Elkjøp rebuilt its ecommerce site from Angular on self-hosted Kubernetes to Next.js on Vercel, and the headline figures are vendor-reported: release cycles compressed from seven weeks to roughly a day, and Core Web Vitals moved to all-green. The interesting part is not the speedup. A Nordic retailer running more than a billion dollars in revenue through its digital properties now routes feature flags, cache invalidation, and edge execution through a single host, which turns a frontend decision into an operational one.
What Elkjøp moved, and from what
Elkjøp moved its ecommerce site from Angular served on self-hosted Kubernetes to Next.js 14 on Vercel. The company, the Nordic subsidiary of Currys PLC that also operates under the Elgiganten brand, runs over $1 billion in revenue through its digital properties, according to Vercel’s customer page. That figure predates the migration and is total digital revenue, not a gain attributed to the platform switch. The stack that came out produced slow page loads, SEO gaps, and a developer experience the team found inefficient; the stack that went in is a hosted framework on a managed edge.
It is worth noting what the previous platform was not. An earlier assumption in some coverage pointed at Sitecore; the documented prior stack is Angular and Kubernetes, not a commercial digital-experience platform. That distinction matters because it changes what the migration actually proves. Elkjøp did not swap one vendor’s opinionated commerce suite for another. It traded self-managed infrastructure control for a hosted framework, which is a different kind of bet, with different recovery costs if it goes wrong.
How much faster, and at what release cadence
The self-reported gains are large on both axes. Release cycles dropped from seven weeks to about a day, per Principal Software Engineer Tomas Jansson, as quoted by Vercel. On the Core Web Vitals side, Elkjøp reports a 93% improvement in Cumulative Layout Shift, 60% on Interaction to Next Paint, and 50% on Largest Contentful Paint, with all-green Lighthouse scores, according to the customer page.
These are vendor-authored case-study numbers, not independently audited results, so the direction is more trustworthy than the exact magnitudes. The cadence claim (seven weeks to a day) is the load-bearing one, because it changes how the product organization behaves: a team that can ship daily runs experiments a seven-week-cycle team cannot. Elkjøp uses feature flags inside the Vercel toolbar for gradual rollouts and internal experimentation, per the same blog post, which embeds the platform directly into the release process rather than leaving it as a deploy target.
Core Web Vitals are not vanity numbers for a retailer. INP tracks how quickly a page responds to input, which is the responsiveness a checkout flow depends on; LCP tracks first paint of the main content, which governs bounce on slow connections; CLS tracks visual stability, which costs a sale the moment a buy button slides under a thumb. Google also folds these signals into ranking, so the SEO gaps the Angular stack carried are partly a function of the same numbers. All-green is the version of the migration that pays off in search visibility rather than just developer morale.
Where does Vercel sit in Elkjøp’s revenue path?
After the migration, Vercel is no longer just a deploy target; it is a runtime dependency in the path between an update and a paying customer. For Elkjøp, the confirmed footprint is ecommerce: the storefront, the feature-flag workflow, the incremental static regeneration that keeps product pages fresh, and the edge functions that run close to users. The case study does not establish that in-store tools or a content pipeline have moved onto Vercel, so treat any broader “retail operations” claim as Vercel’s positioning rather than documented fact.
That positioning is explicit. Vercel has rebranded around “Agentic Infrastructure,” organized into three pillars: Agents (durable orchestration and sandboxed environments), Apps (marketing sites, SaaS backends, storefronts), and Platforms (multi-tenant isolation, custom domains), per the company’s homepage. The Apps pillar is where Elkjøp sits today. The Agents and Platforms pillars are where the revenue expansion comes from. A retailer that has already accepted Vercel inside its ecommerce release loop is a warmer target for the next two pillars than a cold prospect.
The risk profile shifts with that depth. When the cache invalidation, the edge runtime, and the feature flags all sit on one vendor, a single outage or a single pricing change cascades straight into checkout during the traffic windows, like Black Friday, that retailers live or die by. On the old Kubernetes stack, an incident was at least a problem the team owned end to end: it could throttle, roll back, or hotfix without a vendor in the loop, and the blast radius was bounded by what it had provisioned. On a hosted edge platform, several of those levers move to the vendor’s roadmap and status page. That is not automatically worse; a specialist running a global edge network will out-provision most internal teams. But it reframes the question from “can we fix this tonight” to “is the vendor’s incident handling good enough for our peak trading day,” and that answer gets evaluated only when it matters.
What is the lock-in cost, and what breaks off Vercel?
The lock-in is structural rather than contractual: Next.js build artifacts are optimized for Vercel, and features such as next/image, ISR cache invalidation, and edge functions work best, or only, on Vercel. From Next.js 15.1, some features may break on non-Vercel deployments, according to independent review aggregation. That source is a review aggregator rather than an independent audit, so treat the “may break” claim as a flagged risk, not a confirmed regression list.
Both the lock-in framing and the pricing figures come from review aggregators, tryorbye and tool.news, not from independent infrastructure audits. They are consistent with widely reported Vercel cost complaints, but the specific dollar figures should be read as community-reported, not verified. The structural point, that several Next.js features are Vercel-optimized and that bandwidth is metered aggressively, stands regardless of the exact numbers.
The switching cost compounds with usage. A team that has built its release discipline around Vercel previews and toolbar flags is not just deploying an app on Vercel; it has internalized the workflow. Moving off means retraining that muscle and rebuilding the flag and preview machinery on whatever replaces it, which is labor measured in quarters, not a redeployment measured in days. That is why structural lock-in is harder to reverse than contractual lock-in: the contract can be renegotiated, but the workflow is already woven into how the team ships.
Can the alternatives match this?
The alternatives can host a Next.js app, but matching what Vercel sold Elkjøp is a different ask, because the product is the integrated workflow, not the deploy target. The loop Elkjøp bought is preview deploys per pull request, feature flags in a toolbar, ISR and image optimization as managed primitives, and an edge runtime in the same console. A competitor that provides the hosting without that loop leaves the buyer to rebuild the workflow pieces itself.
To match the cadence gain, any of the usual alternatives (Netlify, Cloudflare Pages, or self-hosting on cloud infrastructure) would have to reproduce the preview-and-flag workflow on top of the deploy target, since that workflow is where the seven-weeks-to-a-day gain actually lives. They can plausibly compete on price and portability. They match Vercel on integration depth only by asking the customer to assemble the pieces, which is exactly the internal-infrastructure work the migration was meant to shed. For a team that values the workflow above the unit economics, that is a poor trade. For a team that fears the bandwidth bill more than it values the toolbar, it is the obvious one.
What should Jansson’s NDC Oslo talk answer?
The session is the first public practitioner review of the migration since the 2024 case study, and the listed agenda doubles as a checklist for anyone weighing the same move. At NDC Oslo 2026, Jansson plans to cover why Next.js was chosen, how the app router holds up under production load, what working with Vercel is actually like, and how to pitch such a rewrite inside a large organization.
The vendor case study answers the easy questions. The talk is the place to probe the hard ones. A practitioner weighing this migration should want specifics on the real monthly cost trajectory after launch, including any bandwidth or function-invocation spikes. Which Next.js features turned out to be Vercel-only in practice, and what the team did about them. What is still manual that the case study implies is automated. And what the exit cost would look like: how much of the feature-flag and preview workflow is portable, and how much would have to be rebuilt on a different host.
That last question is the one vendor case studies never answer, and it is the one that determines whether Elkjøp’s move is a template to copy or a bet on a single vendor’s pricing discipline. Vercel, for its part, is sized to make that bet feel safe: $200M-plus in revenue, 823 employees, and a $9.3B Series F valuation, per tool.news’s aggregation. Size is not a guarantee of price stability, but it is the difference between a hosting dependency and a vendor that might not exist in three years.
Frequently Asked Questions
How does Elkjøp’s rewrite compare to reMarkable’s Next.js adoption?
reMarkable, another Vercel-cited Nordic brand, adopted Next.js incrementally alongside its existing stack rather than swapping the whole platform at once. That path preserves a fallback and limits blast radius during migration, where Elkjøp’s full replacement of Angular and Kubernetes traded recovery options for a cleaner cutover and the all-green Core Web Vitals that followed.
Which Next.js features actually require Vercel to function?
On-demand ISR revalidation and the optimized image pipeline route through Vercel’s managed cache and transform service. A self-hosted replacement has to stand up its own revalidation endpoint and a separate image-optimization layer, typically built on the sharp library or a third-party CDN, before those features behave the same way off Vercel.
How should a retailer model the Vercel bandwidth risk before signing?
Project peak-day traffic, not average, against the $150-per-terabyte overage. For an electronics retailer whose load concentrates on Black Friday and launch events, a single surge can dwarf the base plan. The practical mitigation is routing image and asset bandwidth through a CDN under your own billing, since image transforms are the largest metered flow, and setting function-invocation caps before the spike hits.
What would force Elkjøp to revisit the Vercel bet?
Vercel’s rebrand around Agentic Infrastructure signals where its roadmap investment moves next: toward Agents and Platforms pillars, away from the Apps tier where ecommerce storefronts sit. If pricing tiers start bundling agentic features Elkjøp does not use, or if storefront-era capabilities stagnate on the roadmap, the cost-benefit that justified the migration erodes without any single incident to point at.
What Core Web Vitals baseline should a similar retailer verify independently?
INP only became a stable Core Web Vital in March 2024, replacing First Input Delay, so the 60 percent gain Elkjøp reports measures a metric that was not stable for most of the Angular stack’s operational life. A retailer comparing platforms should pull its own CrUX field data for the prior period rather than rely on vendor-reported deltas, since the before and after may be measuring different signals.
