer value of having a recognized AI company publicly anchored on Bedrock.
The award sizes follow a deliberate distribution. AWS calibrates per-cohort budget around the assumption that the median selected startup will receive $300K (the practical center calibrated for 18–24 months of typical AI-startup Bedrock consumption), with the upper tier reserved for the small number of cohort selections where the commercial case justifies a larger investment. Across publicly-disclosed cohort patterns, the tier distribution lands at roughly 15% floor ($200K), 30% lower-median ($250K), 40% median ($300K), 10% upper-median ($500K), and 5% ceiling ($1M).
Translated to absolute numbers per global cohort: of approximately 50 selected startups, 7–8 receive the floor, 15 receive the lower-median, 20 receive the median, 5 receive the upper-median, and 2–3 receive the $1M ceiling. The $1M tier is not rare in the sense of being inaccessible; it is rare in the structural sense of corresponding to a small absolute number of startups per cohort. A founder reading "$1M AWS credits" pages who treats the ceiling as the expected outcome is structurally misreading the program.
The honest framing of the $1M tier is: it exists, it is meaningful, and it is reserved for the specific profile of Series-A AI-native company with substantial commercial signals that AWS treats as platform-defining. The realistic target for the typical AI-native applicant is acceptance at the $300K median. Acceptance into the accelerator at any tier is the meaningful win; the tier within acceptance is determined by signals that are largely set at the time of application and not influenceable through better positioning of an otherwise mid-tier company.
The Generative AI Accelerator operates as a funnel with three stages of selection: application screening (eligibility filter), interview shortlisting (architectural and commercial depth filter), and cohort selection (final award including tier assignment). Each stage narrows the population. The $1M tier sits at the narrowest point.
Stage 1 — Application volume. AWS does not publicly disclose total application volume per cohort, but per partner channel reports, each quarterly cohort receives roughly 800–1,200 applications globally. The volume has trended upward since 2024 as awareness of the program has spread; cohorts in 2026 receive closer to the upper end of the range.
Stage 1 outcome — Initial screening. AWS reviewers apply an eligibility filter against the application: AI-first product (not AI-augmented), pre-Series-B funding stage, named Bedrock commitment, founding team with relevant AI/ML credibility. Approximately 50% of applications pass the initial screening; the failures are typically AI-augmented startups (the product is not AI-first), Series-B+ companies, or applications without a specified Bedrock commitment.
Stage 2 — Interview shortlisting. Applications that pass the initial screening enter a deeper review where AWS's Generative AI Accelerator team evaluates architectural depth, commercial trajectory, and team execution credibility. Roughly 15–20% of screened applications progress to a 30-minute interview with a member of the Bedrock team. The interview phase is the primary qualitative filter; the questions cover Bedrock migration plan specificity, model selection rationale, projected consumption breakdown, and demonstrated commercial signals.
Stage 3 — Cohort selection. Of the applications that complete the interview phase, approximately 25–30% are selected for the cohort. Total selected globally per cohort: ~50 startups. The acceptance rate, calculated as selected / total applicants, lands at roughly 5%. Calculated as selected / interviewed, the rate is closer to 25–30%. The relevant rate for an applicant's expected outcome is the 5% figure — most applications do not reach the interview phase, and most interviewed applications are not selected.
Stage 4 — Tier assignment within acceptance. Selected cohort members receive a tier assignment as part of the formal offer letter. Tier is determined by AWS's evaluation of projected Bedrock consumption, commercial trajectory, and platform-defining potential. The 5% of selected startups that receive the $1M ceiling are determined at this stage; the distribution across tiers is largely set by signals visible in the application and interview, not negotiable post-selection.
The cumulative funnel math: 1,000 applications → 500 pass screening → 75 reach interview → 50 selected → 2–3 receive the $1M tier. The absolute probability of any single applicant receiving the $1M tier, conditional on submitting an application, is approximately 0.25%. The probability of acceptance at any tier is approximately 5%. The relevant target for an applicant is acceptance, not tier.
Within the population of accepted cohort members, the tier assignment is determined by specific signals visible in the application and interview. The signals that distinguish the $1M tier from the median tier are knowable in advance and largely fixed by the company's stage at the time of application.
Funding stage and named institutional backing. The $1M tier consistently corresponds to Series-A AI-native companies with named institutional VC backing — Andreessen Horowitz (a16z), Sequoia, Founders Fund, Lightspeed, Index, Accel, Benchmark, Greylock, NEA, or comparable tier-1 firms. Series-A companies without named backing reach the upper-median ($500K) tier; seed-stage companies typically reach the median ($300K) tier regardless of other signals. The pattern reflects AWS's implicit calibration: a Series-A round with a tier-1 lead signals an external validation event that AWS reviewers treat as predictive of commercial trajectory.
AI-first product profile. All accepted startups are AI-first; the $1M tier additionally requires that the product's core value proposition is something that did not previously exist and was made possible specifically by foundation-model capabilities. Examples that map to the $1M tier: AI-native coding tools where the product is the LLM (Cursor-tier scale); AI-native design tools where the user-facing experience is generation (Lovable-tier products); AI-native infrastructure that provides inference, evaluation, or orchestration as the primary product. Examples that map to the median tier: SaaS products that have added AI features; consumer products that use AI for personalization; vertical AI tools where the AI is one component of a larger product.
Demonstrated traction at scale or LOI commitments. The $1M tier requires demonstrated commercial signals beyond "we have users." Specific signals that map: monthly active user counts in the tens of thousands or higher; revenue at $50K+ MRR with growth visible across the prior 6 months; named enterprise LOI commitments from companies the reviewer recognizes; or analogous commercial validation. Companies at the median tier typically have meaningful but smaller signals — early users, small revenue, design-partner agreements.
Credible projection of $1M+ Bedrock spend over 24 months. The $1M credit award is calibrated against projected Bedrock consumption. AWS reviewers look for a documented projection showing $40K–$50K per month or higher Bedrock consumption within 12 months and continuing scale through month 24. The projection must be defensible: tied to the company's growth model, broken down by inference workload, validated against current consumption trajectory if applicable. Vague projections ("we'll consume substantial credits") map to the median tier; specific projections backed by current consumption data map to the upper tiers.
Bedrock-team engagement signals. The $1M tier consistently corresponds to startups that AWS's Bedrock product team treats as platform-defining. Practical signals include: existing direct relationships with Bedrock product managers; participation in private Bedrock preview programs; co-marketing or case study commitments already in motion; willingness to be the public reference customer for a Bedrock feature. The Bedrock team's endorsement of the application during the interview phase is the most predictive variable for the $1M tier — applications that the Bedrock team flags as platform-defining typically receive the ceiling award.
Accelerator credits issue in tranches against milestones. The standard $300K median award uses a 3-tranche structure across 6 months ($100K per tranche). The $1M ceiling award uses a 3–4 tranche structure across 6–12 months ($250K–$333K per tranche), tied to higher-stakes milestones that reflect AWS's expectation of platform-defining engagement.
The tranche structure for a $1M award is negotiated as part of the acceptance offer letter. Specific tranching depends on the company's stage and projected Bedrock ramp, but the typical structure follows a 4-tranche pattern at $250K per tranche across 6–12 months. The milestones are designed to align the credit issuance with the company's demonstrated Bedrock production engagement, ensuring that the $1M investment maps to actual Bedrock consumption rather than sitting idle in an account.
Tranche 1 — Acceptance ($250K). Issued at the time of formal acceptance into the cohort. The first tranche is the entry-level credit allocation that allows the startup to begin Bedrock POC work and supporting infrastructure provisioning. There is no milestone gate on the first tranche beyond acceptance itself; the credit is applied to the AWS account within 2 weeks of the formal offer letter.
Tranche 2 — Bedrock POC in production at 60 days ($250K). The second tranche releases when the startup has demonstrated a Bedrock workload running in production traffic. The specific bar is: foundation-model inference (typically Claude Sonnet or comparable named model) is serving production user traffic, not just internal evaluation. AWS reviews production deployment evidence — typically a brief technical writeup plus CloudWatch metrics showing real inference volume — and releases the tranche on approval. The 60-day window is calibrated to be achievable for a focused team; the actual median time-to-production-deployment for $1M-tier startups is closer to 45 days.
Tranche 3 — Commercial outcome demonstrated at 120 days ($250K). The third tranche releases when the startup has demonstrated a commercial outcome attached to the Bedrock workload. The outcome can take several forms: revenue specifically attributable to the Bedrock-powered feature, user growth where the Bedrock feature is the conversion driver, enterprise contracts where the Bedrock capability is the differentiating factor, or analogous commercial signals. AWS reviews the outcome evidence — a brief writeup, supporting metrics, customer references where applicable — and releases the tranche on approval.
Tranche 4 — Case study and continued scale at 240 days ($250K). The fourth and final tranche releases when the startup has participated in AWS's public case study program and demonstrated continued scale on Bedrock. The case study commitment is the explicit reciprocal: AWS funds the $1M credit award; the startup publishes a joint case study, participates in re:Invent or comparable AWS event content, and serves as a public reference customer for Bedrock. The continued-scale requirement ensures that the Bedrock workload has grown beyond the initial POC into substantive production consumption — typically measured at $40K+/month inference spend by month 8.
Tranche forfeiture risk. Missing a milestone forfeits the corresponding tranche. A startup that fails to reach Bedrock production by day 60 forfeits tranche 2 ($250K) and is typically downgraded or removed from the program for subsequent tranches. The forfeiture mechanic is enforced — AWS has tightened this through 2025 and 2026 as the program has scaled — and the realistic expected value of a $1M tier acceptance for a startup that does not execute on the milestones is closer to $250K–$500K depending on which tranches are forfeited.
The $1M Generative AI Accelerator award stacks cleanly with the standard partner-filed $150K credit stack. The two programs are operated by separate AWS teams under separate program rules; awards in one do not reduce eligibility in the other. A startup that receives the $1M ceiling can pursue the standard stack in parallel and reach a $1.15M combined credit position.
The structural reason the two paths do not conflict: the Generative AI Accelerator is operated by AWS's Generative AI Accelerator team under a dedicated cohort budget; the standard stack (Activate Portfolio + Build for Startups + Bedrock POC) is operated by AWS Activate under the general partner-filed program budget. The two budgets are separate, the two reviewer pools are separate, and the eligibility rules do not include exclusivity language that would prevent simultaneous participation.
The practical mechanic for a startup pursuing both: the standard stack is filed by an AWS partner via the ACE program in the same week the accelerator application is submitted. The standard stack lands within 11–18 days at $125K–$150K. The accelerator application proceeds on its 60–90 day timeline. If the accelerator returns acceptance at the $1M tier, the stack credits are already applied and consumed against ongoing AWS work; the $1M layers on top across the tranche window.
Day 0–18 — Standard stack execution. Partner-filed Portfolio ($100K) + Build for Startups ($25K) + Bedrock POC ($25K–$50K) applies to the AWS account. The startup spends against these credits for early infrastructure work and Bedrock POC execution. Combined: $150K–$175K applied within 18 days.
Day 60–90 — Accelerator acceptance and first tranche. The accelerator decision arrives at day 60–90. For $1M-tier acceptance, the first tranche of $250K issues at acceptance. Combined credit position at day 90: $400K–$425K (standard stack + first accelerator tranche).
Day 150 — Accelerator second tranche. Bedrock POC in production at day 60 of the accelerator clock unlocks the $250K second tranche. Combined position: $650K–$675K.
Day 210 — Accelerator third tranche. Commercial outcome at day 120 of the accelerator clock unlocks the $250K third tranche. Combined position: $900K–$925K.
Day 330 — Accelerator fourth tranche. Case study and continued scale at day 240 unlocks the final $250K. Combined position: $1.15M–$1.175M. This is the practical ceiling for stacked accelerator + standard stack.
The honest framing: the $1.15M combined position is theoretical for the small population of startups that (a) receive the $1M tier ($1M tier acceptance is roughly 0.25% of total applicants), (b) execute all four accelerator milestones without forfeiture, and (c) maintain standard stack eligibility through the same period. The expected combined position for the typical accelerator-accepted startup at the median tier is $300K accelerator + $150K stack = $450K, which is itself a substantial credit position. The $1.15M ceiling is structurally an outlier outcome, not a typical one.
A subset of AI startups pursue credits both from AWS (via Bedrock and the Generative AI Accelerator) and from Anthropic directly (via the Claude API). The two programs are structurally distinct, target different stages, and stack for startups that operate inference workloads across both Bedrock and Claude API.
Anthropic offers a set of standing credit programs to select AI startups, distinct from the AWS Generative AI Accelerator. The Anthropic programs target startups using Anthropic Claude via the Claude API directly (api.anthropic.com), not via Amazon Bedrock. The credit awards vary by program — Anthropic's startup program, Anthropic for Startups partnerships through select VCs, and discretionary awards for early-access partners — and the typical range sits at $25K–$250K with substantially smaller award sizes than the AWS Generative AI Accelerator.
The structural distinction matters for credit planning. A startup running production inference through Bedrock consumes against AWS credits (standard stack + accelerator tranches). A startup running production inference through Claude API directly consumes against Anthropic credits. A startup running mixed inference — some workloads on Bedrock, some on Claude API direct — can pursue credits from both pools and stack them, with the consumption mapping to wherever the underlying inference call lands.
The strategic question for an AI startup deciding between Bedrock and Claude API direct is partly a credit-availability question: the AWS Generative AI Accelerator path delivers materially larger credit awards ($1M ceiling vs $250K ceiling on the Anthropic side), but requires architectural commitment to Bedrock as the inference path. Anthropic's programs deliver smaller awards but with more flexibility around the architectural commitment — the startup runs Claude on whatever infrastructure they prefer, and the credits apply against direct API consumption.
Some AI startups stack both: route a portion of production inference through Bedrock (claiming AWS credits including accelerator tranches), route a portion through Claude API direct (claiming Anthropic credits), and treat the dual-path architecture as both a credit-optimization mechanic and an inference-availability redundancy mechanic. The dual-path stack is operationally more complex than single-path Bedrock and requires the application architecture to support routing across two API endpoints, but for startups at sufficient scale ($10K+/month combined inference spend) the credit math justifies the engineering effort.
The honest practical advice: most early-stage GenAI startups should commit to Bedrock as the primary inference path, pursue the standard $150K AWS stack and the Generative AI Accelerator track concurrently, and treat Anthropic credits as a secondary consideration to revisit at Series-A scale. The Bedrock-primary architectural commitment is what unlocks the $1M tier of the AWS accelerator; hedging across Bedrock and Claude API direct signals to AWS reviewers that the Bedrock commitment is partial and consistently maps to median-tier acceptance rather than the ceiling.
Most AI founders reading this page will not realistically qualify for the $1M tier. The relevant decision is not "should I pursue the $1M tier" but "is the Generative AI Accelerator worth pursuing for my company given that the realistic outcome is the $300K median." The answer for AI-native startups is consistently yes, for reasons beyond the credit value.
The $300K median accelerator outcome is itself a substantial credit position. Combined with the standard $150K stack, the median accelerator-accepted startup reaches a $450K credit position over a 90-day window — meaningful runway for Bedrock-funded experimentation and production deployment. The credit value alone justifies the application effort for any AI-native startup that fits the eligibility profile.
Beyond the credit value, the accelerator delivers benefits that do not appear on the credit ledger. Accepted startups receive direct engagement with the Bedrock product team, including roadmap visibility, early access to new model launches, and direct channels to AWS engineering for production support issues. The product-team engagement is typically more valuable in qualitative terms than the credit award for startups serious about long-term Bedrock-anchored architecture.
The go-to-market opportunities through the accelerator are meaningful for startups building products that target enterprise buyers. AWS's sales organization can be a co-selling channel for accelerator-accepted startups; the AWS Marketplace integration is smoother for accelerator cohort members; the joint case study and re:Invent visibility deliver outbound pipeline that startup-stage marketing teams cannot generate organically.
The mentorship component, while soft, has selected utility for first-time AI founders. The accelerator program assigns Bedrock-team mentors to cohort members; the mentorship covers architectural review, model selection guidance, evaluation methodology, and production-deployment best practices. The mentorship is valuable for teams without prior AI/ML production deployment experience; less valuable for teams that already have deep technical AI credibility.
The decision framework: pursue the accelerator if (a) the company is AI-native and fits the eligibility profile, (b) the team can commit to Bedrock as the primary inference architecture for 18–24 months, (c) the founder time required for the application is acceptable (typically 8–15 hours of focused founder time), and (d) the rejection-case outcome is acceptable (the standard $150K stack remains available regardless of accelerator outcome). For most AI-native startups that meet these conditions, the accelerator is the highest-expected-value credit path available regardless of whether the realistic outcome is $300K, $500K, or $1M.
The decision framework recommends against the accelerator if (a) the company is AI-augmented rather than AI-native, in which case the standard stack path is more appropriate and the accelerator application will not pass screening, (b) the team is unwilling to commit to Bedrock as the primary inference architecture, in which case the accelerator commitment is incompatible with the company's technical direction, or (c) the company is Series-B+, in which case the accelerator targets earlier stages and EDP (committed-spend discounts) is the more relevant AWS commercial structure.
AWS does not publicly identify award sizes per accelerator cohort member. The award sizes are private between AWS and the startup. However, certain accelerator cohort members have been press-mentioned in connection with the $1M tier, and the pattern across these mentions is consistent with the eligibility signals described above.
Across the 2024 and 2025 Generative AI Accelerator cohorts, the publicly-mentioned $1M-tier acceptances have followed a recognizable pattern: Series-A AI-native companies with named tier-1 institutional backing (a16z, Sequoia, Founders Fund typical), product categories where the LLM is the core product rather than a feature (AI-native developer tools, AI-native design tools, AI-native infrastructure), and demonstrated scale at the time of accelerator selection (tens of thousands of users, $50K+ MRR, or comparable commercial validation).
The product categories that have most consistently corresponded to $1M-tier acceptance: AI-native code generation and developer productivity tools (Cursor and the broader category of LLM-powered IDEs); AI-native design and creative tools (Lovable and the broader category of generative design products); AI-native infrastructure for inference, evaluation, or orchestration (the picks-and-shovels layer of the AI stack); AI-native vertical applications in regulated industries where the foundation-model capability unlocks a previously-unsolvable problem (healthcare, legal, financial services where the LLM enables novel automation).
The pattern across these categories: the product would not exist without foundation-model capabilities, the company has reached commercial scale that AWS reviewers can identify as platform-defining, and the projected Bedrock consumption maps to a long-term substantial inference customer relationship. The $1M tier is, in practice, AWS's mechanism for anchoring the recognizable AI-native companies of the cohort year to the Bedrock platform.
The companies that the broader press has associated with cohort highlight status — without confirmation of specific award tier — typically share these characteristics. The names that appear in cohort coverage as headline acceptances are the candidates most likely to have received the $1M tier; the names that appear in cohort coverage without headline emphasis are the candidates most likely to have received the median.
The structural takeaway: the $1M tier is not allocated randomly within the cohort. It corresponds to the specific subset of accepted startups that AWS treats as cohort-defining and platform-anchoring. For an applicant evaluating realistic tier expectations, the relevant question is whether the company would be a cohort-defining acceptance — substantially recognized at the time of application as a notable AI company, with named backing, demonstrated scale, and platform commitment. Companies that would not be cohort-defining will not receive the $1M tier regardless of application quality; companies that would be cohort-defining will receive it as the default outcome of acceptance.
What separates the ceiling from the floor within the population of accepted cohort members.
| Tier | Recipient profile | % of accepted | Approximate award | Tranche structure |
|---|---|---|---|---|
| Floor | Pre-seed AI-native; credible plan; modest projected Bedrock consumption | ~15% | $200K | 2 tranches over 6 months |
| Lower-median | Seed-stage AI-native; named seed VC; early traction signals | ~30% | $250K | 3 tranches over 6 months |
| Median | Seed to early Series-A AI-native; demonstrated Bedrock POC; commercial outcome visible | ~40% | $300K | 3 tranches over 6 months |
| Upper-median | Series-A AI-native; strong VCs; user scale at the tens of thousands or LOI commitments | ~10% | $500K | 3–4 tranches over 6–9 months |
| $1M ceiling | Series-A AI-native; named tier-1 institutional VC (a16z, Sequoia, Founders Fund); cohort-defining scale or commercial signals; $1M+ projected Bedrock spend over 24 months | ~5% | $1M | 3–4 tranches over 6–12 months at $250K–$333K per tranche |
Situation: Series-A AI infrastructure startup providing inference orchestration and evaluation tooling for production AI applications. Raised $40M Series-A led by a tier-1 institutional VC with notable co-investors. Production traffic at the time of accelerator application: roughly $80K/month combined inference and supporting infrastructure on Bedrock, with the projection model showing $200K+/month within 12 months and continuing scale through month 24. The company was already a recognized name in the AI infrastructure category with substantial commercial pipeline.
What CloudRoute did: Standard stack engagement filed in week 1 via AWS partner — Activate Portfolio ($100K) + Build for Startups ($25K) + Bedrock POC ($50K, the ceiling for the standard pool) — landing $175K within 14 days. Generative AI Accelerator application submitted directly to AWS in week 1, with partner advisory on the Bedrock migration plan, the projected consumption model, and the architectural depth sections. Application progressed through initial screening within 30 days, interview phase at day 52, and cohort acceptance at the $1M tier at day 71.
Outcome: Standard stack credits $175K applied by day 14. Accelerator first tranche $250K issued at acceptance (day 71). Bedrock POC in production at day 47 of the accelerator clock (well ahead of the 60-day milestone), unlocking second tranche $250K at day 131. Commercial outcome demonstrated at day 108 (within the 120-day window), unlocking third tranche $250K at day 191. Case study published and continued scale demonstrated at day 232, unlocking final tranche $250K at day 254. Total credit position at completion: $175K stack + $1M accelerator = $1.175M across the 8-month execution window. Founder time invested across both paths: approximately 22 hours.
engagement window: 8 months · founder time: ~22 hours · credits secured: $1.175M (stack + $1M accelerator)
CloudRoute routes AI-native startups to AWS partners filing the standard $150K stack via ACE (lands in 11–18 days) and advises on Generative AI Accelerator application scoping. The two paths run on independent timelines; the realistic combined ceiling is $450K at the median tier and $1.15M at the $1M-tier outlier outcome.