# Buying storage during the chip shortage: five lessons to avoid getting stuck > Source: > Published: 2026-09-02 13:38:56+00:00 **Anyone ordering servers or disks today faces lead times of twenty weeks and quotes that remain valid for only two weeks. Said Boukhizou, VP of Products at DataCore and someone with hands-on experience in storage since 2001, explains how his customers deal with this in five lessons.** The chip shortage is everywhere. DataCore itself buys hardware for its labs and waits more than twenty weeks for it. In the market, there are many other examples with similar or even much longer waiting times. Suppliers keep their quotes valid for two weeks; according to Boukhizou, those who sign too late pay two to five times more or see their projects stall. At the same time, demand is growing: NVMe has become the standard but costs a multiple of what it used to. The reason is simple: [AI is swallowing up everything the factories produce](https://itdaily.com/blogs/datacenter/billions-in-investments-barely-bring-us-closer-to-ending-memory-loss/). Boukhizou recently spoke to a customer who wants to buy 250 petabytes on day one for AI training, with an exabyte as the ultimate goal. CEO Dave Zabrowski [already predicted such data explosions two years ago](https://itdaily.be/blogs/software/datacore-van-dataopslag-naar-levens-redden-via-ai/). What do you do as an IT manager with a normal budget and an array that is filling up? Boukhizou is a salesperson, so his lessons all point toward software-defined storage (SDS). Nevertheless, they are concrete enough to be considered independently of his product. **Lesson 1: Buy what you can get, and mix it** The classic model—one brand and one generation of which you order the same box—becomes a trap during a shortage. “With Nutanix or HPE SimpliVity, you have to buy the exact same generation for expansion, at the new price, and wait.” The alternative is a software layer that accepts any x86 server—[the freedom DataCore has been preaching for years](https://itdaily.com/blogs/software/datacore-days-2025-avignon/)—so you choose Lenovo if HPE doesn’t deliver, place an older generation alongside a new one, and buy refurbished if necessary. Hardware availability becomes an input, not a dependency trap At DataCore’s partner conference later this month, a German partner who is doing good business with second-hand servers will speak. “Hardware availability becomes an input, not a dependency trap,” he summarizes. The downside: those who mix become their own integrator and must monitor compatibility themselves. Freedom shifts the work; it does not eliminate it. **Lesson 2: Let software sort the disks** The example that sticks most: a customer with 50 terabytes wants to expand, but flash is too expensive and has an eight-month lead time. Instead of waiting, he buys what is available: NVMe, SSD, and nearline SAS (NL-SAS). Everything is connected to SANsymphony. That software automatically distributes the data across up to fifteen layers based on access frequency and policy. “You might think I’m hallucinating, but this is real life. This isn’t marketing.” Automatic tiering is not exclusive to DataCore, but the lesson is more general: buy capacity where it is available and affordable and leave the performance distribution to software. That is always better than dimensioning your entire array on the most expensive medium. **Lesson 3: Do not encapsulate your data** A shortage makes every dependency more expensive, including vendor dependency. Boukhizou therefore emphasizes standards: data remains accessible via SCSI and S3, without a proprietary protocol or file system, so you can leave whenever you want. The same logic applies to the hypervisor. Now that Broadcom [is making small VMware customers more expensive or dumping them](https://itdaily.com/news/business/broadcom-vmware-licentie-16-72/), DataCore’s migration tool converts VMs to Hyper-V, Xen, XCP-ng, Proxmox, or KVM, using the over-provisioned space that every customer has anyway as a temporary buffer. [Proxmox](https://itdaily.be/nieuws/datacenter/proxmox-lokt-vmware-klanten-met-esxi-alternatief/) usage under SANsymphony increased by more than 200 percent. ArcelorMittal in Fos-sur-Mer demonstrates how far that goes: the French IT manager chose VMware, his Spanish colleague chose Hyper-V, and the same application for the blast furnaces runs under both. Can you read your data without the vendor’s software? ‘No lock-in’ is, of course, a sales argument used by every vendor. The test is simple: can you read your data without the vendor’s software? **Lesson 4: Delete what you do not need** The cheapest terabyte is the terabyte you do not buy. “We are good at creating data and bad at deleting it,” says Boukhizou. He found an Excel 97 file at a CTO’s office with eight-digit phone numbers of employees who had been gone for decades, and a veterinary clinic that keeps MRIs of hamsters for thirty years because the law requires it for humans: one to two petabytes of deceased pets. Regulations such as [NIS2](https://itdaily.be/blogs/security/nis2-wat-waarom/) and DORA require proof that critical data is protected and recoverable, not that you keep everything. What you must keep does not belong on NVMe. Therefore, move cold data to inexpensive object storage. According to Boukhizou, ArcelorMittal does this with the IoT telemetry from its chimneys. Over time, these are moved from SANsymphony to Swarm and remain there immutably and indefinitely for environmental audits. The difficulty is rarely technical: no one in an organization wants to sign off on a deletion. **Lesson 5: Build small and isolated at the edge** Those with hundreds of locations do not have to wait for one large central order. Boukhizou uses a French supermarket as an example. Each store runs on about fifteen VMs on modest hardware, completely isolated. “Only the payment terminal goes onto the internet. Pharma factories and vaccine labs work locally for the same reason.” Small, isolated islands are built with whatever is in stock, including refurbished equipment. They limit the attack surface, and a hacked store does not infect the rest. In times of scarcity, this is also a procurement strategy: thirty small orders spread over time are easier to find than one large one. **Common sense is worth money** None of these lessons are new. Mixing brands, tiering, demanding standards, deleting data, and building decentrally: it was always wise, but as long as hardware was cheap and deliverable within two weeks, it wasn’t worth the effort. The shortage changes that calculation. As long as hardware was cheap and deliverable within two weeks, no one was concerned about it. The shortage changes that calculation. At the same time, a grain of salt is needed for a salesperson whose product happens to be the answer to each of the five points. DataCore itself relies on Windows as the foundation for SANsymphony, on Proxmox and Vates for the hypervisor, and has yet to forge its [five products](https://itdaily.com/news/business/datacore-expands-hyperconverged-infrastructure-acquisition-starwind/) into a single management layer. At ArcelorMittal, this is still done today with two consoles. Furthermore, shortages are volatile: [HPE is already freezing its server prices](https://itdaily.com/blogs/datacenter/server-prices-stabilize-hpe/), and in two years, the lead time could be two weeks again. The habits you learn now to survive the scarcity will continue to save money afterward. That is perhaps the real lesson: the shortage forces a discipline that should never have been optional in the first place.