The Industry Map of Cloud: Fashion Runs to AWS, Big Retail Runs Away From It

“Fashion uses AWS and tech uses Azure” is half right. The half that is right is interesting; the half that is wrong is more interesting still. Cloud choices really do split by sector, but the cause is not the sector itself — it is the constraints attached to that sector. Two companies can both sell clothes, and Zalando and Walmart still made opposite decisions. The reason is not clothing. It is Amazon.

Force ①. Competition — who gets your money and your data

This is the bluntest force, and the strongest.

Large US retailers avoid AWS as a matter of policy. Walmart signed a five-year cloud agreement with Microsoft in 2018 and also runs on Google Cloud. Kroger split its work between Google and Microsoft, while Target, eBay, Best Buy and L.L.Bean went to Google. The summary an industry figure gave when Gap picked Azure in 2019 is close to a definition of this force — they want “a partner that is not going to be a competitor of theirs in any other parts of their businesses.”

The arithmetic is simple. Money spent on AWS becomes ammunition for Amazon’s retail arm, and infrastructure usage patterns are themselves business intelligence. Peak-season traffic curves, new region build-outs and the shape of a data pipeline are exactly the sort of fingerprint a competitor should not be holding.

Fashion brands, though, move the other way. Zalando announced in 2020 that it had selected AWS as its preferred cloud provider, and it runs machine learning workloads there for personalisation, pricing and supply chain. It even moved its SAP systems from on-premises to AWS. Nike has been an AWS customer for years.

The difference is the angle of competition. Walmart and Amazon collide head-on over the same shopping basket, but for a fashion brand Amazon is one channel among many, or a different game entirely. When the threat is not frontal, what remains is technical maturity and the hiring market — and AWS wins that fight. In short, “fashion uses AWS” is the outcome; the cause is that a sector which does not collide with Amazon head-on is free to choose AWS.

Force ②. Contracts already in place — Azure’s real weapon is not technology

The idea that “tech companies use Azure” points in the wrong direction. Silicon Valley-style tech companies and startups actually skew towards AWS and Google Cloud. Where Azure is strong is large incumbents that were already running Microsoft.

For a company that manages accounts in Active Directory, works in Office 365, runs internal systems on Windows Server and buys licences through an Enterprise Agreement, Azure is not a new vendor but an additional line item on an existing contract. Procurement is short, identity is already wired up, and a sales organisation is already sitting at the table.

Two deals show the structure in numbers.

• Coca-Cola signed a five-year, $1.1 billion strategic partnership with Microsoft in 2024 and committed to migrating all of its applications to Azure. Azure OpenAI Service and Copilot were bundled into the same agreement.
• The London Stock Exchange Group (LSEG) entered a ten-year partnership with Microsoft in 2022 with a minimum cloud spend commitment of $2.8 billion, and Microsoft acquired roughly a 4% equity stake in LSEG. A cloud contract that extended into a capital relationship.

What to read here is not “who sells the better virtual machine.” It is that bundles and relationships decide infrastructure. In the AI era this structure has only grown stronger. Once Copilot or OpenAI access rides along with the infrastructure contract, a pure infrastructure benchmark drops down the list of things that decide the deal.

Force ③. Data gravity and the AI stack — the gap Google works

Google Cloud is third, yet its presence in certain sectors is different in kind. It enters through analytics and AI and works its way down to infrastructure.

LVMH is the clearest case. The luxury group, which holds 75 maisons, built a data and AI platform with Google Cloud to gather data that had been scattered brand by brand into one layer, and is putting predictive, generative and agentic tools on top of it. Data consolidation is hard in luxury for organisational rather than technical reasons: each maison operates independently and is reluctant to share brand assets. That problem is not solved by “our virtual machines are cheaper.” The data warehouse and the governance layer sell first, and compute follows them in.

In media, Spotify has remained a reference case since moving to Google Cloud in 2016, and in retail the anti-Amazon sentiment described above flowed towards Google. Google’s sector map, in other words, is drawn where “places whose analytics problem is urgent” overlaps with “places that must avoid Amazon.”

Force ④. Regulation and sovereignty — the force that deletes options

If the first three forces are preferences, this one is a constraint.

Finance. In the US, Capital One closed its data centres and moved entirely to AWS, and is cited as close to the only large bank to have done so. JPMorgan and Wells Fargo, by contrast, went multicloud. The larger the institution, the more concentration risk on a single provider becomes a supervisory matter in its own right — and a forty-year-old mainframe core cannot simply be lifted and shifted. A bank’s hybrid setup is not a matter of taste but an equilibrium produced by regulators, legacy systems and negotiating leverage.

Public sector and defence. The US Department of Defense’s JWCC is a contract worth up to $9 billion opened simultaneously to AWS, Microsoft, Google and Oracle in December 2022. In June 2026 the Navy issued task orders to all four providers, covering processing up to the highest classification level (IL6) and air-gapped edge options. Here multicloud is not an efficiency play but the result of banning single points of failure and vendor lock-in at the procurement stage.

Sovereignty. Europe changed shape in 2026. AWS launched its European Sovereign Cloud on 15 January 2026 — physically and logically separated from existing regions, operated exclusively by EU residents, with a stated €7.8 billion investment in Germany alone through 2040. Microsoft plans general availability of its sovereign public cloud across European regions during 2026, and Google is taking the partner route with T-Systems in Germany, S3NS in France, Minsait in Spain and Telecom Italia. The EU pinned down eight requirements in its Cloud Sovereignty Framework in October 2025.

One point deserves a cold reading, though. As of 2026 there is still no law that neutralises the extraterritorial reach of the US CLOUD Act. Keeping data in Europe and having Europeans operate it does not change the fact that the parent company is American. That is why some European public-sector and healthcare organisations still insist on domestic providers or on-premises.

Healthcare. All three clouds support a HIPAA business associate agreement, so regulation does not pick the vendor. What splits the field is the shape of the workload. Large-scale genomics and simulation lean towards AWS (Moderna’s rapid scale-up during the pandemic is the canonical example), research and federated data platforms towards Google (Mayo Clinic), and hospital administration with heavy existing Microsoft estates towards Azure.

Force ⑤. Physics — latency, the shop floor, egress

The last force cannot be negotiated away.

Manufacturing. Volkswagen’s Industrial Cloud runs on AWS. It began as a structure for pooling machine and equipment data from plants worldwide onto one platform, and by 2026 had expanded into a digital production platform connecting 43 factories across three continents with 1,200 AI systems, with the agreement extended for another five years. Factory data is enormous, and most of it has to be judged on site in milliseconds. So the answer in manufacturing is always edge plus central cloud, and in that pairing the cloud is not “the place that replaces the factory” but “the place where what the factory produces is gathered and trained on.”

Media and gaming. Netflix is a flagship AWS customer, yet it delivers video through its own CDN, Open Connect, because the pressure point in streaming economics is not compute but egress volume. Gaming is the same. Xbox Cloud Gaming runs on Azure and Amazon Luna on AWS. Here the cloud choice is not an infrastructure decision but a parent-company decision.

The sector map on one page

Hybrid is not one strategy

“We run hybrid” mixes together three things of entirely different character.

First, the workload split. Core on-premises, front end and analytics in public cloud. Most banks and manufacturers sit here, and it is the easiest form to defend.

Second, the negotiating card. A second cloud kept for price negotiations rather than genuinely used. It produces real discounts at renewal, but the cost of one team maintaining two sets of operational knowledge accumulates quietly.

Third, accidental multicloud. The residue of acquisitions and business units buying on their own. It looks like strategy but is really several invoices. Most “multicloud companies” are in this category, and admitting it is the healthier position.

The real cost of multicloud usually shows up in egress fees and headcount. That friction is starting to be filed down, though. AWS and Google Cloud have jointly launched multicloud networking, with talk of Azure being included during 2026. The moment vendors reduce the friction, “hybrid is too hard” stops working as an excuse, and the burden shifts to explaining why you should split at all.

The flow back deserves attention too

If cloud were always the answer, the word repatriation would not exist. 37signals documented its 2023 departure from public cloud onto its own hardware and claims savings of roughly $1.3–1.5 million a year, about $7 million over five years. In a Barclays CIO survey, 86% of respondents said they planned to move some workloads from public cloud back to private or on-premises environments.

Both numbers need careful reading. 37signals is a company with predictable traffic and a team capable of operating infrastructure itself, and “some workloads” in a survey is not a full retreat. What comes back is generally predictable workloads that run near peak load all the time. Workloads with heavy seasonality and frequent experimentation still favour public cloud. By sector, repatriation is discussed most in SaaS and media back ends with flat traffic, and almost never in retail and e-commerce, where peaks jump tenfold.

Korea’s map looks somewhat different

Two additional forces operate here.

First, the public sector is effectively domestic territory. CSAP (cloud security certification) requirements have hardened a structure in which Naver Cloud, NHN Cloud and KT Cloud divide public-sector work between them. That framework is being reworked: guidelines are to be revised during 2026 with full enforcement from July 2027, so how the threshold for the public market shifts after that is the thing to watch.

Second, financial network separation rules are loosening. In January 2026 the financial authorities gave advance notice of an amendment to the enforcement rules of the Electronic Financial Supervisory Regulation, explicitly exempting SaaS use on a financial institution’s internal business network from network separation requirements. Korean financial firms chose hybrid not out of architectural preference but because of that single rule. If the condition eases, the sector’s cloud map may be redrawn over the next few years.

In the private market AWS remains far ahead. The data is somewhat dated, but a 2023 government survey of value-added telecommunications businesses found AWS used by 60.2% of Korean companies that use cloud, and Naver Cloud by 20.5% (multiple responses allowed). It reflects the AWS concentration in gaming, commerce and startups overlapping with the in-house IT subsidiary structure of the large conglomerates.

So what should you be asking?

Choosing a cloud by reading the label on your industry is the wrong order of operations. First find which of the five forces is actually binding on your organisation.

1. Does the company threatening your revenue own that cloud? If so, the technical comparison matters less.
2. What contracts are you already tied into? Identity, Office, ERP and database licences are already half of the answer.
3. Where does your heaviest data sit right now? Data is the most expensive thing to move, and compute follows data.
4. Do the rules that govern you delete options? If they do, compare only within what remains.
5. Is your pressure point milliseconds or egress volume? The former is an edge problem and the latter a CDN design problem — neither is a question about cloud brands.

Answer those five and the field usually narrows to one or two. That is where the real comparison begins.

⚠️ The cases and figures in this article are drawn from public announcements by the companies and institutions involved and from reporting available as of September 2026. Cloud market share varies widely by research firm (roughly AWS 28%, Azure 20% and Google 15% in Q2 2026), and much about contract size and scope of use is not disclosed. Repatriation figures rest on self-reporting and surveys, so generalise from them with care. A company using one cloud does not mean it uses no other.

Frequently asked questions

Do fashion brands really use AWS that heavily?

AWS cases are noticeably common in fashion e-commerce. Zalando announced AWS as its preferred cloud provider in 2020 and moved machine learning workloads and even SAP there, and Nike has long been known as an AWS customer. It is hard to explain this as “because it is fashion,” though. The closer reasons are that sectors which do not compete head-on with Amazon are free to choose AWS, and that ML services for personalisation and demand forecasting were available there early.

Is it true that tech companies use Azure?

The received wisdom points slightly the wrong way. Startups and internet companies actually skew towards AWS and Google Cloud. Azure is strong among large incumbents already running Active Directory, Office 365 and Windows Server, and in finance, manufacturing and the public sector — because it extends an existing contract rather than introducing a new vendor, which shortens both procurement and accreditation.

Does multicloud protect you from vendor lock-in?

Only partly. Few organisations can genuinely run the same service on two clouds at once; most split workloads or keep a second contract for negotiating purposes. In exchange for reduced lock-in you take on egress fees, two sets of operational knowledge and duplicated security policy. The realistic approach is not making everything portable, but writing the parts you might actually move in standard technology.

Why do Korean financial firms run hybrid?

Regulation was the main reason. Network separation rules made it difficult to move internal business systems to public cloud or SaaS, so a configuration of channels and analytics in the cloud with the core inside the perimeter became standard. That premise began to shift in January 2026, when the authorities gave advance notice of an amendment explicitly exempting internal-network SaaS use. Easing the rule does not move core systems overnight, however.

Is European sovereign cloud actually safe?

In terms of physical data location, the nationality of operating staff and governance structure, it is clear progress. AWS launched its European Sovereign Cloud in January 2026, and Microsoft and Google are responding in their own ways. But there is still no legal instrument that removes the extraterritorial effect of the US CLOUD Act. So for an organisation that must strike “possible US government access” off its risk register entirely, a sovereign cloud may not be enough on its own.