Cloud Costs Will Rise Next Year and the Year After. Do You Know Why?

To put it simply, about 40% of data center construction projects in the U.S. will not be completed on schedule. The shortage of RAM will not be resolved until 2027. The combination of these two factors means that the costs of cloud services used by small and medium-sized enterprises (SMEs) are likely to continue to rise structurally over the next 2 to 3 years.

This is not just a temporary price increase; it is a story of infrastructure-level congestion where “supply will be insufficient for several years.”

—

What’s Happening: The Contradiction of a Construction Boom Yet ‘Insufficient’

Currently, hundreds of billions of dollars are being poured into data center construction in the U.S., driven by the AI boom. Major companies like Microsoft, Oracle, and OpenAI are launching large-scale projects one after another. At first glance, one might think, “Supply should increase” based on the investment amounts.

However, the reality is different.

According to the latest research using satellite imagery, many projects are expected to be delayed by more than three months from their planned completion dates. There are three main reasons for this.

1. Labor Shortage — There is a severe shortage of skilled technicians needed for data center construction. Electrical work, cooling systems, and high-voltage wiring all require advanced expertise, and with a simultaneous construction boom across the country, there is fierce competition for workers.
2. Delays in Power Supply — AI data centers consume vastly more power compared to traditional ones. To operate and cool tens of thousands of GPUs, new substations and enhancements to the power grid are necessary, but developing power infrastructure takes longer than constructing buildings. In some cases, permits alone can take 1 to 2 years.
3. Shortage of Equipment and Materials — The lead times for critical equipment such as transformers, cooling systems, and backup generators have lengthened. Demand significantly exceeds supply, and delays of six months to a year have become the norm.

Investment is increasing. However, the number of completed and operational data centers is not growing as planned. “There is money, but no buildings are being constructed”—this is what is happening now.

—

RAM Shortage: Another Bottleneck That Will Persist Until 2027

It’s not just the “boxes” of data centers that are delayed; the contents are also lacking.

The most serious issue is the shortage of HBM (High Bandwidth Memory), a type of high-performance memory. GPUs used for AI training and inference (such as NVIDIA’s H100 and H200) require a large amount of HBM. Currently, only three companies in the world—Samsung, SK Hynix, and Micron—can mass-produce HBM.

All three companies are working to increase production. However, building new manufacturing lines to boost semiconductor memory production takes at least 18 to 24 months from capital investment to mass production. Industry forecasts suggest that it will be at least 2027 before HBM supply catches up with demand.

The impact of this memory shortage is already reflected in prices. A clear example is Meta’s Quest VR headset. Due to rising costs of RAM and other components, a price increase of $50 to $100 (approximately 12% to 20%) has been announced. Even consumer products are feeling this impact.

The effects on enterprise servers and cloud infrastructure are, of course, even greater.

—

Impact on Small Businesses: The Most Dangerous Mindset is ‘It Doesn’t Concern Us’

“We don’t own a data center and we’re not using AI”—if there are small business owners thinking this way, this is the most crucial point.

The cloud services your company uses operate on these data centers.

AWS, Azure, Google Cloud, or SaaS built on top of them—accounting software, customer management, email distribution, online storage—all consume resources from data centers.

As supply becomes constrained, major tech companies will prioritize resources for AI workloads because they have higher profit margins. What happens then? Resources for general cloud services will relatively ‘take a hit.’

Let’s consider this structure numerically.

• Assume a small business is currently paying 50,000 yen per month for cloud services.
• If cloud providers implement annual price increases of 5% to 15%, the monthly cost would rise to between 52,500 yen and 57,500 yen.
• If compounded over three years, the monthly cost of 50,000 yen could reach the 60,000 to 70,000 yen range.
• Annually, this translates to an increase in costs of 120,000 to 240,000 yen.

For a single company, this might seem manageable. However, if multiple cloud services are used, the impact multiplies. If five SaaS applications all increase by 5% to 10%, annual IT expenses could balloon by several hundred thousand yen. For small businesses with thin profit margins, this is a number that cannot be ignored.

—

So, What Should We Do?

I won’t write an article that just ends with “It’s tough.” Here are four actions small businesses can take now.

1. Immediately Take Stock of Your Cloud ‘Usage’

The first step is to assess the current situation. Open the management screens of AWS or Azure and check the gap between the resources you are actually using and those you are contracted for. Surprisingly many small businesses are paying monthly for instances and storage that they are not using.

Specifically, if there are instances with an average CPU usage of less than 20% over the past three months, consider downsizing or stopping them. This alone can often reduce monthly costs by 10% to 30%.

2. Consider Long-Term Contracts (Reserved Instances)

AWS’s Reserved Instances or Azure’s Reserved VM Instances offer 30% to 60% discounts off on-demand rates when you commit to using them for one or three years.

Now that future price increases are on the horizon, it makes sense to make the decision to “enter into a long-term contract at current price levels.” Of course, accurate forecasting of usage is required, but if there are workloads that are used consistently, it is worth considering.

3. Seriously Consider Returning to Local Processing

The era of putting everything in the cloud is reaching a turning point from a cost perspective.

For example, for internal file sharing or backups, by introducing a single NAS (Network Attached Storage), you can significantly reduce monthly cloud storage costs with an initial investment of around 50,000 to 150,000 yen. If you use cloud storage at 3,000 yen per month for five years, that totals 180,000 yen. A NAS can cover equivalent or greater capacity with just the initial investment.

The same goes for AI inference processing. Recent local LLMs (like Llama 3, Phi-3, etc.) can run practically on PCs equipped with GPUs costing around 100,000 to 200,000 yen. Compared to cloud AI that charges per API call, there is a turning point where companies with high monthly processing volumes find local processing to be cheaper.

4. Maintain Negotiating Power with Multi-Cloud and Multi-Vendor Strategies

If you rely on a single cloud provider, you have no escape if they raise prices. Using multiple providers and maintaining a state where you can switch at any time becomes your greatest weapon in price negotiations.

Can you say, “If your company raises prices, I will move this part to another service”? Just being able to say this can change how sales representatives respond.

—

The Essence: How Will Small Businesses Compete in an Era of Rising Costs?

For the past decade, cloud prices have generally continued to decline. Many small businesses have made IT investment decisions based on the premise that “as long as it’s in the cloud, it’s cheap.”

That premise is beginning to crumble.

The AI bubble is consuming data center resources, memory shortages are driving up prices, and construction delays are slowing the recovery of supply. These three issues are occurring simultaneously, and all are structural problems that will not be resolved in the short term.

However, there is no need for pessimism. The timing of changes in cost structures presents an opportunity for small businesses.

Large corporations cannot easily reassess their massive cloud contracts. The larger the organization, the longer decision-making takes. In contrast, small businesses can decide with a single judgment from the owner to “stop this service next month” or “switch to local.”

This speed is the greatest weapon of small businesses.

“Cloud costs are expected to rise”—with this information in mind, what will you do by next month? That will determine your cost structure three years from now.

If you’re going to act, now is the time.