The Memory Shortage and What It Means for Your Hosting Costs
If your hosting renewal came in higher this year, or your provider quietly changed what a plan includes, there is a specific reason and it is not local. The memory that goes into servers has become dramatically more expensive, and the cause is a deliberate industry-wide reallocation of manufacturing capacity toward artificial intelligence hardware.
This matters to anyone running a website because server RAM is a primary input cost in hosting. When the price of that input rises steeply and stays there, it eventually shows up in what plans cost, what resources they include, and how readily a provider will hand you more capacity. Understanding the mechanism helps you plan rather than simply absorb it.
As a Canadian web hosting provider we buy this hardware, so this is a cost pressure we are dealing with directly rather than reporting on from a distance. What follows is the honest version: what is actually happening in the memory market, why this shortage is structurally different from the one in 2020 to 2023, how long the people who make the chips expect it to last, and the practical steps that reduce your exposure. Several of those steps will save you money whether or not memory prices ever fall.
One note on figures. Every number below is attributed to its source, and the memory market is moving quickly enough that some will be out of date within a quarter. Where forecasts are cited they are forecasts, not outcomes.
What is actually happening in the memory market
The short version: three companies make almost all the world's DRAM, and they have moved a large share of their manufacturing capacity to a different, more profitable product. Everything else follows from that.
High Bandwidth Memory, or HBM, is a specialised form of DRAM used in AI accelerators. It commands substantially more revenue per silicon wafer than conventional server and consumer memory. Faced with that gap, Samsung, SK hynix and Micron have reallocated wafer capacity toward HBM, which is a rational commercial decision and which reduces the supply of the ordinary DRAM that goes into servers, laptops and phones.
The resulting price movement has been severe. J.P. Morgan Global Research has estimated that DRAM prices will have risen more than 400% from the start of 2024 to the end of 2026. J.P. Morgan also notes that the Consumer Price Index for software and accessories and the Producer Price Index for storage devices have each risen 23% since the end of 2024, which is the point at which a component shortage becomes a visible economic effect rather than an industry story.
More recent quarterly data suggests the steepest phase may be easing without reversing. TrendForce's memory pricing survey, as reported by Tom's Hardware, projected conventional DRAM contract prices rising 13% to 18% quarter over quarter in the third quarter of 2026, with NAND flash contract prices up 10% to 15%. Those are still substantial increases, and they represent a marked slowdown from the far larger jumps recorded in the second quarter.
On the supply side, IDC's analysis indicated 2026 DRAM bit supply growing only around 16% year over year, with NAND around 17%. Set against AI data centre demand capable of absorbing that entire increment, the arithmetic explains why the situation does not resolve simply by making more chips.
Why this is not the 2020 to 2023 chip shortage
It is tempting to treat this as a continuation of the shortage that disrupted cars and consumer electronics earlier in the decade. It is a different phenomenon with a different cause, and conflating them leads to the wrong expectations about how it ends.
| 2020–2023 chip shortage | 2025–2026 memory shortage | |
| Primary cause | Pandemic demand swings, supply chain disruption, and order cancellations followed by simultaneous re-ordering | Deliberate reallocation of wafer capacity from conventional DRAM to higher-margin HBM for AI accelerators |
| Nature | Largely a mismatch of timing and logistics | A rational allocation decision under sustained margin differences |
| What resolved or would resolve it | Demand normalising and supply chains restabilising | New fabrication capacity, or AI demand slowing relative to projections |
| Timeframe to fix | Resolved over roughly three years | New memory fabs take years to build; manufacturers cite 2027 and beyond |
| Effect on hosting | Mostly indirect, affecting hardware lead times | Direct, since server memory is a primary input cost |
The distinction that matters most: the earlier shortage was something happening to the industry, and this one is something the industry is choosing, because the economics favour it. That means it does not self-correct as demand normalises. It corrects when either new capacity arrives or the margin gap closes.
Which is why the timelines being quoted are long. Manufacturers and analysts have pointed variously toward 2027, 2028 and beyond for normalisation, with SK hynix's leadership warning that demand will exceed supply well into the next decade. Forecasts that far out are genuinely uncertain and should be treated as such, but the direction of the guidance is consistent: this is not a quarter or two.
Why a memory shortage reaches your hosting bill
The path from a wafer allocation decision to your invoice is shorter than it looks.
Server memory is one of the largest line items in the cost of a hosting platform. A machine configured for virtualised hosting carries a substantial amount of RAM, because memory is typically what limits how many customer environments a host can run on one server before performance degrades. When memory prices multiply, the cost of building or refreshing that hardware multiplies with them.
Providers have a limited set of responses, and it is worth recognising them because you may be experiencing one without it being announced:
- Raise prices on new plans, or at renewal.
- Hold prices and reduce what a plan includes, which is the least visible option and the most common.
- Extend hardware refresh cycles, running existing servers longer than planned.
- Increase density on existing hardware, which is where contention and the noisy-neighbour problem become more likely.
- Restrict or slow resource upgrades, because adding RAM to a plan now costs the provider materially more than it did.
- Move toward longer contract terms, to make input costs predictable.
Only the first is transparent. The one worth watching for as a customer is the fourth, because increasing density is how a provider protects margin without changing the price list, and the consequence lands on you as inconsistent performance at peak rather than as a higher bill.
| The question worth asking your provider Not “are you raising prices”, which invites a reassuring answer, but “has the number of customer environments per physical server changed, and is my CPU and memory allocation guaranteed or burstable?” Guaranteed allocation is what insulates you from a provider's density decisions. Burstable allocation is what exposes you to them, and it becomes a materially worse deal in a market where providers are under pressure to pack machines more tightly. |
What to do about it: reduce what you need before you pay more for it
Here is the useful part, and the reason this is not simply bad news. Most small business sites use more memory than they need to, and the interventions that reduce memory consumption are the same ones that improve performance. They pay off regardless of what happens to DRAM prices.
1. Cache properly, so requests never reach the application
Full-page caching is the highest-leverage change available on most content-managed sites. A cached page is served without executing your application or querying your database, which means it consumes a fraction of the memory and CPU an uncached request does. A site serving most of its traffic from cache needs materially less capacity than the same site serving everything dynamically.
If your caching is currently a plugin bolted onto an untuned stack, there is usually a substantial gain available from moving it to the server level. We cover the difference in server-side caching versus plugin caching.
2. Fix the database rather than feeding it
Memory pressure on a typical WordPress or ecommerce site is frequently a symptom of database inefficiency rather than genuine scale. Unindexed queries, accumulated post revisions, expired transients, orphaned metadata and abandoned plugin tables all consume memory and CPU on every uncached request.
This is the intervention with the best economics in the current market, because it converts a recurring hardware cost into a one-off piece of work. Adding RAM to mask a slow query has always been wasteful; at current memory prices it is considerably more so. Start with the slow query log, and see slow database queries.
3. Audit what is actually running
Every plugin, extension, background worker, monitoring agent and scheduled task holds memory. On sites that have accumulated over years, a meaningful share of resident memory belongs to things nobody uses and nobody remembers installing.
Practical sequence: list what is installed, identify what is genuinely used, remove rather than deactivate the rest, and consolidate overlapping tools. Deactivated plugins are still files on your server, and several categories of tool duplicate each other's function while each holding its own memory.
4. Right-size rather than over-provision
The traditional advice to buy headroom was reasonable when memory was cheap. It is an expensive habit now. The better approach is to measure actual peak memory use, provision for that plus a defined margin, and monitor so you can add capacity when the data says to rather than in anticipation.
That requires monitoring you may not currently have, which is itself worth setting up. Knowing your real peak memory consumption is what lets you decline an upgrade with confidence, and in this market declining an unnecessary upgrade is a direct saving.
5. Match the hosting type to the workload
Memory economics change which hosting choice makes sense. Where your traffic is largely cacheable, a well-configured shared or managed WordPress hosting environment can serve it using a fraction of the dedicated memory a VPS hosting would require, because the provider amortises capacity across many sites that are mostly serving cached content.
Where you genuinely need guaranteed resources, a VPS remains the right answer and the guarantee is worth more than it used to be. And where you need substantial memory for a large database or sustained load, Canadian dedicated servers may be more economical per gigabyte than repeatedly upgrading a virtual instance. The decision has not changed in structure; the cost of getting it wrong has increased.
6. Time your commitments deliberately
If forecasts are right that prices remain elevated into 2027 and beyond, a longer term at today's rate is a hedge rather than merely a discount. That is a genuine argument for a longer commitment, and it is also exactly what a provider would like you to conclude, so weigh it accordingly.
The honest framing: lock in a longer term if you are confident in the provider and the fit, because you are buying price certainty in a volatile input market. Do not lock in a longer term to obtain a discount from a provider you have not tested, because the switching cost during the term is what you would be paying for the discount. If you are considering a move, changing hosts without losing rankings covers doing it without damaging your search visibility.
The numbers, with sources
Rather than scatter figures through the text, here they are together with attribution so you can check each one. Memory pricing moves quarterly, so treat this as a snapshot with a date on it rather than a fixed picture.
| Figure | Source | What it describes |
| DRAM prices rising more than 400% from the start of 2024 to the end of 2026 | J.P. Morgan Global Research | The headline repricing. An estimate spanning a three-year window, not a single-year move |
| CPI for software and accessories and PPI for storage devices each up 23% since end of 2024 | J.P. Morgan Global Research | Where a component shortage becomes a measurable economic effect |
| Conventional DRAM contract prices up 13–18% quarter over quarter in Q3 2026; NAND up 10–15% | TrendForce, reported by Tom's Hardware | Still rising, but a marked slowdown from the far larger increases of the previous quarter |
| 2026 DRAM bit supply growth around 16%; NAND around 17% | IDC | Why simply making more chips does not resolve it quickly, given AI demand can absorb the entire increment |
| Memory can be 15–20% of the bill of materials for a mid-range smartphone, 10–15% for a flagship | IDC | A useful proxy for how heavily memory weights a hardware cost structure |
| Demand expected to exceed supply well into the next decade | SK hynix leadership, July 2026 | The longest of the published views, and from a manufacturer rather than an analyst |
| Normalisation cited variously at 2027, 2028 and beyond | Multiple manufacturers and analysts | The range of published expectations. No consensus date exists |
Two cautions on reading these. Forecasts made during a shortage have a poor historical accuracy record, and the incentive to shift capacity back toward conventional DRAM grows as the margin gap narrows, so the longer projections may prove pessimistic. Equally, every estimate assumes AI infrastructure spending does not accelerate further, which is not guaranteed. The honest position is a wide range with a clear direction.
It is not only RAM: NAND and storage are affected too
Most coverage focuses on DRAM, but the same dynamic reaches storage, and that matters because storage is the other major component in a hosting platform.
NAND flash, the technology behind solid-state drives, has seen its own contract price increases, with TrendForce projecting 10% to 15% quarter over quarter in the third quarter of 2026 following larger earlier jumps. The mechanism is related: AI workloads generate enormous storage requirements alongside their memory requirements, and manufacturers face similar allocation decisions about where to direct capacity.
The practical consequences for hosting customers are threefold. Storage allocations on plans may become less generous, since a provider giving away a large amount of SSD space is now giving away something more expensive. Backup retention becomes a real cost consideration rather than an afterthought, because backups consume the same storage. And the price gap between storage tiers can widen, which changes the calculus on whether a faster storage tier is worth paying for.
What to do about it is unglamorous and effective. Audit what you are actually storing: old backups nobody will restore, uncompressed original-resolution image uploads, log files never rotated, staging copies of the production database, and media from campaigns that ended years ago. On a long-lived site this routinely accounts for a large share of consumed space. Compress and correctly size images before upload rather than storing full-resolution originals you never serve. Set a deliberate backup retention policy instead of keeping everything indefinitely. And keep at least one backup off-site regardless, because the storage saving is not worth the risk of having only local copies.
Our SSD hosting pages set out what current plans include. The general point stands whoever you host with: storage efficiency has become worth attention in a way it was not when space was effectively free.
How to find out what you actually need
Every recommendation in this article depends on knowing your real consumption, and most site owners do not. This is how to establish it, and it is the step that converts general advice into a specific decision.
Establish your baseline
- Record peak memory use across a full week, not a single reading. Memory consumption tracks traffic and background tasks, so a Tuesday afternoon and a Sunday morning are different numbers. Most control panels expose this; if yours does not, that is worth raising with your provider.
- Note what was running at peak. Backups, cron jobs, imports and search-engine crawls all consume memory, and a peak caused by a nightly backup is a scheduling problem rather than a capacity problem.
- Measure server response time on a cached page and an uncached page, at your quietest hour and your busiest. Four numbers. This tells you whether you have a contention problem separate from a capacity one.
- Record the proportion of requests served from cache. If it is low on a site that is mostly public content, you have found your largest available saving before spending anything.
- Check your database size and look at the slow query log. A database that has grown far beyond what your content volume justifies is carrying overhead you can remove.
- Write all of it down with a date. Without a dated baseline you cannot tell whether a later change helped, and you will be arguing from impressions at renewal time.
Then interpret it
| What you find | What it means | What to do |
| Peak memory well below your allocation | You are over-provisioned | Consider a smaller plan, but keep a defined margin. Do not cut to the measured peak exactly |
| Peak memory near your allocation, response times fine | Correctly sized | Leave it alone. Monitor, and revisit if traffic grows |
| Peak memory near allocation, response times degrade at peak | Genuine capacity constraint | Reduce demand first through caching and database work, then upgrade if it persists |
| Low cache hit rate on public content | Configuration problem, not a capacity problem | Fix caching. This is usually the largest single saving available |
| Peak driven by scheduled tasks | Scheduling problem | Move backups and imports away from traffic peaks before buying anything |
| Response times vary by time of day but memory is fine | Contention on shared hardware | This is about your provider's density, not your consumption. Ask about guaranteed allocation |
What this means for different kinds of site
Memory pressure lands unevenly, because how much a site depends on dynamic requests determines how much RAM it consumes per visitor.
| Site type | Exposure | Where the saving is |
| Brochure or small business site | Low | Almost all traffic is cacheable. If you are on a plan sized for dynamic load, you are probably over-provisioned |
| Blog or publisher | Low to moderate | Caching should carry nearly everything. Check cache hit rate before considering an upgrade |
| Ecommerce | High | Carts, accounts and checkout cannot be page-cached, so those requests execute the application. Object caching is the lever here, not more RAM |
| Membership or course site | High | Logged-in traffic bypasses page caching almost entirely. This is the profile that genuinely needs memory, and where efficiency work pays most |
| Multi-site or agency | Moderate to high | Many sites sharing an allocation. Consolidating plugins across sites and standardising caching produces compounding savings |
| Web application or SaaS | High | Dynamic by nature. Focus on application-level efficiency and connection pooling rather than provisioning |
The pattern worth noticing: the sites most exposed to memory costs are the ones with the most logged-in traffic, and for those, object caching rather than page caching is the intervention that matters. Adding RAM without addressing repeated identical database queries is paying a monthly premium to avoid a one-off configuration change.
What not to do
| Reaction | Why it backfires |
| Downgrade to a cheaper plan without measuring | If you drop below your actual peak memory use, the result is errors and slow responses at exactly your busiest moments. Measure first |
| Move to the cheapest provider available | In a market where memory is the constrained input, unusually low pricing is being funded by density, oversubscription or deferred hardware refresh. You will pay in performance variance |
| Accept burstable allocation to save money | Burstable is the mechanism by which a provider's density decisions become your performance problem. It is a worse trade now than it was |
| Add RAM to fix a slow site | If the cause is an unindexed query or missing caching, you are paying a recurring premium to avoid a one-off fix |
| Panic-buy capacity you do not need | Provisioning for imagined growth at current prices is an expensive form of reassurance |
| Assume prices will fall back soon | Manufacturers and analysts are pointing to 2027 and beyond. Planning on a near-term reversal is planning on a forecast nobody is making |
The Canadian angle
A few considerations specific to Canadian businesses, and one claim to be careful about.
Currency exposure is real. Server hardware is priced in US dollars, so Canadian hosting providers absorb both the component price increase and any movement in the exchange rate. That is part of why Canadian hosting costs can move independently of headline component prices, and it is a legitimate explanation rather than an excuse.
Data centre capacity in Canada is also subject to the same AI-driven pressure as everywhere else, since operators building for AI workloads compete for the same power, space and hardware as conventional hosting. If you have specific requirements about where your data resides, it is worth confirming current availability rather than assuming capacity is unconstrained. Our Canadian data centres pages set out locations, and if residency matters to your organisation the time to confirm it is before you need to move.
The claim to be careful about, including when we or our competitors make it: none of this means Canadian hosting is insulated from global component markets. It is not. A provider implying that domestic hosting escapes international hardware pricing is describing something that is not true, and the honest version is that Canadian providers face the same input costs with an added currency dimension.
What to expect, stated as uncertainty
Anyone offering you a confident forecast on this is overreaching. What can be described is the range of views and what each would mean.
The consistent element across analyst and manufacturer commentary is that relief depends on new manufacturing capacity, and that new memory fabrication plants take years to build and bring to volume. That is the structural reason the timelines being quoted extend to 2027, 2028 and in some commentary beyond. Micron's own capacity additions have been reported as not expected to deliver meaningful volume until around mid-2027.
The paths to easing, in rough order of how often they are cited: new capacity arriving; AI infrastructure spending slowing relative to current projections, which would reduce HBM demand pressure and is not something analysts are currently forecasting; and incremental efficiency gains in manufacturing, which deliver gradual rather than dramatic relief.
What almost nobody is predicting is a return to early-2025 pricing in the near term. Even the more optimistic commentary describes stabilisation rather than reversal. For planning purposes, the reasonable assumption is that memory remains materially more expensive than it was, and that you should therefore treat memory efficiency as a durable discipline rather than a temporary response.
There is also a reasonable case that some of the current commentary overstates the duration, since forecasts made during a shortage have a poor historical record and the incentive to reallocate capacity back toward conventional DRAM grows as the margin gap narrows. Holding both possibilities is more useful than committing to either.
The performance connection worth understanding
There is a search and user-experience dimension here that is easy to miss, and it is the reason this is worth attention rather than just budgeting.
If providers respond to memory costs by increasing density on existing hardware, the practical effect on customers is more resource contention. More contention means server response times that degrade at peak, and server response time feeds directly into Largest Contentful Paint, which is one of the three Core Web Vitals Google assesses. Those are measured at the 75th percentile of real user visits, which means peak-hour degradation shows up in the assessment even when your averages look healthy.
So the chain runs from a wafer allocation decision to your Core Web Vitals, via your provider's margin pressure. That is not a reason to panic; it is a reason to measure. Establish your current server response time on a cached and an uncached page, at your quietest hour and your busiest, and record the numbers with a date. If your provider's density changes, you will be able to see it rather than suspect it. We cover the mechanics in Core Web Vitals and hosting and the WordPress-specific side in speeding up slow WordPress sites.
For anyone whose hosting is currently adequate, this is also an argument for not moving unnecessarily. A known-good environment has more value in a constrained market than it does in a comfortable one, and SSD hosting or a guaranteed-resource plan you have already tested is worth more than a cheaper unknown.
How to have the renewal conversation
If your renewal has gone up, you have more room than you might think, and the useful approach is not to demand the old price.
Start by knowing your position. If you measured your consumption and you are over-provisioned, the strongest move is to ask for a smaller plan at a lower price rather than to argue about the increase on a plan you do not need. Providers would generally rather right-size a customer than lose one, and you arrive with data rather than a complaint.
Questions worth asking, in roughly this order:
- Is my current allocation guaranteed or burstable, and has that changed?
- Given my actual peak usage, which is X, is there a plan that fits better than my current one?
- What does a longer term do to the price, and what happens if I need to change plans mid-term?
- Is server-level caching available on my plan, and is it currently active on my site?
- Are backups included in my storage allocation, and what is the retention policy?
- If I reduce my resource usage, can I move down a tier before renewal rather than at it?
What not to do is threaten to leave as an opening position. In a market where providers are managing constrained capacity, a customer who is difficult and unprofitable is one they can afford to lose, and you will have spent your leverage before establishing whether a sensible option exists. Ask what is available first.
One thing genuinely worth pushing on: if a provider has increased density without disclosing it, and your performance has measurably degraded, that is a legitimate complaint with evidence behind it. Which is the practical argument for having taken those four response-time measurements before you needed them.
The wider context, briefly
It is worth understanding what is on the other side of this, because it explains why the pressure is unlikely to ease on its own.
The demand absorbing memory capacity is AI infrastructure construction, and it operates at a scale that dwarfs conventional hosting. Individual large-scale AI infrastructure projects have been described as capable of absorbing a substantial share of global DRAM output at peak build-out. Manufacturers have reported high-bandwidth memory as effectively sold out for the year, and one major producer retired a consumer-facing brand in early 2026 to concentrate on data centre and HBM products, which is about as clear a signal of priority as an industry gives.
Two implications follow for anyone buying hosting. First, conventional web hosting is not the customer these manufacturers are optimising for, and it will not become one while the margin difference persists. Second, the constraint is genuine rather than manufactured scarcity in the pejorative sense: there really is not enough conventional DRAM being produced to meet demand at previous prices, because the capacity is making something else.
That is neither reassuring nor alarming; it is just the shape of the market. The reason to know it is that it tells you what kind of problem you are solving. This is not a supplier being opportunistic that you can negotiate around. It is an input cost change that you address by needing less of the input.
Conclusion
The memory market has repriced, and the cause is a deliberate industry-wide shift of manufacturing capacity toward AI hardware rather than a temporary disruption. J.P. Morgan Global Research has estimated DRAM prices rising more than 400% from the start of 2024 to the end of 2026, supply growth is running well below what demand would absorb, and the people who make the chips are pointing to 2027 and beyond before conditions normalise.
For a business running a website, the practical consequence is that memory is now something to use efficiently rather than to buy generously. The good news is that everything which reduces memory consumption also improves performance: caching properly so requests never reach your application, fixing database inefficiency instead of masking it with hardware, removing what you no longer use, and provisioning against measured peaks rather than imagined growth. None of that depends on component prices to be worth doing.
The risk to watch is not the price list. It is density. A provider under input cost pressure can hold prices and pack machines more tightly, and that arrives as inconsistent performance at your busiest hour rather than as a higher invoice. Guaranteed rather than burstable resource allocation is the protection, and it is worth more now than it was.
If you want to know where you actually stand, measure before you decide anything: server response time on cached and uncached pages, at your quietest and busiest hours, plus your real peak memory use. Those numbers tell you whether to optimise, upgrade, stay put or move. Reliable Canadian web hosting with guaranteed resources is the foundation either way, and the measurement costs nothing.
FAQ
Why has my hosting price gone up in 2026?
The most likely underlying reason is the cost of server memory. Samsung, SK hynix and Micron have reallocated manufacturing capacity from conventional DRAM toward High Bandwidth Memory for AI accelerators, which carries far higher revenue per wafer. J.P. Morgan Global Research has estimated DRAM prices rising more than 400% from the start of 2024 to the end of 2026. Since RAM is a primary input cost in hosting, that eventually reaches plan pricing.
Is the 2026 memory shortage the same as the earlier chip shortage?
No, and the difference matters. The 2020 to 2023 shortage was largely caused by pandemic demand swings and supply chain disruption, and it resolved as those normalised. The current shortage is caused by manufacturers deliberately reallocating capacity to higher-margin AI memory. That does not self-correct as demand normalises; it corrects when new capacity arrives or the margin gap closes.
How long will the memory shortage last?
Nobody knows, and the guidance points long. New memory fabrication plants take years to build and bring to volume, which is why manufacturers and analysts cite 2027, 2028 and in some cases beyond, with SK hynix's leadership warning demand will exceed supply well into the next decade. Treat all of that as forecast rather than fact, and note that almost nobody is predicting a return to early-2025 pricing in the near term.
Will hosting prices come back down?
Analyst commentary generally describes eventual stabilisation rather than reversal. The reasonable planning assumption is that memory stays materially more expensive than it was, which makes memory efficiency a durable discipline rather than a temporary response. If prices do ease, the work you did to reduce consumption still leaves you with a faster site and a smaller bill.
Should I downgrade my hosting plan to save money?
Not without measuring first. If you drop below your actual peak memory use, the result is errors and slow responses at your busiest moments, which is the most expensive time to have them. Measure real peak consumption, then decide. Often the better route is reducing what you need through caching and database work rather than reducing what you have.
What is HBM and why does it affect my server?
High Bandwidth Memory is a specialised form of DRAM used in AI accelerators. It commands several times the revenue per silicon wafer that conventional memory does, so manufacturers have shifted capacity toward it. The conventional DRAM that goes into hosting servers, laptops and phones is what became scarcer as a result.
How can I reduce my server memory usage?
In order of impact: enable server-level full-page caching so most requests never execute your application, fix database inefficiency including unindexed queries and accumulated revisions and transients, remove rather than deactivate plugins and extensions you no longer use, and consolidate overlapping tools. All four improve performance as well as reducing consumption.
Should I lock in a longer hosting contract now?
It is a defensible hedge if you are already confident in the provider and the fit, because you are buying price certainty in a volatile input market. It is a poor decision if you are locking in with a provider you have not tested, since the switching cost during the term is effectively what you are paying for the discount.
Does this affect Canadian hosting differently?
Canadian providers face the same global component prices plus currency exposure, since server hardware is priced in US dollars. So Canadian hosting costs can move partly independently of headline component prices. What is not true is that domestic hosting is insulated from international hardware markets, and any provider implying otherwise is overstating things.
Could this affect my site's Core Web Vitals?
Indirectly, yes, and it is worth watching. If providers respond to input costs by increasing density on existing hardware, customers experience more resource contention, which degrades server response time at peak. Response time feeds Largest Contentful Paint, and Core Web Vitals are assessed at the 75th percentile of real visits, so peak-hour degradation registers even when averages look fine. Guaranteed rather than burstable allocation is the protection.
What should I ask my hosting provider right now?
Two questions. Whether your CPU and memory allocation is guaranteed or burstable, since guaranteed allocation insulates you from their density decisions. And whether the number of customer environments per physical server has changed, since increasing density is how a provider protects margin without changing the price list.
Is it a good time to move hosts?
Only if you have a specific reason beyond price. In a constrained market, unusually cheap pricing is generally funded by density, oversubscription or deferred hardware refresh, and a known-good environment is worth more than a cheaper unknown. If you do move, measure response time before and after, and handle the migration carefully to avoid losing search visibility.
Key takeaways
- This is not the 2020–2023 chip shortage. That was disruption; this is deliberate reallocation of wafer capacity to higher-margin AI memory, so it does not self-correct as demand normalises.
- P. Morgan Global Research estimates DRAM prices rising more than 400% from the start of 2024 to the end of 2026.
- IDC puts 2026 DRAM bit supply growth at around 16%, which AI demand can absorb entirely.
- TrendForce data shows increases continuing into Q3 2026 but slowing markedly from the prior quarter.
- Manufacturers and analysts point to 2027, 2028 and beyond for normalisation. Almost nobody forecasts a return to early-2025 pricing.
- Server RAM is a primary input cost in hosting, which is how a wafer allocation decision reaches your invoice.
- The risk to watch is not the price list, it is density. A provider can hold prices and pack machines more tightly, which arrives as performance variance at peak.
- Guaranteed rather than burstable resource allocation is the protection, and it is worth more now than it was.
- Caching properly is the largest available saving on most sites, because a cached request never executes your application.
- Adding RAM to mask an unindexed query converts a one-off fix into a recurring premium.
- Logged-in traffic bypasses page caching, so membership and ecommerce sites are the most exposed and object caching is their lever.
- Storage is affected too. Backup retention and unoptimised image uploads are now real costs.
- Canadian providers face the same global component prices plus currency exposure. Domestic hosting is not insulated from international hardware markets.
