On this page
AWS: EDP, PPA, and the commitment stack#
First, AWS's economics depend on commitment capture. Every dollar of committed spend on EDP is deeply valuable to AWS internally because it locks long-term revenue and reduces churn modeling risk. That economic reality gives strategic-account teams substantial discount authority to close commitment volume, especially against credible competitive alternatives. The rep will not advertise this, deal desk and the strategic-account team will approve it when the buyer demonstrates credible walk-away posture.
Second, AWS maintains a parallel commercial structure beyond EDP: the Private Pricing Agreement (PPA). PPAs layer custom per-service discounts on top of EDP, typically on EC2, S3, data transfer, and RDS. PPAs are not advertised, are not in the standard AWS pricing literature, and are routinely denied at the AE level. They exist, they are approved frequently for strategic accounts, and they can add another 10 to 25 percentage points of discount to specific services that represent the bulk of enterprise cloud spend. For additional benchmark context, compare with our Microsoft Azure pricing page.
Third, Reserved Instances, Compute Savings Plans, and Spot Instances stack on top of EDP and PPA. A well-structured AWS commitment uses EDP for platform-level discount, PPA for service-level discount, Compute Savings Plans for steady-state compute discount (up to 66% off on-demand), Reserved Instances for predictable database workloads, and Spot for fault-tolerant batch. Buyers who only negotiate EDP and ignore the commitment-layer optimization routinely leave 20 to 35% of effective savings unrealized.
EDP discount scales with commitment length and size. A 1-year $1M EDP earns 5 to 10%; a 3-year $5M earns 13 to 18%; a 5-year $25M+ with competitive pressure reaches 22 to 32%. Critically, negotiate the Monthly Aggregate Commitment Cover (MACC) down from AWS's default 100% to 85 to 90% with quarterly flexibility. A 100% MACC means you pay the committed spend even if actual usage falls, a trap when product launches slip or workloads migrate. On commitments above $10M, AWS will accept 85% MACC with quarterly true-up on strategic deals.
PPAs are AWS's strategic retention tool. They layer custom per-service discounts on EC2, S3, data transfer, RDS, and occasionally DynamoDB or EKS. PPAs are not offered by default, they must be requested, usually require escalation to the strategic-account team, and require a competitive alternative to unlock. Well-negotiated PPAs add 10 to 25 percentage points of discount to the services that represent 60 to 80% of enterprise AWS spend.
On top of EDP/PPA, layer commitment discounts. Compute Savings Plans (up to 66% off on-demand) for steady-state compute, with 1-year or 3-year no-upfront terms for maximum flexibility. Reserved Instances for predictable database workloads (RDS reserved). Spot Instances for fault-tolerant batch and ML training (70 to 90% off on-demand). Avoid standard RI lock-in on anything that might migrate regions or instance families, Compute Savings Plans offer equivalent discount with portability.
AWS egress fees are one of the largest hidden costs in cloud. Standard egress runs $0.08 to $0.12 per GB in most regions, which dwarfs compute costs for data-heavy workloads. On EDP contracts above $5M annual spend, negotiate egress caps, tiered egress pricing, or dedicated direct-connect rates. Strategic accounts routinely secure 40 to 70% egress reductions through PPAs, particularly for CloudFront and S3 transfer-out.
AWS bundles “soft costs” into EDP deals: Professional Services credits ($50K to $500K), Training credits, Enterprise Support at discounted tiers, architecture review engagements, and migration credits. These are soft margin and routinely approved on competitive deals. Treat the total EDP package, licenses + ProServe + support + training, as the benchmark, not just the discount percentage.
Default EDP contracts include effective uplift in year 2 and year 3 through rate-card changes. Negotiate flat committed spend with year-over-year rate-card protection, meaning if AWS raises published prices, your committed-rate discount depth remains constant. Separately, negotiate commitment-shift rights: the ability to reallocate committed spend between services (EC2 ↔ S3 ↔ RDS) at year-end without penalty.
— amazon-web-services-aws-discount-negotiation
The framework looks like this: On-demand list price (the highest tier) → discounted by Reserved Instances, Savings Plans, and the Enterprise Discount Program (EDP), each stacking independently. Enterprises typically combine all three to reach their final blended rate.
- 1-year RIs: Up to 40% off on-demand pricing
- 3-year RIs: Up to 60% off on-demand pricing
- Compute Savings Plans: Cover EC2, Lambda, and Fargate, up to 66% off on-demand
- EC2 Instance Savings Plans: Lock to instance family, up to 72% off on-demand
This is the big negotiation lever. EDP is not published, you only unlock it by reaching a minimum annual spend threshold (typically $500K to $1M). Once there, AWS assigns you an Enterprise Account Manager and makes a custom deal.
EDP discounts typically stack on top of RIs and Savings Plans. Enterprises report EDP discounts of 5 to 15% off list, depending on annual commitment, competitive situation, and account size. The largest enterprises (multi-million-dollar spenders) sometimes negotiate higher discounts, but AWS gets tighter as you push past 20% off their published RI rates.
- Business Support: $100/month minimum or 3% of monthly bill (whichever is higher)
- Enterprise Support: $15,000/month minimum or 10% of monthly bill (whichever is higher)
Most enterprises over $1M annual spend are pushed toward Enterprise Support, adding $150K to $500K+ annually depending on usage.
— amazon-web-services-aws-pricing
Microsoft Azure: MACC, Hybrid Benefit, EA bundle, reservations#
First, Microsoft's strategic goal on Azure is AWS displacement, not margin optimization. Strategic-account teams have explicit authority to concede on MACC pricing when a credible AWS workload-migration case exists. "Credible" means named workloads, scoped SOWs, SI partnership, executive sponsorship, the same bar as AWS reverse-negotiation. Without this credible pressure, Azure reps default to 4 to 8% MACC discount depth. With it, 18 to 28% becomes achievable.
Second, Azure Hybrid Benefit is structurally under-used. Customers with Windows Server and SQL Server Software Assurance can apply those licenses to Azure workloads and save 40 to 55% on compute. This is on top of MACC. Many Azure deployments do not apply AHB correctly, the Azure portal allows you to flag AHB, but it requires active license management and SKU-level attention. Our benchmark data shows that 35 to 50% of eligible Azure workloads are billed without AHB applied, which represents millions of dollars in unnecessary spend on any meaningful Windows or SQL footprint.
Third, Microsoft's Enterprise Agreement structure bundles Azure with M365, Dynamics 365, and Power Platform. This creates both opportunity and risk. Opportunity: multi-product EA deals unlock 5 to 12 additional discount points on each component because field comp rewards portfolio attach. Risk: Microsoft prefers portfolio-level bundling that obscures line-item pricing, which blocks benchmarking and increases switching cost at renewal. Force line-item pricing on every product and benchmark each component independently. For related context, see our Microsoft 365 pricing page.
Fourth, Azure Reserved Instances and Savings Plans stack on MACC. A 3-year Reserved Instance saves 40 to 72% off pay-as-you-go, which then stacks on MACC discount and AHB. Buyers who negotiate only MACC and ignore the RI/SP commitment layer leave 20 to 35% of effective savings unrealized.
AHB is the single most valuable structural lever in Azure. Windows Server and SQL Server with Software Assurance applied to Azure save 40 to 55% on those workloads. Audit your Azure estate at signing and at renewal; ensure every eligible VM has AHB flagged. If you don't have Software Assurance on Windows/SQL, negotiate AHB into the EA as part of the Azure MACC discussion. AHB is under-utilized by default and routinely unlocked simply by asking.
A multi-product EA unlocks 5 to 12 additional discount points on each component. Microsoft field comp strongly rewards portfolio attach. On a $20M combined EA (Azure MACC $8M + M365 E5 $10M + Dynamics $2M), expect 18 to 25% blended discount versus 12 to 18% on each component negotiated separately. Critical caveat: insist on decomposed line-item pricing even in a bundled EA. Never sign a portfolio bundle that obscures per-product economics.
Azure RIs save 40 to 72% off pay-as-you-go for 1 or 3-year commitments and stack on MACC. Savings Plans for Compute save 17 to 65% with cross-region and cross-VM-family flexibility. Use Savings Plans for workloads that might migrate, RIs for stable database and compute commitments. A well-structured Azure commitment layers: MACC (platform-level discount) + RI/SP (service-level commitment discount) + AHB (existing-license leverage) for effective discounts 28 to 40% below pay-as-you-go.
MACC discount scales with term length. A 1-year $1M MACC earns 4 to 8%; a 3-year $5M MACC earns 12 to 18%; a 3-year $25M+ MACC earns 20 to 28%. For 5-year commitments with AWS pressure, 30 to 38% is achievable. Critically: negotiate a mid-term reduction right (e.g., 15% reduction at year-2 without penalty) to preserve optionality if workloads migrate or projects slip. Microsoft defaults to fixed commitment but will accept reduction rights on strategic accounts.
Default MACC contracts charge true-up at list price for any consumption above committed spend. Negotiate true-up at committed rate (MACC discount applied to overage) for any spend within 15 to 20% of commitment. Separately, negotiate service-level credit rights for downtime and performance issues that can be applied as credit against future MACC consumption. Both are regularly approved on strategic deals.
Microsoft bundles soft costs into strategic EAs: Azure Fast Track engagements, Cloud Adoption Framework workshops, FastTrack for Microsoft 365, architecture reviews, and migration credits. On a $10M+ EA, expect $200K to $1M of soft consulting value when negotiated. These are budget-funded internally and routinely approved on competitive deals.
— microsoft-azure-discount-negotiation
EA pricing is fundamentally different from pay-as-you-go. You commit to an Azure Monetary Commitment (AMC), typically a 3-year agreement, at a negotiated discount, then consume against that commitment. Unused commitment doesn't roll over; you lose it at year-end.
An Enterprise Agreement works like this: Microsoft says, "Commit $5M annually for three years ($15M total), and we'll give you 15 to 20% off list pricing for all Azure services." You sign, and for the next three years:
Microsoft is migrating customers from EA to MCA (Customer Agreement). MCA terms are often similar to EA, but MCA is more flexible and allows quarterly true-up (adjust commitment mid-year). EA is older but sometimes locks in better rates if grandfathered. In negotiations, you may have a choice between EA and MCA-E (MCA with Enterprise terms). Choose MCA-E if available, the flexibility is worth it.
- 1-year reservations: Up to 34% off pay-as-you-go
- 3-year reservations: Up to 60% off pay-as-you-go
Reservations are separate from EA discounts, they stack. So you can commit to an EA with 15% off, then buy 3-year RIs for another 30 to 40% off on top, for blended discounts exceeding 45% on certain compute workloads.
Azure also offers Savings Plans (similar to AWS). Compute Savings Plans provide up to 65% off pay-as-you-go for a 1-year or 3-year commitment, without locking to specific instance types. These compete with RIs and stack with EA discounts.
This is where Azure has a structural advantage over AWS. If you already own Windows Server or SQL Server licenses, you can bring them to Azure and get massive discounts:
- Windows Server VMs: 40% off Azure compute (you bring the license)
- SQL Server licenses: 85% off Azure SQL Database or SQL MI (if you bring the license)
This is huge. A company with 500 Windows Server licenses on-premises that moves to Azure can cut VM costs by 40% immediately. SQL Server licensing is expensive; moving it to Azure with your existing licenses saves 85%.
Key point: To claim Hybrid Benefit, you must already own the licenses (or be part of a Software Assurance program). Microsoft aggressively audits this.
Microsoft's sales model incentivizes bundling Azure with M365 (Microsoft 365) and Dynamics 365. A single contract covering Azure + M365 + Dynamics often unlocks additional credits or stepped discounts that wouldn't exist in standalone negotiations.
Google Cloud: custom contract, CUDs, sustained use, BigQuery, AI#
First, GCP strategic-account teams have written authority to compete aggressively on headline discount to win AWS or Azure migration workloads. We have seen 3-year GCP custom contracts close at 30%+ discount when a credible AWS alternative was scoped. That discount depth is genuinely hard to achieve with AWS or Azure at comparable commitment volumes.
Second, GCP's Committed Use Discount (CUD) structure is more flexible than AWS Reserved Instances or Azure RIs in one important way: spend-based CUDs apply across VM families without family-specific lock-in. For workloads that evolve over a 3-year horizon, spend-based CUDs avoid the "stranded RI" problem that plagues long-term AWS and Azure commitments. That flexibility is a structural advantage worth real money for customers with changing workload profiles.
Third, GCP's BigQuery Editions (flat-rate slots) offer 30 to 60% savings versus on-demand query pricing for steady analytics workloads. BigQuery is genuinely differentiated as an enterprise analytics warehouse, and Google knows it, which means aggressive BigQuery pricing is available when you're displacing Redshift or Synapse. For buyers running more than ~500 TB of query volume monthly, BigQuery Editions can reshape the entire cloud economics comparison.
Fourth, Vertex AI and the Google AI ecosystem (Gemini API, custom models, TPU access) increasingly feature in strategic cloud negotiations. Google offers aggressive AI commitment discounts and TPU access that AWS and Azure cannot fully match, and these become powerful leverage when AI is in the customer's roadmap. For additional context, see our Vertex AI pricing page.
Spend-based CUDs (applied across VM families) are generally better than resource-based CUDs (tied to specific VM families) for workloads that might evolve. Discount depth is slightly lower, 20 to 46% on spend-based versus 37 to 57% on resource-based, but the flexibility to shift workloads between VM families without losing commitment discount is worth the tradeoff for most enterprise deployments. Use resource-based CUDs only for workloads that are truly stable across the commitment term.
For buyers running meaningful analytics volume, BigQuery Editions (flat-rate slot commitments) deliver 30 to 60% savings versus on-demand query pricing. The negotiation lever: slot-count commitment scales with discount. A 3-year 500-slot BigQuery Editions commitment routinely closes at 40 to 55% below the equivalent on-demand query spend. Model both structures against actual query patterns, if queries are bursty and low-volume, on-demand remains cheaper. If queries are steady and high-volume, flat-rate dominates.
GCP offers three stacking discount mechanisms. Custom-contract (negotiated platform-level discount), Committed Use Discount (commitment-based), and Sustained Use Discount (automatically applied for workloads running >25% of the month). A well-structured GCP commitment layers all three for effective discount depth of 35 to 55% below pay-as-you-go on steady compute workloads. Most buyers negotiate only custom-contract and ignore the commitment-layer optimization.
Google offers aggressive commitment discounts on Vertex AI, Gemini API, and TPU access, areas where AWS Bedrock and Azure OpenAI cannot fully match the product depth. On a $5M+ GCP commitment with a credible AI workload plan, expect 25 to 40% discount on Vertex AI and Gemini consumption, with TPU access at favorable rates for model training. These discounts are under-utilized because most buyers negotiate compute CUDs but ignore the AI commitment layer.
GCP egress, like AWS and Azure, runs $0.08 to $0.12 per GB on standard rates, which dominates total cloud spend for data-heavy workloads. On custom contracts above $5M annual spend, negotiate tiered egress pricing, dedicated Interconnect rates, and Cloud CDN pricing. Strategic accounts routinely secure 40 to 70% effective egress reductions. Google is generally more flexible on egress negotiation than AWS because egress pricing is a known competitive weakness they want to neutralize.
Google bundles soft costs into strategic custom contracts: Professional Services credits ($100K to $1M), AI specialist engagement, architecture review, migration credits, and Anthos or GKE deployment support. Google's PSO and specialist organizations are strong sources of technical value, and they are budget-funded internally for strategic accounts. On a $10M+ GCP commitment, expect $300K to $1.5M of soft consulting value when negotiated.
— google-cloud-platform-gcp-discount-negotiation
- Pay-as-you-go: Baseline pricing for all resources (compute, storage, networking, services). This is what the pricing calculator shows, but almost no enterprise pays this.
- Sustained Use Discounts (SUDs): Automatic discounts applied monthly for virtual machines and SQL databases running 25% or more of the month. You get these automatically, no commitment required, up to 30% off for some VM types.
- Committed Use Discounts (CUDs): You commit to 1-year or 3-year terms and receive 25 to 37% off (1-year) or 30 to 55% off (3-year) for compute, memory, and specific accelerators.
- Private Agreements: For enterprises spending $1M+ annually, Google's account teams negotiate custom pricing via formal agreements. This is where the largest discounts occur, often 40 to 50% off or better.
If you run a VM for 25% of the month, you get an automatic SUD. The reduction increases progressively:
- 25% uptime: 10% discount
- 50% uptime: 20% discount
- 75% uptime: 30% discount
- 1-year CUD: 25 to 37% off list price (varies by resource type). Lower risk; easy to exit.
- 3-year CUD: 30 to 55% off list price. Significant commitment, but if your workload is stable, the savings are meaningful.
At $1M+ annual spend, Google treats you as a strategic account. Private Agreements allow:
- Custom per-resource pricing (not just tier-based discounts)
- Volume-based multipliers (e.g., "all compute in region X gets 48% off")
- Bundled pricing with Google Workspace Enterprise
- Google Cloud credits for specific products (Vertex AI, BigQuery, Looker)
- Multi-year pricing locks with guaranteed price ceilings
- On-demand: $6.25 to $6.90 per TB of data scanned (varies by region). You pay per query.
- Flat-rate slots: $2,000/month for 100 slot-hours, up to $40,000+/month for larger reservations. Fixed cost, unlimited queries.
— google-cloud-platform-gcp-pricing
Databricks: DBUs, capacity commitments, workload SKUs#
First, Databricks is optimizing for IPO readiness. Public bookings growth narrative and logo-count metrics matter more than per-deal margin. Deal desk has standing authority to offer aggressive discounts on strategic accounts, particularly competitive displacements from Snowflake and Google BigQuery. Displacement deals routinely clear 45 to 55% off list, substantially deeper than renewals of existing Databricks customers.
Second, the Snowflake vs. Databricks competitive narrative is existential for both vendors. Databricks positions as the lakehouse that subsumes the warehouse (Snowflake's core product). Snowflake positions as the data cloud that subsumes the lakehouse (Databricks' core product). Deal desks on both sides have elevated authority for competitive displacements. A written Snowflake proposal is the single most effective leverage in a Databricks negotiation, and vice versa.
Third, workload expansion is the primary growth vector. Databricks started as an ML/data engineering platform and expanded into SQL (Databricks SQL, formerly SQL Analytics), Workflows, Model Serving, Vector Search, Unity Catalog, and Delta Live Tables. Each workload is a separate DBU SKU with its own per-DBU rate. Customers who commit only to ML workloads miss the discount capacity available on multi-workload bundled commitments, typically 10 to 20 points deeper.
Fourth, serverless pricing is a hidden cost risk. Databricks Serverless SQL and Serverless Model Serving carry premium per-DBU rates (1.5 to 2x non-serverless equivalent) and are typically priced at list in default contracts. Customers who shift workloads to serverless post-signing, often without explicit negotiation, see 40 to 80% effective cost increases. Negotiate serverless commitment pricing at signing with published rates for serverless consumption.
Fifth, the data platform landscape is now genuinely competitive. Snowflake has matured its ML capabilities with Snowpark, closing the gap with Databricks. Microsoft Fabric is a credible bundle in Microsoft 365 E5 environments. Google BigQuery plus Vertex AI is a credible alternative for ML-heavy workloads on GCP. AWS SageMaker plus Redshift is a credible alternative for AWS-native customers. A written RFP with proposals from two of these alternatives unlocks discount capacity that verbal pressure does not.
Written competitive proposals from Snowflake or Microsoft Fabric are the single largest lever. Databricks reps treat competitive threats as bluff until a written alternative surfaces. Once it does, they model line by line and price 5 to 10 points below the next-best alternative on strategic accounts. For Snowflake displacements in particular, Databricks has elevated deal-desk authority, the competitive narrative is existential for both vendors.
Databricks bundles SQL, ML, Workflows, and Model Serving at 10 to 20 points of additional discount vs. single-workload commitments. But the bundle discount is only valuable with cross-workload DBU fungibility, committed DBUs in one workload SKU should offset overage in another. Without fungibility, a customer over-consuming on SQL while under-consuming on ML pays list overage on SQL despite having unused ML commitment. Negotiate a unified DBU commitment pool that flows across all workloads.
Serverless SQL and Serverless Model Serving carry 1.5 to 2x the per-DBU rate of non-serverless. Default contracts price serverless at list with no commitment protection. Negotiate serverless DBU commitments at signing with published rates, rollover, and cross-workload fungibility. For customers planning serverless adoption during term, this clause is worth 20 to 40% of 3-year effective cost.
Workload expansion during term, new users, new pipelines, new ML models, new dashboards, is typically priced at list for overage above committed DBUs. Negotiate expansion at the same discount tier as base commitment, with published per-DBU rates in the order form, and automatic re-tiering into higher commitment bands at parity. For customers with active ML programs, this clause alone is worth 20 to 35% of 3-year effective cost.
Databricks' standard renewal uplift is 5 to 8% on per-DBU pricing. Cap at lower of US CPI or 3%, applied to effective per-DBU rates across all workload SKUs. Negotiate multi-year per-DBU price protection, rates locked for the full term regardless of workload mix changes during the period. Without price protection, Databricks can re-price mid-term under workload-mix language.
Unity Catalog, Delta Sharing, Delta Live Tables, and governance tooling are priced as add-ons to base DBU commitments. Databricks wants broad Unity Catalog adoption as the competitive moat vs. Snowflake's Horizon Catalog. Bundled with DBU commitment at signing, these tools discount 35 to 55%, often deeper than the DBU base. Post-signing, they discount 15 to 25%. Commit at signing with phased deployment milestones and deactivation rights if adoption stalls.
— databricks-discount-negotiation
Databricks pricing is consumption-based, structured around Databricks Units (DBUs), a standardized measure of processing capability per hour. The actual cost per DBU varies based on three dimensions: the workload type (All-Purpose Compute, Jobs Compute, SQL Warehouse, Delta Live Tables, Model Serving), the product edition (Standard, Premium, Enterprise), and the cloud provider and region where the workload runs. This three-dimensional pricing matrix is one of the most complex in enterprise software, and organizations that do not actively govern their Databricks usage routinely pay 2 to 3x more than necessary for equivalent work.
Unlike most enterprise software vendors, Databricks has published pricing for most workload types on its website, though enterprise negotiated pricing bears little resemblance to the published rates. The published rates serve as the on-demand anchor from which capacity commitment discounts are calculated. Understanding the published rate for each workload type is the starting point for Databricks negotiation, not the endpoint. The full data and analytics pricing benchmark covers Databricks, Snowflake, BigQuery, and the broader modern data stack for complete market context.
Databricks capacity commitments, pre-purchased DBU packages negotiated directly with Databricks enterprise sales, provide the primary pricing mechanism for enterprise customers. Capacity commitments specify an annual dollar commitment, a per-DBU rate for each workload type within scope, and the cloud provider and region. Organizations with meaningful Databricks workloads (above $300,000 annually) should have a capacity commitment agreement in place, paying on-demand rates on a workload of this scale leaves 30 to 55% of cost reduction on the table. The negotiation is straightforward: the larger the annual commitment, the lower the per-DBU rate for each workload type.
Typical Contract Length1 to 3 years; capacity commitments paid annuallyDiscount vs On-Demand30 to 55% for enterprise capacity commitmentsFiscal Year EndJanuary 31, strong Q4 discount authority Nov to Jan Cloud Providers
Data engineering-primary deployments, ELT pipelines, batch processing, data lake management using Delta Lake, streaming ingestion, at mid-market enterprise scale ($300K to $1M annual run rate) represent the most common entry point. Jobs Compute is the dominant workload type, consuming DBUs at the lowest per-DBU rate. At negotiated capacity commitment pricing, Jobs Compute runs $0.15 to $0.22 per DBU for Standard/Premium tier. These deployments are highly price-competitive versus alternatives like AWS EMR or Azure HDInsight, particularly at scale where Databricks' Delta Lake performance improvements reduce total compute hours required.
Analytics and SQL-intensive deployments, where data science teams use notebooks, BI tools connect via SQL Warehouse, and business analysts run ad-hoc SQL queries, are the fastest-growing Databricks use case. SQL Warehouse DBU pricing at enterprise discount runs $0.22 to $0.35 per DBU. This is where Databricks competes most directly with Snowflake. At enterprise scale (50,000+ DBUs per month in SQL Warehouse), Databricks is typically 20 to 35% less expensive than equivalent Snowflake capacity commitments for the same query volumes. The Snowflake vs Databricks SQL comparison is the most consequential data platform cost decision most enterprises make in 2026.
AI and ML platform deployments, including model training on GPU clusters, Mosaic AI model serving (Model Serving endpoints), Vector Search for RAG applications, and Feature Engineering pipelines, represent Databricks' highest-growth and highest-per-DBU workload tier. GPU cluster DBU rates are 4x to 8x standard compute rates. Organizations scaling AI workloads on Databricks without specific AI-tier contract pricing discover that model training and inference costs represent 40 to 60% of total Databricks spend despite being a small fraction of workload volume. Negotiate AI workload DBU rates explicitly as a separate line item in capacity commitment agreements.
Snowflake is the single most powerful lever for extracting Databricks discounts. In our benchmark database, organizations that conducted a formal Databricks vs Snowflake evaluation, presenting both vendors with workload requirements and receiving competitive proposals, achieved Databricks discounts that averaged 12% deeper than organizations that negotiated Databricks standalone. Databricks has specific "competitive displacement" pricing tiers for situations where Snowflake is the incumbent or a formal alternative. These tiers require documentation (a signed NDA with Snowflake for evaluation, Snowflake pricing proposals), but the discount authorization they unlock is genuine and substantial.
The jobs-to-all-purpose workload migration is a commercial lever that works both for cost optimization and negotiation. Organizations that commit in their capacity agreement to migrate production pipelines from All-Purpose Compute to Jobs Compute, reducing their per-DBU cost by 40 to 50% for those workloads, can negotiate a higher total DBU commitment at the lower Jobs rate, creating a larger commitment that unlocks deeper platform-level pricing. This restructuring of workload allocation is operationally beneficial (production pipelines should not run on All-Purpose Compute anyway) and commercially beneficial, reducing the effective per-unit cost for the entire platform agreement.