Selecting a Data Warehouse Provider: A Daunting Task

ao Ferreira

, Nuno Lourenc¸o

and Jo

ao R. Campos

CISUC/LASI, DEI, University of Coimbra, Coimbra, Portugal

Keywords:

Data Warehouses, ETL Tools, Cost Comparison, Pricing Models.

Abstract:

In the contemporary landscape of rapid data accumulation, organizations increasingly rely on data warehouses

to process and store vast datasets efﬁciently. Although the most challenging task is appropriately designing

a data warehouse, selecting a provider is far from the trivial task it should be. Each provider offers a distinct

array of services, each with its pricing model, which requires signiﬁcant effort to analyze and determine which

conﬁguration meets the speciﬁc needs of the organization. In this paper, we highlight the inherent challenges

of making fair comparisons among data warehouse solutions, providing the context of a start-up in the space

trafﬁc management industry as a case study. We deﬁned several critical attributes for corporate decision-

making: cost, processing capabilities, and data storage capacity. We systematically compare four leading

technologies: Google BigQuery, AWS Redshift, Azure Synapse, and Snowﬂake. Our methodology employs a

set of metrics designed to assess warehouse solutions, encompassing storage pricing, processing capabilities,

scalability, and the integration of ETL tools. The process and the results highlight the challenges of this

evaluation. It underscores the need for a standard approach to characterize the provided service speciﬁcations

and pricing to allow for a fair and systematic assessment and comparison of alternative solutions.

1 INTRODUCTION

The exponential escalation in the volume of data gen-

erated globally reﬂects an era marked by the contin-

uous increase in data storage capacity, with the ma-

jority being hosted in cloud environments. Techno-

logical advances and the increasing digitization of in-

formation in all sectors of society drive this growth

(Aftab and Siddiqui, 2018). As a result, there has

been a signiﬁcant increase in the use of Data Ware-

house (DW) solutions, which are essential to handling

and extracting knowledge and value from these data

(Berisha et al., 2022).

Data warehouses extend beyond the traditional

concept of centralized data repositories, functioning

as dynamic and sophisticated structures. Designed to

store and organize large volumes of data from vari-

ous sources, such as transactional systems, relational

databases, or heterogeneous sources, DWs are fun-

damental in various areas, including business, health,

science, and technology (Serra, 2024).

The explosion in the number of companies that

build data architectures has been remarkable in recent

https://orcid.org/0000-0003-4242-7961

https://orcid.org/0000-0002-2154-0642

https://orcid.org/0000-0002-4623-764X

years, and this trend is expected to continue. Daily

data generation is estimated at around 44 zettabytes

(Berisha et al., 2022), driven by sources like social

media, IoT devices, and third-party software (Kim

et al., 2017). From 2018 to 2021, the data vol-

ume nearly doubled to approximately 84 zettabytes,

with projections suggesting that it could reach 149

zettabytes by 2024 (Ahlawat et al., 2023).

Companies can enhance revenue by analyzing

data to detect trends and make forecasts (Serra, 2024).

However, designing a DW that effectively supports

business processes and analytics remains challeng-

ing (Santos and Costa, 2022). Selecting a suit-

able provider requires evaluating processing capac-

ity, hardware, scalability, and cost. This process is

often complex and time-consuming due to heteroge-

neous architectures and pricing models, which hin-

der objective comparisons (Soma, 2022). The lack

of standardization further complicates the task, espe-

cially for startups with limited resources and time.

This paper adopts the perspective of a startup se-

lecting its initial cloud-based DW provider, where

the investment decision must balance current needs

with future scalability and operational costs. Rather

than offering a comprehensive assessment of platform

functionalities, the focus is on evaluating costs and

552

Ferreira, J., Lourenço, N., Campos and J. R.

Selecting a Data Warehouse Provider: A Daunting Task.

DOI: 10.5220/0013568700003967

In Proceedings of the 14th International Conference on Data Science, Technology and Applications (DATA 2025), pages 552-559

ISBN: 978-989-758-758-0; ISSN: 2184-285X

conﬁguration aspects that impact the initial setup and

expected evolution of the data infrastructure.

Recent studies have expanded knowledge of data

warehouses and cloud services, highlighting the value

of comparative analyses. Uddin (Uddin and Hos-

san, 2024) examined AI integration in DWs for big

data optimization, summarizing topics from 25 pa-

pers. Villamizar (Villamizar et al., 2017) analyzed

the economics of microservices and AWS Lambda,

noting cost-efﬁciency despite management complex-

ity. AlJamal (Aljamal et al., 2019) and Biplob (Badi-

uzzaman Biplob et al., 2018) explored IaaS and HPC,

focusing on virtual machines and real-time ETL tools.

Harby (Harby and Zulkernine, 2022) introduced the

Data Lakehouse model to handle both structured and

unstructured data. However, we did not ﬁnd studies

in the literature that address the speciﬁc challenges of

selecting a suitable DW provider from a startup per-

spective.

In this paper, we systematically compare four

DW solutions: Google Big Query (Google, 2025),

AWS Redshift (Services, 2025), Azure Synapse (Mi-

crosoft, 2025), and Snowﬂake (Inc., 2025), highlight-

ing the complexities of cost comparison. Our analysis

revealed signiﬁcant challenges stemming from non-

standardized virtual machine hardware, diverse ser-

vices, and varying pricing policies among providers.

This lack of uniformity complicates objective evalu-

ations of DW solutions, as performance and capacity

differ signiﬁcantly across similar categories. With the

challenges highlighted in this work, we aim to con-

tribute towards the standardization of DW solution

providers.

2 DW PRICE MODEL

Data warehouse service providers typically adopt

multifactor pricing models that are designed to reﬂect

resource usage, scalability, and performance. Here

are some of the most common pricing models adopted

by these providers (sourced from the ofﬁcial DW

websites):

• Instance or Node-Based Pricing: This model de-

pends on the number and compute nodes or in-

stances, suitable for platforms like Amazon Red-

shift and Azure Synapse Analytics, allowing users

to choose conﬁgurations based on their workload

needs.

• Resource-Based Pricing: Charges are based on

actual resource usage such as CPU and memory.

This model is often used by services that offer au-

tomatic scalability, like Google BigQuery, which

also considers the volume of data processed and

storage.

• Storage-Based Pricing: Costs are calculated

based on the amount of data stored in the data

warehouse, typical for platforms that separate

compute and storage resources, such as Snowﬂake

and Google BigQuery.

• Performance or Speed-Based Pricing: Pricing

is determined by the speed of query processing

and the performance level selected, allowing cus-

tomers to pay more for higher speeds and lower

latency.

• Subscription: Involves a ﬁxed monthly or an-

nual fee, providing access to a predeﬁned package

of features and capabilities, with discounts often

available for longer-term commitments.

Although these various pricing approaches might

provide more ﬂexible and granular billing because

there are no standards or uniﬁed practices, compar-

ing the pricing of multiple providers becomes time-

consuming and difﬁcult.

3 METHODOLOGY

This study emerged from the development of a DW

solution for a company in the space trafﬁc manage-

ment sector. A systematic methodology was adopted

to compare existing DW platforms, aligning the anal-

ysis with the company’s requirements and the capa-

bilities of major providers.

Requirements and Contextual Analysis: The

initial evaluation focused on four leading providers

with infrastructure in Europe, considering scalabil-

ity, ease of deployment, and potential for future ex-

pansion. Key criteria included storage, processing,

backup options, integration with ETL tools, and cost

models, favoring on-demand pricing and ﬁxed-rate

plans for predictable budgeting. The European loca-

tion was selected due to the company’s base of oper-

ations, alongside regional advantages such as strong

data protection laws, robust infrastructure, and prox-

imity to end users.

Chosen Providers: Amazon Redshift, Google

BigQuery, Microsoft Azure Synapse, and Snowﬂake

were selected for their market leadership, global

reach, and advanced technical capabilities. All of-

fer scalable solutions, integrated services (including

ETL tools), and robust infrastructure in Europe, align-

ing with the company’s geographic and operational

needs.

Deﬁning the Evaluation Process: We selected

baseline hardware conﬁgurations aligned across

Selecting a Data Warehouse Provider: A Daunting Task

553

providers in terms of vCPU, memory, and storage.

The analysis compared technical features and costs,

including capacity-based and serverless pricing, long-

term contract discounts (1–3 years), and regional data

transfer costs. We also evaluated ETL-related ex-

penses, such as pipeline orchestration and ﬂow exe-

cution.

Data Adequacy Check: We then veriﬁed whether

the collected data supported a fair and reliable com-

parison. If gaps were identiﬁed, a Collect More

Data step was triggered to obtain further details from

providers, ensuring the dataset was robust enough to

support informed and accurate conclusions.

For Data Analysis, given the recognized dispar-

ity in the units of measurement, scales, and pricing

models provided by the different providers, it was es-

sential to carry out data adjustments and transforma-

tions. Among others, this process involved standard-

izing some units of measurement.

4 PLATFORMS

This section examines DW solutions from leading

providers: Google BigQuery (Google, 2025), AWS

Redshift (Services, 2025), Azure Synapse (Microsoft,

2025), and Snowﬂake (Inc., 2025), selected for their

compatibility with the partner company’s require-

ments. We analyze pricing and hardware speciﬁ-

cations to highlight each platform’s capabilities and

cost structure. All prices were obtained from ofﬁcial

provider websites during the ﬁrst half of 2025.

4.1 Google BigQuery

Google BigQuery is a serverless data warehouse that

uses standard SQL, easing adoption for users of tradi-

tional databases. It integrates with Google’s analytics

and machine learning tools and operates on a scal-

able architecture. The service provides various com-

puting nodes tailored to speciﬁc processing needs, as

outlined in Table 1.

Table 1: BigQuery Data Warehouse Node Conﬁgurations.

Name vCPU GB per CPU

E2 up to 32 8

N2 up to 128 8

N2D up to 224 8

C3 up to 176 2 or 4 or 8

C3D up to 360 2 or 4 or 8

Table 2 shows the pricing structure of various Big-

Query editions, quantiﬁed by the use of slots, which

represent fundamental units of computational power

similar to virtual CPUs. Queries are broken down into

tasks and processed in parallel across multiple slots,

with the number of slots enhancing processing speed

by enabling more simultaneous task execution.

Table 2: Pricing and Commitment Models.

On Demand - Access 2000 slots Simultaneous

Standard Edition

Model Hourly cost Details

Pay as you go $0.044 / slot hour $/per sec

Enterprise Edition

Pay as you go $0.066 / slot hour $/per sec

1 yr commit $0.0528 / slot hour $/for 1 yr

3 yr commit $0.0396 / slot hour $/for 3 yrs

Enterprise Plus Edition

Pay as you go $0.11 / slot hour $/per sec

1 yr commit $0.088 / slot hour $/for 1 yr

3 yr commit $0.066 / slot hour $/for 3 yrs

The lack of detailed machine conﬁgurations may

create uncertainty for users needing to align perfor-

mance with budget, which is essential for precise

planning or speciﬁc technical demands. BigQuery’s

one- and three-year commitments offer pricing incen-

tives for stable workloads, enabling signiﬁcant sav-

ings over pay-as-you-go rates.

4.2 Amazon AWS

Amazon Redshift (Services, 2025), part of AWS, of-

fers RA3 instances with managed storage and DC2

Dense Compute instances, as shown in Table 3 for

the Europe (Spain) region. RA3 enables indepen-

dent scaling of compute and storage, while DC2

synchronizes both, favoring compute-intensive work-

loads with steady data volumes. Key attributes in-

clude vCPU, ECU, memory, I/O (GB/s), and hourly

cost.

Table 3: Instance types and speciﬁcations for Redshift.

Type vCPU Mem I/O Price

RA3 with Redshift Managed Storage

ra3.16xlarge 48 384 G 8.00 $14.424

ra3.4xlarge 12 96 G 2.00 $3.606

ra3.xlplus 4 32 G 0.65 $1.202

ra3.large 2 16 G 0.36 $0.601

Dense Compute DC2

dc2.8xlarge 32 244 G 7.50 $5.6

dc2.large 2 15 G 0.60 $0.30

Table 4 provides an analysis of the price and com-

mitment models for Amazon Redshift instances in a

European region (Spain). It categorizes costs into

various commitment terms: one-year commitments

DATA 2025 - 14th International Conference on Data Science, Technology and Applications

554

with no initial payment, partial initial payment, and

full initial payment. Each category describes not only

the initial fees and monthly fees but also the effective

hourly rates, the annual costs per terabyte, and the

comparative savings compared to on-demand rates

(due to space constraints, on-demand rates, and three-

year payment values are omitted in this article but are

indirectly represented in the savings column).

Table 4: RI Pricing and Commitment Models for Redshift.

Instance

RI upfront

Fee

Monthly

Fees

Effective

Hourly Rate

Savings on

demanded

No advance payment - One Year (EU - Spain)

dc2.large USD 0 USD 175.20 USD 0.240 20%

ra3.4xlarge USD 0 USD 1,839.60 USD 2.520 30%

dc2.8xlarge USD 0 USD 3,212.00 USD 4.400 21%

ra3.16xlarge USD 0 USD 7,373.00 USD 10.100 30%

Partial advance - One Year

dc2.large USD 858 USD 73.00 USD 0.198 34%

ra3.4xlarge USD 10,600 USD 883.30 USD 2.420 33%

dc2.8xlarge USD 16,951 USD 1,416.20 USD 3.875 31%

ra3.16xlarge USD 42,311 USD 3,525.90 USD 9.660 33%

Full upfront payment - One Year

dc2.large USD 1,682 USD 0.00 USD 0.192 36%

ra3.4xlarge USD 20,849 USD 0.00 USD 2.380 34%

dc2.8xlarge

USD 33,218 USD 0.00 USD 3.792 32%

ra3.16xlarge USD 83,395 USD 0.00 USD 9.520 34%

Table 5 provides a detailed overview of the costs

associated with data transfers within Amazon Red-

shift, describing a pricing model that differentiates be-

tween inbound and outbound data movements.

Table 5: Data Transfer Costs for Amazon Redshift.

Category Cost

Inbound Free

Outbound - From Redshift to Internet

First 10 TB per month $ 0.09 per GB

Next 40 TB per month $ 0.085 per GB

Next 100 TB per month $ 0.07 per GB

More than 150 TB per month $ 0.05 per GB

Outbound to Other Regions $ 0.02 per GB

Inbound transfers are free, encouraging users to

upload data without incurring charges. In contrast,

outbound transfers are tiered, with costs decreasing as

data volumes start at USD 0.09 per GB for the ﬁrst 10

TB and reducing to USD 0.05 per GB beyond 150 TB

per month; additionally, a reduced rate of USD 0.02

per GB for outbound transfers to other AWS regions.

4.3 SnowFlake

Snowﬂake is a cloud-based DW that decouples com-

pute from storage (Inc., 2025). Table 6 presents spec-

iﬁcations for various instance types, from basic (e.g.,

‘CPU - XS’) to advanced (e.g., ‘GPU - L’), detailing

vCPUs, memory (GiB), storage, GPU type and mem-

ory, and resource usage limits where applicable.

Table 6: Snowﬂake Node Speciﬁcations.

Instance vCPU Memory GPU Type GPU Memory

CPU-XS 2 8 GiB n/a n/a

CPU-S 4 16 GiB n/a n/a

CPU-M 8 32 GiB n/a n/a

CPU-L 32 128 GiB n/a n/a

HighMem-S 8 64 GiB n/a n/a

HighMem-M 32 256 GiB n/a n/a

HighMem-L 128 1024 GiB n/a n/a

GPU-S 8 32 GiB 1 NVIDIA A10G 24 GiB

GPU-M 48 192 GiB 4 NVIDIA A10G 96 GiB

GPU-L 96 1152 GiB 8 NVIDIA A100 320 GiB

The complexity of comparisons increases due

to the differing computational needs between CPU-

centric and GPU-enhanced instances. Snowﬂake’s

pricing includes Credit Costs for Virtual Storage Ser-

vices, where usage is quantiﬁed in ”credits” as de-

tailed in Table 7, and On-Demand Storage Costs

based on data volume, as outlined in Table 8. Storage

costs are segmented by tiers, with prices varying ac-

cording to the amount of data stored. The tiered pric-

ing encourages larger data storage by offering lower

rates for higher volumes.

Table 7: Credits Per Hour for Snowﬂake.

Service XS S M L 2XL 4XL

Standard Warehouse 1 2 4 8 32 128

Snowpark Optimized N/A N/A 6 12 48 192

SnowPark Container Service Compute

CPU 0.11 0.22 0.43 1.65 - -

High-Memory CPU N/A 0.56 2.22 8.88 - -

GPU N/A 1.14 5.36 28.24 - -

Table 8: On Demand - Credit and Standard Storage Pricing

- SnowFlake.

On Demand - Credit Pricing

Provider Region Standard Enterprise Business Critical VPN

AWS Dublin 2.6 3.9 5.2 7.8

AZURE Ireland 2.6 3.9 5.2 7.8

GCP Netherlands 2.6 3.9 5.2 7.8

On Demand - Standard Storage Pricing

Tier 1 Tier 3 Tier 5 Tier 7

AWS Dublin 23 19.94 16.86 13.8

AZURE Ireland 23 19.94 16.86 13.8

GCP Netherlands 20 20 20 20

4.4 Microsoft Azure Synapse Analytics

Azure Synapse Analytics integrates big data and

data warehousing, supporting sources from relational

databases to data lakes. It offers over 125 VM

types across six categories (e.g., General Purpose,

Compute-Optimized, GPU) and 31 OS options on

Windows and Linux, ensuring broad conﬁgurability.

For this analysis, we focus on the Azure D2ads

Selecting a Data Warehouse Provider: A Daunting Task

555

v5 instance (Table 9) because its hardware closely

aligns with the discussed platforms, facilitating direct

comparison and consistent data presentation. Other

instances, which are not detailed here, have similar

descriptions but vary in values and speciﬁcations.

Table 9: Azure Dads v5 Series Pricing.

Instance vCPU/RAM/Storage Pay as you go Savings (% off)

D2ads v5 2 / 8 GiB / 75 GiB $75.19/mo 1-Yr: 31, 3-Yr: 54

D4ads v5 4 / 16 GiB / 150 GiB $150.38/mo 1-Yr: 31, 3-Yr: 54

D8ads v5 8 / 32 GiB / 300 GiB $300.76/mo 1-Yr: 31, 3-Yr: 54

D16ads v5 16 / 64 GiB / 600 GiB $601.52/mo 1-Yr: 31, 3-Yr: 54

D32ads v5 32 / 128 GiB / 1.2 TB $1,203.04/mo 1-Yr: 31, 3-Yr: 54

D48ads v5 48 / 192 GiB / 1.8 TB $1,804.56/mo 1-Yr: 31, 3-Yr: 54

D64ads v5 64 / 256 GiB / 2.4 TB $2,406.08/mo 1-Yr: 31, 3-Yr: 54

D96ads v5 96 / 384 GiB / 3.6 TB $3,609.12/mo 1-Yr: 31, 3-Yr: 54

Table 9 outlines the pricing models for the Azure

Dads v5 series, featuring a tiered pricing structure

based on vCPUs, RAM, and storage capacities across

different payment plans. Pay-as-you-go rates increase

with resource capacity, reﬂecting the direct costs of

computing resources. In contrast, long-term commit-

ment plans, such as the 1-year and 3-year savings

plans, offer cost reductions ranging from 31% to 54%.

5 DISCUSSION

A meticulous comprehension and rigorous evaluation

of the ﬁnancial implications and the Return on In-

vestment (ROI) are imperative prerequisites prior to

the initiation of a Data Warehouse (DW) develop-

ment within a cloud environment. The genesis of

this study was a case analysis that reviewed these as-

pects. The evaluation process necessitates an estima-

tion of operational expenses encompassing storage,

data processing, and information transfer, alongside

an examination of the cost reductions afforded by the

scalability and elasticity inherent to cloud infrastruc-

tures. Of equal signiﬁcance is the analysis of both

tangible and intangible beneﬁts derived from the de-

ployment of a cloud-based DW, such as enhanced re-

sponse times for analytical queries, improved opera-

tional efﬁciency, and bolstered support for data-driven

decision-making.

However, conducting a balanced comparison

across available market solutions poses a substantial

challenge due to the variance in technical conﬁgu-

rations among platforms. These differences encom-

pass distinct processing architectures, data partition-

ing techniques, and query optimization strategies (as

presented in detail in section 4).

To ensure a fair and accurate assessment, we

strove to standardize the conﬁgurations between the

different services, keeping the vCPU, memory, and

storage capacity speciﬁcations as closely aligned as

possible. This methodological approach allows us to

provide a clear view of the variations in cost and per-

formance. Table 10 summarizes the conﬁgurations

and hourly prices of four popular cloud DW plat-

forms. The price column for BigQuery is empty, in-

dicating missing data.

Table 10: Warehouse characteristics and its conﬁgurations.

Warehouse Node Name vCPU Memory Storage Price/h

BigQuery E2 up to 32 8 GB 128 GB -

Redshift dc2.large 2 15 GB - $0.3

SnowFlake CPU X64 S 4 16 GB - $0.22

Azure D4ads v5 4 16 GB 150 GB $0.22

Table 11 shows a comparison of data warehouse

solutions according to cost based on a consultation

carried out in March 2024 for Snowﬂake, BigQuery,

Redshift, and Azure Synapse.

Table 11: Comparative Costs of Data Warehouse Solutions.

Warehouse

Storage

(1TB)/Mo

Compute

Cost/h

Serverless

Cost/TB

1-Yr

Saving

3-Yr

Saving

SnowFlake $23 $2.3 - - -

BigQuery $20 $0.044/slot $6 - -

Redshift $24 $0.3 $5 ∼34% ∼63%

Azure Synapse $20 $0.2216 $4.6 ∼24% ∼47%

5.1 Processing and Storage Costs

In this section, we explore in detail the results ob-

tained in the comparisons, focusing on the variations

in hardware conﬁgurations and the pricing models of

the selected data warehouses.

5.1.1 Storage Cost per TB per Month

BigQuery and Azure Synapse have an identical

monthly cost of $20 per TB per month, which is also

in line with Snowﬂake $23. Redshift is slightly more

expensive, at $24.

5.1.2 Capacity-Based Cost

This model requires customers to choose and pay

for a speciﬁc capacity of resources that are reserved

regardless of usage. Snowﬂake offers hourly-based

costs that are considerably higher ($2.3/h) compared

to other options. BigQuery offer ($0.044 per slot per

hour) and Redshift ($0.3/h). Azure Synapse offers the

best choice ($0.2216/h).

5.1.3 Serverless Computing Costs

This model automates resource scaling and charges

based on actual usage, eliminating server manage-

DATA 2025 - 14th International Conference on Data Science, Technology and Applications

556

ment. BigQuery and Redshift offer serverless com-

puting at competitive rates ($6/TB and $5/TB, re-

spectively), suitable for variable workloads. Azure

Synapse offers the lowest rate at $4.6/TB. Snowﬂake

does not provide a serverless option. This approach

is advantageous for companies with ﬂuctuating data

needs, allowing them to pay only for utilized re-

sources without managing infrastructure.

5.1.4 Discounts for Contracts

Redshift and Azure Synapse offer notable discounts

for one- and three-year commitments, up to 63% and

47%, respectively, making Redshift attractive despite

its higher base price.

Each DW solution presents distinct advantages.

BigQuery’s ﬂexible pricing suits variable workloads.

Snowﬂake and Synapse offer versatile conﬁgurations

and competitive discounts, while Redshift’s long-term

savings beneﬁt organizations are able to commit in

advance.

Table 12 shows the values associated with stor-

ing and backing up data in the same region (Eu-

rope/Europe) and another continent (Europe/North

America).

Table 12: Comparative costs of In/Out storage solutions.

Warehouse

Storage($)

1TB/Mo

Inbound

Out. EU/EU

($/TB)

Out. EU/US

($/TB)

Est.

Cost ($)

SnowFlake 23 Provider 20 / TB 50 / TB C2,000

BigQuery 20 Free 0.2 / GB 0.5 / GB C300

Redshift 24 Free Free 0.2 / GB C200

Azure 20 Free 0.019 / GB 0.047 / GB C320

5.1.5 Variability in In/Out Costs

Inbound: All providers except Snowﬂake offer free

inbound data. Snowﬂake indicates a ’Cloud Provider

Cost,’ implying potential costs from the cloud stor-

age provider, which varies based on the customer’s

choice, making direct comparisons challenging.

Outbound: Outbound rates differ signiﬁcantly in

values and units (per GB vs. per TB) and depend

on the storage region and destination. For instance,

a lower per GB rate may become substantial when

scaled to TB.

5.2 Regional Storage Implications

In this subsection, we will explore the ﬁnancial, le-

gal, and logistical complexities associated with stor-

ing data in Europe with different backup strategies,

whether within the same continent or transcontinen-

tally in the United States.

5.2.1 Europe/Europe

This cost refers to data storage within Europe, in-

cluding backups in other European regions. Price

variations reﬂect differing policies and infrastructures

across providers. Redshift, for example, offers the

ﬁrst 10 TB per month free. For this analysis, ge-

ographically close servers were selected: Snowﬂake

(Dublin, Ireland, Netherlands), AWS (Frankfurt,

Ireland, London, Milan, Paris, Spain, Stockholm,

Zurich), Azure (Paris, Frankfurt, Milan, Ireland,

Madrid, London, Netherlands), and BigQuery (Bel-

gium, Berlin, Frankfurt, London, Madrid, Milan,

Paris).

5.2.2 Europe/America

Table 12 presents the cost of storing data in Europe

with backups in the U.S. Transfer costs vary signif-

icantly due to inconsistent pricing units and models.

Snowﬂake charges C50/TB, reﬂecting high interna-

tional bandwidth and security costs. BigQuery ap-

plies C0.5/GB, suiting variable transfer needs. Red-

shift is more competitive at C0.2/GB, indicating in-

frastructure efﬁciency. Azure offers the lowest rate at

C0.047/GB, making it an appealing option for users

with frequent or large-scale transatlantic transfers.

These costs are shaped by multiple complex fac-

tors that directly affect data strategy and ﬁnancial

planning.

Physical Distance and Infrastructure: Varia-

tions in distance between European and U.S. data cen-

ters impact latency and transfer costs, which tend to

rise with reliance on long-distance networks and in-

ternational gateways.

International Transfer Fees: Cross-border trans-

fers incur higher fees due to ISP charges and regula-

tory overheads tied to international data movement.

Data Protection Regulations: Divergent laws,

like the GDPR in Europe and U.S. state-speciﬁc

rules, require additional compliance and security

measures, increasing operational complexity and as-

sociated costs.

5.3 ETL Tools

Table 13 shows information on the costs associated

with the ETL tools Azure Data Factory, AWS Glue,

and Cloud Data Flow. Each tool offers a unique

billing model that can signiﬁcantly impact the total

cost of operation, depending on the volume of data,

frequency of task execution, and complexity of the

data operations involved.

Selecting a Data Warehouse Provider: A Daunting Task

557

Table 13: Comparative Costs of ETL Tools.

ETL Tools

Orchestration

($/run)

Execution

($/unit)

Azure Data Factory $0.001 per run $0.000045 per min

AWS Glue $0.025 per second $0.00025 per GB

Cloud Data Flow $0.05 per Vcpu-hour $0.023 per GB

5.3.1 Azure Data Factory

Features an advantageous pricing model for large-

scale operations, with a charge of 1 dollar per 1,000

executions. This cost makes it ideal for scenarios

involving numerous pipelines that are executed fre-

quently but are relatively light in terms of data pro-

cessing. In addition, the cost of 0.168 per hour for

executing and debugging data ﬂows effectively sup-

ports processes that require less processing time, of-

fering a cost-effective option for maintaining contin-

uous operations without incurring high costs. How-

ever, it is essential to note that there is a charge for in-

active pipelines, stipulated at 0.80 per month, which

can introduce additional costs if not well managed.

5.3.2 AWS Glue

Adopts a billing methodology that emphasizes pro-

cessing time, with a rate of 0.29 dollars per second.

This model can result in high costs for processing

that takes long periods or is highly complex. Such a

pricing structure can prove disadvantageous for ETL

projects that require substantial processing power or

extend over long time intervals, making cost manage-

ment a critical aspect. In addition, charging 0.00025

dollars per GB of data processed presents a compet-

itive cost for operations that handle large volumes of

data. This pricing per volume of data processed of-

fers cost predictability that beneﬁts large-scale opera-

tions, facilitating ﬁnancial planning for organizations

that handle large amounts of data.

5.3.3 Cloud Data Flow

The cost per vCPU-hour in Cloud Data Flow, set at

US$0.05, can make this tool a relatively more expen-

sive option for CPU-intensive processing, especially

when compared to alternative solutions that do not

implement vCPU-based charging. This charging is

based on the allocated processing capacity, applied in-

dependently of the actual utilization of resources.

Cloud Data Flow offers a cost of 0.023 dollars per

GB of stored data, which remains constant regardless

of data usage or activity. This pricing model is similar

to the one adopted by AWS Glue, providing a reason-

able and predictable cost for data storage.

5.4 DW Selection

To validate the proposed comparison methodology,

we applied it to a real-world scenario involving the

selection of a cloud DW platform for a startup in the

early stages of building its analytics infrastructure.

The main requirements of this scenario included: (i)

limited initial investment capacity, (ii) expectation of

workload growth over time, (iii) preference for server-

less capabilities to simplify management, and (iv)

ease of integration with existing tools such as Power

BI.

Among the evaluated options, Azure Synapse was

identiﬁed as the most suitable solution given this con-

text. Azure presented a well-balanced hardware con-

ﬁguration and efﬁcient costs for 1 TB per month stor-

age at $4.60/TB in its serverless model, in addition

to progressive offers for long-term contracts (detailed

in Table 11). It stood out for its competitive cost of

serverless computing, as presented in section 5.1.3,

and moderate prices for capacity-based operations,

detailed in section 5.1.2. The native integration with

Power BI adds even more value to our decision.

The selection process followed the evaluation cri-

teria and cost models described in the Methodology

section. This included standardizing hardware conﬁg-

urations (Table 10), comparing operational cost struc-

tures, and considering regional storage implications

(Table 12). The structured comparison supported the

decision, and the methodology proved helpful in nar-

rowing the options in a practical and reproducible

way.

However, it is essential to note that this choice is

mainly linked to the speciﬁc characteristics of our task

and could vary for other problems.

Professionals in similar startup contexts, where

budget constraints, platform simplicity, and native in-

tegration with existing tools are essential, may beneﬁt

from applying the same methodology to guide their

selection.

Another decisive factor in selecting the DW to

use is prior knowledge of the providers’ cloud plat-

forms. Although this is ultimately a qualitative factor,

if the company already uses other services from the

same providers, it might reduce learning and adapta-

tion costs.

As for the other solutions, we highlight that Big-

Query is suitable for large-scale analysis within the

Google Cloud ecosystem, with a query-based pricing

model, which is ideal for companies that need real-

time streaming analysis and integrations with ma-

chine learning. AWS Redshift, on the other hand,

is best for companies that can plan data usage and

commit to the long term. It offers signiﬁcantly re-

DATA 2025 - 14th International Conference on Data Science, Technology and Applications

558

duced prices through one- and three-year commit-

ments, ensuring long-term cost-effectiveness. Finally,

Snowﬂake stands out for its performance in multi-

cloud environments and its architecture, which sep-

arates storage from computing. It enables nearly un-

limited scalability and a strictly pay-as-you-go model,

ideal for companies with variable data processing

needs.

6 CONCLUSIONS

This position paper highlights the challenges faced by

startups when selecting a cloud-based data warehouse

(DW) provider. While modeling and implementing a

DW are complex tasks, the provider selection process,

though expected to be straightforward, is in practice

intricate due to the diversity in technical conﬁgura-

tions, pricing models, and service structures.

To support this decision, we conducted a system-

atic comparison of four leading DW technologies:

Google BigQuery, AWS Redshift, Azure Synapse,

and Snowﬂake. The analysis focused on critical at-

tributes such as cost, processing capacity, storage, and

integration with ETL tools. We applied the method-

ology in the context of a real-world startup project in

the space trafﬁc management domain, demonstrating

how such a structured evaluation can guide informed

decision-making.

Our ﬁndings underscore the signiﬁcant hetero-

geneity across platforms, which complicates fair com-

parisons and increases the cognitive load on decision-

makers. Based on this evidence, we advocate for

greater standardization in the description of DW ser-

vice offerings, particularly regarding resource speci-

ﬁcations and pricing transparency, to facilitate more

accessible and equitable evaluations across different

organizational contexts.

ACKNOWLEDGEMENTS

This work was supported by Project No. 7059

- Neuraspace - AI ﬁghts Space Debris, reference

C644877546-00000020, supported by the RRP - Re-

covery and Resilience Plan and the European Next

Generation EU Funds, following Notice No. 02/C05-

i01/2022, Component 5 - Capitalization and Business

Innovation - Mobilizing Agendas for Business Inno-

vation.

This work was also partially ﬁnanced through na-

tional funds by FCT - Fundac¸

ao para a Ci

encia e

a Tecnologia, I.P., in the framework of the Project

UIDB/00326/2025 and UIDP/00326/2025.

REFERENCES

Aftab, U. and Siddiqui, G. F. (2018). Big data augmentation

with data warehouse: A survey. In 2018 IEEE Inter-

national Conference on Big Data (Big Data), pages

2785–2794.

Ahlawat, P., Borgman, J., Eden, S., Huels, S., Iandiorio, J.,

Kumar, A., and Zakahi, P. (2023). A new architecture

to manage data costs and complexity. Boston Consult-

ing Group (BCG), pages 1–12.

Aljamal, R., El-Mousa, A., and Jubair, F. (2019). A user

perspective overview of the top infrastructure as a ser-

vice and high performance computing cloud service

providers. In 2019 IEEE Jordan International Joint

Conference on Electrical Engineering and Informa-

tion Technology (JEEIT), pages 244–249.

Badiuzzaman Biplob, M., Sheraji, G. A., and Khan, S. I.

(2018). Comparison of different extraction transfor-

mation and loading tools for data warehousing. In

2018 International Conference on Innovations in Sci-

ence, Engineering and Technology (ICISET), pages

262–267.

Berisha, B., M

eziu, E., and Shabani, I. (2022). Big data

analytics in cloud computing: an overview. Journal of

Cloud Computing, 11(1):24.

Google (2025). Bigquery: Cloud data warehouse. Ac-

cessed: 2025-10-02.

Harby, A. A. and Zulkernine, F. (2022). From data ware-

house to lakehouse: A comparative review. In 2022

IEEE International Conference on Big Data (Big

Data), pages 389–395.

Inc., S. (2025). Snowﬂake: The data cloud. Accessed:

2025-02-02.

Kim, T.-h., Ramos, C., and Mohammed, S. (2017). Smart

city and iot.

Microsoft (2025). Azure synapse analytics. Accessed:

2025-01-03.

Santos, M. Y. and Costa, C. (2022). Big data: concepts,

warehousing, and analytics. CRC Press.

Serra, J. (2024). Deciphering Data Architectures. ”

O’Reilly Media, Inc.”.

Services, A. W. (2025). Aws ofﬁcial site. Accessed: 2025-

10-01.

Soma, V. (2022). Comparative study of big query, redshift,

and snowﬂake. North American Journal of Engineer-

ing Research, 3(2).

Uddin, M. K. S. and Hossan, K. M. R. (2024). A review of

implementing ai-powered data warehouse solutions to

optimize big data management and utilization. Aca-

demic Journal on Business Administration, Innova-

tion & Sustainability, 4(3):10–69593.

Villamizar, M., Garc

es, O., Ochoa, L., Castro, H., Sala-

manca, L., Verano, M., Casallas, R., Gil, S., Valen-

cia, C., Zambrano, A., et al. (2017). Cost compar-

ison of running web applications in the cloud using

monolithic, microservice, and aws lambda architec-

tures. Service Oriented Computing and Applications,

11:233–247.

Selecting a Data Warehouse Provider: A Daunting Task

559