# ERP cost 2026: the hardware changing your IT budget

**ERP cost 2026 hardware** is a budget line most companies never agreed to. Every company using management software signed a contract with a vendor at some point. That contract defined features, licences and maintenance fees. Almost no contract mentioned the hardware architecture the software was designed for, or the economic consequences of that choice over time. In 2026, for many companies, that omission is turning into a real and growing cost.

## The inflation arriving from far away

The market data this August are unequivocal. Enterprise SSD prices have reached about 6.5 times the value of the third quarter of 2025, with 30 TB units quoted around 22,600 dollars. Nvidia's main customers have received notices of increases above 15% on AI systems for 2027. Server DDR5 memory is up 307% since September 2025, according to TrendForce. OVHcloud, one of the main European cloud providers, has already announced increases of 5% to 10% on its services because of these upstream costs.

The cause is known: demand for AI hardware — datacentres, GPU clusters, inference systems — has absorbed the production capacity of the main semiconductor makers, taking components away from the traditional enterprise market and pushing prices to structurally higher levels. Samsung, SK Hynix and Micron have redirected production towards the high-bandwidth memory AI requires, reducing supply for ordinary business segments.

So far, a macroeconomic phenomenon. The point that matters for anyone running a company is another: this shock transfers to the operating cost of management software through channels most owners have not yet mapped.

## The SAP HANA case: an architectural bet the market disproved

The most illustrative case is SAP HANA, the in-memory database launched by SAP between 2010 and 2011 and now the mandatory engine of S/4HANA, the current generation of the German multinational's ERP, announced in 2015 and made progressively mandatory with the end of support for the previous version, SAP ECC, set for 31 December 2027 with paid extensions to 2030 and 2033 for some configurations.

The idea behind SAP HANA was technically consistent with the context at the time: keep the entire operational dataset in RAM rather than on disk, removing read times and making processing almost instantaneous. In 2012-2015, with RAM steadily falling in price and the semiconductor trajectory seemingly predictable, that bet looked reasonable.

SAP HANA's licence model reflects this architecture very directly: **the cost of on-premise licences is calculated on the amount of physical RAM allocated to the system**, measured in gigabytes. Not on the number of users, not on activated modules, not on transaction volumes — on RAM. Each time the company grows, data increases, or the system is scaled, the basis for calculating the fee rises with it.

According to recent analysis published by SAP licensing specialists, about 35% of companies using Enterprise Edition licences could requalify to Standard Edition, with estimated savings of 25% to 40% on HANA licence costs alone. The figure shows how widespread over-sizing is, often unintentional, a direct consequence of organic data growth that was never planned in terms of licensing impact.

Companies migrating from SAP ECC to S/4HANA — now under pressure from the end-of-support deadlines — have to assess RAM hardware investments in a market radically different from the one where SAP HANA was designed. Consulting costs for migrations represent on average 60-70% of the total project budget, according to industry data. On top of that comes the hardware component, which in 2026 weighs significantly more than any migration plan drafted before 2024 had foreseen.

## The general mechanism: architectural assumptions and market risk

The SAP case is the most visible, but the principle it illustrates applies to any software designed around hardware assumptions specific to the moment it was conceived.

Every software system embeds, in its architecture, hypotheses about the technology context in which it will operate: the availability and cost of components, expected performance, the licence models of the layers below. When those hypotheses hold over time, the system works as intended. When the market disproves them, the customer ends up paying a cost it did not choose and cannot control.

Standardised management software — designed to be sold to thousands of companies with different structures, in different sectors, with different volumes — necessarily embeds those assumptions rigidly. The vendor optimises the architecture for an average profile, not for each client's specific profile. When the hardware market changes, every customer of that platform suffers the same exposure, regardless of their real needs.

A system designed around the company's actual specifications — real volumes, existing processes, the hardware available at the time of construction — works with a different logic. It can be sized consistently with the context, updated progressively when the context changes and recalibrated without going through vendor-imposed migration cycles. Customisation is not an aesthetic preference: it is a concrete form of long-term control over technology risk.

## What is happening to small-company IT budgets

The dynamics described show up specifically for smaller organisations through three main channels.

The first is the **rise in SaaS fees**. Cloud providers and SaaS vendors that deliver services on their own infrastructure are starting to pass higher hardware costs on to customers. OVHcloud was among the first to say so publicly. By 2027, according to available market analysis, many providers will renew the hardware of their clusters with components bought at post-crisis prices, no longer able to rely on stock accumulated earlier. That impact will reach their price lists.

The second channel is the **refresh of on-premise hardware**. Companies managing their own infrastructure — internal servers, NAS, storage systems for business databases — now face a market where the same server that cost a certain amount in 2024 costs noticeably more in 2026, with RAM and storage components where inflation is structural rather than seasonal.

The third channel, less visible but potentially more expensive, concerns **forced migrations**. When a vendor imposes a migration to a new version of the platform — as SAP is doing with the sunset of ECC — the customer does not choose the moment. It suffers it in a market context that can be unfavourable, with hardware costs adding to already substantial consulting and training costs.

## The hidden cost no quote shows

The point that rarely emerges in preliminary assessments is this: the total cost of ownership of management software includes variables that depend on markets over which the customer has no lever.

Standard management software carries a structural, not accidental, exposure to market risk. Every architectural choice by the vendor — which database it uses, how it manages memory, which components it is optimised for — affects the customer's costs for the whole duration of the contract, and often beyond.

Correct assessment of a management-software investment should include, alongside the cost of the licence and implementation, an estimate of architectural risk: how much does the system depend on hardware components subject to price volatility? What is the cost of a forced migration in five years? How much does the hardware needed to sustain the promised performance weigh in the current price context?

These questions rarely appear in vendor demos. They emerge, with all their concreteness, when the contract renewal, the migration quote or the price-increase notice arrives. Designing a software system calibrated on the company's real needs, the available hardware and its actual growth prospects remains the choice that offers the greatest control over costs over time. Market variables cannot be eliminated, but their exposure can be governed — provided it was considered before signing, not after.

Hardware is only the visible part of the bill, because [the vendor lock-in behind the model choice](/en/ecorner/2025/ai-vendor-lock-in-risk.html) decides how much it costs to leave.

## Frequently asked questions

**Why are SSD and RAM prices rising so much in 2026?**
Demand for AI datacentre hardware has taken production capacity from traditional enterprise segments. The main manufacturers — Samsung, SK Hynix, Micron — have redirected production towards high-bandwidth memory for GPUs, reducing supply for the ordinary business market and pushing prices to structurally higher levels than in 2024.

**What is SAP HANA and why does it matter for Italian SMEs?**
SAP HANA is the in-memory database on which S/4HANA, the current SAP ERP version, is based. Support for the previous version, SAP ECC, ends on 31 December 2027, with paid extensions to 2030-2033. Companies using SAP ECC are therefore in a forced migration window, at a time when hardware costs — central to the HANA architecture — are significantly higher than forecast a few years ago.

**How do AI hardware increases transfer to the cost of SaaS management software?**
SaaS vendors deliver their services on cloud infrastructure, whose cost depends directly on RAM, storage and compute prices. When those prices rise, providers pass the costs on to their price lists, typically with a few quarters' delay. OVHcloud has already announced increases of 5% to 10% for 2026. The phenomenon will extend progressively to other providers as stock bought at earlier prices runs out.

**What is vendor lock-in for management software?**
Vendor lock-in is the condition in which changing software vendor becomes economically or technically prohibitive. For management software, it shows up through data in proprietary formats, dependence on vendor-specific infrastructure, high migration costs and loss of customisations built over time. According to available analysis, migration costs between systems average 50-80% of the original development or implementation cost.

**Does custom software really cost less in the long run?**
The correct comparison is not the initial cost of the two approaches, but total cost of ownership over five years. A custom system requires a higher initial investment, but it has no rising licence fees, imposes no forced migrations and can be sized on the hardware most suited at the time of construction. Industry studies indicate that over two to four years the TCO of custom software is generally lower than that of an equivalent standard package.

**How do you assess the architectural risk of management software before choosing it?**
The relevant questions are: what kind of database is the system built on, and how does it manage memory? What is the structure of the licence fee — does it depend on RAM, users, modules, transactions? Does the vendor have a history of forced migrations, and how often? What are the estimated exit costs from the platform? This information does not appear in commercial demos but emerges from the technical and contractual analysis before the choice.

## Sources

StorageReview — enterprise SSD prices run at 6.5x last year — https://www.storagereview.com/news/enterprise-ssd-prices-run-at-6-5x-last-year-vdura-pegs-a-30tb-tlc-drive-at-22600
Bloomberg — Nvidia customers notified about AI-related price hikes above 15% — https://www.bloomberg.com/news/articles/2026-08-22/nvidia-customers-notified-about-ai-related-price-hikes-above-15
TrendForce — DRAM DDR4 and DDR5 price increases since September 2025, cloud prices up 5-10% by 2026 — https://www.tomshw.it/business/ovh-prezzi-cloud-in-rialzo-del-5-10-entro-il-2026-2025-11-24
SAP Licensing Experts — SAP HANA licensing: in-memory platform costs, metrics and negotiation strategies for 2026 — https://saplicensingexperts.com/blog/sap-hana-licensing-guide
K2 Partnering Solutions — SAP S/4HANA migration: deadlines and costs — https://k2partnering.com/it/consulting/practice-areas/sap-consulting-services/s-4hana-migration/
Pizero.dev — the real TCO and hidden costs of custom software for SMEs, 2026 — https://pizero.dev/blog/quanto-costa-un-software-su-misura-tco-reale-e-costi-nascosti-per-pmi-2026/