AI Demand Is Tightening IT Supply

Before approving hardware for an AI project, verify five things: the exact configuration, where it is physically available, the honest lead time, the permitted substitutions and the support route after delivery.

A product page or portal result is not proof that the required machine can be supplied on the project's timeline.

CloudCoCo's laptop example makes the problem concrete

On 19 August 2026, CloudCoCo described a customer requirement for an AI-workload laptop whose standard OEM configuration did not meet the required memory specification. CloudCoCo said it sourced the memory separately and built the machines to the required specification rather than asking the customer to wait months. Read CloudCoCo's post.

The useful lesson is broader: the buyer's requirement was fixed by the workload, but the obvious purchasing route could not fulfil it. Somebody had to verify compatibility, find an alternative supply path and remain accountable for the finished configuration.

That is different from casually substituting whatever is available. A viable alternative must still meet performance, compatibility, warranty, security and support requirements.

What memory manufacturers are actually saying

In my May 2026 LinkedIn post, I argued that AI was absorbing semiconductor capacity and changing the economics of enterprise refresh cycles.

The precise numbers move quickly, so the public article should lean on current primary evidence rather than repeat every market estimate.

Micron's fiscal first-quarter 2026 prepared remarks said customers' AI data-centre plans had sharply increased demand forecasts for memory and storage. It expected aggregate industry supply to remain substantially short of demand, with tight DRAM and NAND conditions persisting through and beyond calendar 2026. Read Micron's prepared remarks.

SK hynix's July 2026 results likewise attributed record performance to strong AI demand and high-value DRAM, HBM and enterprise SSD products. It described a structural broadening of demand across AI and conventional memory. Read SK hynix's Q2 2026 results.

These sources support the direction of travel: AI demand is tightening parts of the memory and storage market. They do not prove that every laptop or server is unavailable, or that one lead time applies to every buyer.

An old friend's view of the GPU-configuration problem

I have also been discussing a related problem with an old friend, Deep Toor, Director at CloudCoCo.

Deep and I were trying to source a Lenovo ThinkPad P16 with an NVIDIA RTX 5000-series professional laptop GPU. The availability constraint we encountered was that those GPUs had been ring-fenced for AI data-centre build-outs. A configuration that existed in the range was therefore not necessarily available to our project through the normal route.

That is our experience of a specific requirement, not a claim that every RTX 5000-series card or every Lenovo P16 configuration is unavailable.

That point should remain model-specific. It would be easy to turn a real availability problem into a sweeping claim that every OEM has removed every high-end option. The evidence does not support that.

The practical buying question is narrower and more useful:

Can this exact CPU, GPU, memory, storage, display, security and warranty configuration be supplied in the required quantity, country and timeframe?

If the answer is no, the next questions are whether a different factory configuration, validated memory upgrade, alternative model or revised delivery sequence still satisfies the workload and support requirements.

Lenovo illustrates why the exact model matters

Lenovo's UK pages currently show that the ThinkPad P16 Gen 3 is configurable and can include discrete NVIDIA RTX PRO graphics and high memory configurations. Lenovo also describes the ThinkPad P16s Gen 5 as supporting customer-replaceable memory and other components. Those are useful product facts, not evidence that every desired configuration is in stock. See Lenovo's ThinkPad P16 Gen 3 and ThinkPad P16s Gen 5.

For an AI or engineering workload, buyers may need to document:

• GPU model, VRAM and required driver or application certification;
• system memory capacity, type, channel population and upgrade route;
• CPU, NPU and thermal requirements;
• storage capacity, performance, encryption and spare slots;
• power supply, dock and display requirements;
• operating-system and endpoint-management compatibility;
• warranty and the effect of any third-party component installation;
• quantity, delivery country and latest acceptable date; and
• an approved fallback configuration.

The Lenovo P16 discussion with Deep is separate from CloudCoCo's published customer example.

Can the RAM or GPU be upgraded after purchase?

Sometimes — but the answer belongs to the exact machine type, service manual and warranty terms, not the badge on the lid.

RAM and GPU changes should also be treated differently. Selected business laptops and mobile workstations have accessible SODIMM memory slots. A laptop GPU is more often fixed to the system board, supplied as a factory configuration or classed as a service part rather than a customer upgrade. A few workstation designs use a removable GPU card, but removability does not by itself prove that a different card is validated, orderable or customer-installable.

Lenovo

Lenovo's specification for the ThinkPad P16 Gen 3 lists DDR5 SODIMM memory and factory GPU options up to the NVIDIA RTX PRO 5000 Blackwell Generation Laptop GPU. Lenovo also uses a Customer Replaceable Unit model for parts such as memory on supported machines. The exact P16 machine type and its hardware-maintenance documentation still need to confirm which parts may be changed, by whom and with which supported part numbers. ThinkPad P16 Gen 3 specification and Lenovo CRU guidance.

Dell

Dell publishes service procedures for replaceable memory and, on selected Precision mobile workstations, removable GPU cards. Some newer graphics procedures are explicitly intended for authorised service technicians. Dell also states that upgraded third-party parts do not take on the computer's Dell warranty. That makes Dell Precision a model-by-model engineering and warranty check, not a blanket promise of customer GPU upgrades. Dell Precision 7780 GPU procedure and Dell guidance on upgraded parts.

HP

HP's warranty framework includes Customer Self-Repair parts, but the classification varies by product and country. HP may require third-party options to be removed during fault isolation, and the separately fitted component and any installation damage should not be assumed to be covered by the base-system warranty. Check the maintenance guide, part classification and UK warranty for the exact EliteBook or ZBook. HP worldwide limited warranty.

Apple

Current Apple silicon systems generally use memory integrated into the chip package, so it cannot be upgraded after purchase. Apple advises checking the exact model: if its system information does not show upgradeable memory, an upgrade is not available. Apple memory guidance.

Will an upgrade void the warranty?

There is no safe universal answer. The practical distinction is usually between:

• the original machine and parts covered by its OEM warranty;
• an approved OEM option or Customer Replaceable Unit;
• a third-party component carrying its own warranty;
• damage caused during installation; and
• a later fault the OEM believes is connected to the modification.

Before changing a machine outside the factory, record the exact model and serial number, current warranty, service-manual procedure, approved part numbers, component supplier, installer, test results and rollback route. Keep the original parts where practical. Obtain a written answer from the OEM or warranty provider when continued entitlement matters to the business case.

This is also where independent support can be useful. Myrmekes can assess whether a modified or post-warranty system can be covered for defined hardware faults, parts and labour through the available delivery network. That support cannot reinstate an OEM warranty, create access to proprietary firmware or make an unvalidated GPU compatible. The quotation must state which base system and added components are covered, who diagnoses cross-boundary faults and what remains with the manufacturer or component supplier.

A pre-purchase-order checklist

1. Freeze the workload requirement

Record what the system must run, the data size, performance target, concurrent users, required certifications and acceptable compromises. “AI-ready” is marketing language until it is tied to a workload.

2. Verify the bill of materials

Capture manufacturer part numbers, not only a family name. Check which items are factory-installed, field-replaceable or supplied separately, and who validates the finished build.

3. Separate catalogue, distributor and physical stock

A listed product may be orderable without being available. Ask where the units are, how many exist, whether they are allocated, and how long the quoted position remains valid.

4. Test every substitution

An alternative GPU, memory module or model must be checked against workload performance, compatibility, warranty, driver support, security controls and the customer's standard build.

5. Confirm ownership after delivery

Who supports the base machine, the separately sourced component and the completed configuration? Who holds spares? Which party diagnoses a fault that crosses those boundaries?

6. Write an expiry into the quote

In a constrained market, availability evidence ages quickly. State when stock and pricing must be rechecked before order.

If the refresh slips, extend deliberately

Supply pressure does not mean every existing server or laptop should be kept indefinitely. It means the refresh decision should be tested against the real estate.

For servers and storage, review:

• asset condition and capacity;
• redundancy and failure history;
• parts availability and positioning;
• OEM lifecycle and active entitlement;
• firmware, software and security dependencies;
• business criticality and restoration requirement; and
• the planned refresh or migration trigger.

Assessed third-party maintenance may provide hardware cover for suitable equipment during a delayed refresh. It does not automatically supply OEM firmware, proprietary software rights, licensing or vendor escalation. A hybrid arrangement may therefore be the honest answer.

For laptops, the equivalent decision includes battery condition, security baseline, operating-system eligibility, repairability, user role, prepared spares and data handling. Extending a device that is secure, serviceable and fit for the workload is different from leaving a user on an unreliable machine because procurement is difficult.

How Myrmekes can help

Myrmekes provides a route into hardware fulfilment, infrastructure support and engineering across multi-vendor estates. UK delivery is provided through Cantel's network, and sourcing is coordinated through relevant partners. Exact stock, configuration, delivery party, warranty position and support scope must be confirmed for each requirement.

If an AI, workstation or infrastructure project is being held up by availability, send the exact specification, quantity, locations and required date. We can start by testing the requirement and identifying the questions that must be answered before anybody promises a machine or a service.

• Explore Myrmekes fulfilment
• Review services and engineering options
• Compare infrastructure support-contract options
• Discuss the requirement

Sources

• Duncan Brown's original LinkedIn post, 18 May 2026
• CloudCoCo LinkedIn post, 19 August 2026
• Deep Toor's LinkedIn profile
• Micron fiscal Q1 2026 prepared remarks
• SK hynix Q2 2026 results
• Lenovo ThinkPad P16 Gen 3, UK
• Lenovo ThinkPad P16s Gen 5, UK
• Lenovo ThinkPad P16 Gen 3 specification
• Dell Precision 7780 GPU service procedure
• HP worldwide limited warranty
• Apple memory-upgrade guidance