Cloud Services

3 Hidden Cost Drivers Behind an Overcharged Mid-Market

Summary

• Three major forces are driving up technology costs: rising AI consumption, memory shortages and Broadcom’s VMware licensing changes.
• AI costs are becoming harder to predict as agentic AI consumes significantly more tokens and pricing models vary across providers.
• Server memory prices have surged as manufacturers prioritise high-margin AI memory, reducing the supply available for traditional enterprise infrastructure.
• VMware licensing changes are increasing costs, with subscription-based bundles, higher core minimums and potentially significant renewal increases.
• Mid-market organisations feel these increases more heavily because they typically have less procurement leverage and fewer specialist resources than large enterprises.
• A fairer total cost of ownership requires greater transparency, proactive pricing reviews and choosing platforms based on workload requirements rather than vendor incentives.
• Businesses should review costs before upcoming renewals or refreshes, including AI consumption, current memory pricing, licensing utilisation and switching costs.
• The key takeaway: organisations should assess the full cost of their technology environment—not just headline pricing—to identify unnecessary costs and make better infrastructure decisions.

Your infrastructure budget didn’t blow out because you made a bad decision. It blew out because three forces collided at once, and the scale caught almost everyone off guard.

AI is now billed by consumption, not by seat. Memory chip prices have doubled inside twelve months. And Broadcom has redrawn the VMware rulebook entirely. None of these forces started with you, yet you’re the one explaining the variance to your CFO.

This is what an overcharged market looks like from the inside. Not one bad supplier. A structural shift that lands hardest on organisations without the scale to push back.

Here’s what’s actually driving the increase, why the mid-market absorbs more of it than anyone else, and what a fairer model looks like.

What's Actually Driving Up Your Technology Costs

Three separate shocks are hitting infrastructure budgets simultaneously. Each one alone would be manageable. Together, they compound.

AI Token Costs Are the New Variable Nobody Budgeted For

AI is increasingly billed by the token: the small units a model uses to read and generate text. Analysis of enterprise usage shows the underlying price per million tokens has actually fallen sharply over the past year. Yet enterprise AI bills keep rising.

The gap isn’t pricing. It’s volume. Agentic AI systems that reason, use tools and iterate through multiple steps consume tokens far faster than a simple chatbot ever did. A single interaction that cost a few cents in 2023 can now cost more than a dollar once orchestration and tool use are involved.

Governance compounds the problem. AI pricing spans tokens, credits and vendor-specific units that make providers genuinely difficult to compare. In Australia, commentators have pointed out that organisations often can’t say whether they’re paying US$2 or US$18 per million tokens for comparable capability.

For mid-market CIOs already stretched thin, this is a new cost category arriving with no internal playbook to manage it.

The Memory Shortage That Outpaced Every Forecast

While AI reshapes how software gets billed, a second and largely invisible shock has been building in hardware. It has nothing to do with your business and everything to do with your next invoice.

Why Memory Prices Have Spiked

The driver is AI demand, but not in the way most IT teams expect. Building AI accelerators requires high-bandwidth memory, a specialised chip category that consumes roughly three times the wafer capacity of standard server memory to produce.

Samsung, SK Hynix and Micron control the vast majority of global memory production. Faced with far higher margins on AI-grade memory, all three have redirected manufacturing lines away from the standard DDR5 modules that fill everyday servers. Supply for ordinary infrastructure has shrunk as a direct result.

Server-grade DDR5 has roughly doubled in price through 2026. Some enterprise modules have moved considerably further than that, and DRAM now represents up to a quarter of a typical server’s total cost, more on hosts running 256GB or 512GB of RAM.

The Supply Chain Bottleneck Behind the Numbers

This isn’t a simple demand spike that a factory can quickly absorb. It’s a structural reallocation, and three separate pressures are reinforcing it at once.

First, capacity itself is being redirected. Industry estimates put 2026 memory production allocation at roughly 70% toward AI data centres, leaving the remaining 30% for every other use combined, from laptops to enterprise servers.

Second, hyperscalers moved first. Large cloud providers pulled forward enormous DRAM orders in late 2025 to protect their own AI buildouts. That front-loading pulled inventory out of the market before mid-market buyers had a chance to lock in supply.

Third, new capacity is years away. Building a memory fabrication plant is capital-intensive and slow. The new capacity announced by manufacturers won’t reach meaningful production until late 2027 or 2028 at the earliest, regardless of how fast demand grows in the meantime.

A tariff layer has added further friction. Section 232 tariffs on advanced computing chips, introduced in January 2026, have added another cost layer to imported hardware for Australian and global buyers alike.

What This Means for Future Pricing

The rate of increase is starting to slow. Recent forecasts show DRAM contract prices rising in the low double digits quarter-on-quarter through Q3 2026, a marked cooling from the 90%+ jumps seen earlier in the year.

Slower growth is not the same as falling prices. Major manufacturers have signalled that 2026 capital spending will prioritise AI memory and process upgrades over expanding commodity DRAM output. Several have indicated they expect elevated pricing to persist well beyond 2028.

For mid-market IT teams, the practical takeaway is this: a refresh budgeted eighteen months ago no longer reflects reality, and waiting for prices to fall back to pre-2025 levels is not currently a credible plan.

Broadcom's VMware Overhaul Is Rewriting the Rules

The third shock sits inside the software layer, and it has moved just as fast as the hardware market above it.

Since acquiring VMware, Broadcom has ended perpetual licensing entirely. Every customer now sits on a subscription, bundled into one of four packages whether they need the full feature set or not.

The core-count math has changed too. From April 2025, Broadcom introduced a 72-core minimum per CPU, regardless of how many cores your server actually uses. A modest five-host cluster that once cost an Australian business $5,000 to $15,000 a year can now quote at $30,000 to $60,000 or more.

Industry reporting has tracked renewal increases ranging from 150% to over 1,000%, with smaller deployments hit hardest. Some organisations have taken legal action over the disruption this has caused to their operations. A late renewal now attracts a 20% surcharge on top.

Why the Mid-Market Feels This Hardest

Enterprise customers have procurement teams built to absorb shocks like these. They negotiate, escalate and force supplier accountability through sheer commercial weight.

You don’t have that luxury. A small IT team manages a technology estate that carries enterprise-level consequences without enterprise-level leverage. There’s no dedicated licensing specialist tracking core-count changes. No one is benchmarking DRAM pricing against last quarter, on top of everything else on your plate.

So the increase simply lands. It arrives as a renewal quote, a refresh budget or a support bill, and you’re the one explaining the variance, whether that’s to a CFO, a board or an auditor asking why the number moved.

This is precisely the injustice that overcharging describes. It isn’t only about unit price. It’s about who has the resources to challenge a bill and who doesn’t. You consistently sit on the wrong side of that line, absorbing cost shifts that larger buyers simply negotiate away.

What a Fair Total Cost of Ownership Actually Looks Like

A different kind of provider doesn’t just quote lower. It changes what you can see and what you can control.

That starts with billing transparency. You should be able to trace every dollar on your invoice back to something you actually consumed, not a bundled feature you never asked for. Proactive re-rating matters too. A provider that reviews your commercial structure before you have to ask is doing part of the job your stretched team doesn’t have time for, freeing your people for work that actually moves the business forward.

Platform choice should follow the workload, not the vendor’s margin. Private cloud, public cloud, SaaS and colocation each suit different applications. The right answer is rarely “move everything,” and a credible provider will say so even when it costs them the sale.

Here’s what this looks like in practice at Macquarie Cloud Services:

Managed Azure customers save an average of 26%, with migration included and no recurring professional services friction for routine changes.
Launch® Private Cloud runs approximately 30% below equivalent public cloud pricing, with predictable charges instead of consumption surprises.
Migration is built into the commercial model, not billed as a separate project every time your environment needs to move.[ND1.1]

None of this depends on cutting corners. It depends on refusing to let opacity, inertia and bundled complexity become the business model.

Questions to Ask Before Your Next Renewal

If you’re facing an AI rollout, a hardware refresh or a Broadcom renewal in the next twelve months, these questions will tell you more than any vendor pitch.

Who is tracking my AI token consumption, and against what baseline? If the answer is nobody, that’s the first gap to close before the next agentic project scales.

Does my memory budget reflect current pricing, or a quote from eighteen months ago? DRAM costs have moved fast enough that stale assumptions can undermine an entire refresh business case.

What am I actually licensed to use, versus what am I paying for? Bundled licensing models mean many mid-market environments are over-provisioned for capability nobody uses.

Can my current provider explain my bill in one conversation? If it takes multiple calls and a spreadsheet to understand what you’re paying for, the opacity itself is a cost.

What would moving away actually cost me? If the honest answer is “more than staying,” ask why switching has been made that difficult.

The Bottom Line

AI token economics, the memory shortage and Broadcom’s licensing changes are three separate stories. But they share a pattern that mid-market organisations know well: costs shift upward, and the customers least equipped to challenge them absorb the difference.

Being overcharged isn’t only about the number on the invoice. It’s about opacity, inertia and platforms built around what the vendor wants to sell rather than what your workload actually needs.

A fairer model starts with total cost of ownership, not unit price. It starts with a provider willing to show you the whole bill, re-rate proactively and put the right workload on the right platform, even when that means recommending less.

If your next renewal, refresh or AI rollout has you wondering whether you’re paying for capability or paying for someone else’s margin, that’s a conversation worth having early, before you’re the one presenting the variance upward.

Get in touch with the Macquarie Cloud Services team to talk through your total cost of ownership.

Josh Dominguez

Josh holds a Bachelor of Computer Science from Western Sydney University. He began his technical career at Qantas, specializing in cloud computing. Later, he joined Macquarie Cloud Services, starting in the Hosting Management Centre (HMC) for two years before becoming a Senior SOC Analyst in their Cyber Security Practice.