Cloud Prices Have Dropped. AI Costs Have Increased. Do You Understand What This Means?

Microsoft has reduced the prices of Azure Virtual Desktop (cloud desktop) by up to 20%. At the same time, the prices of AI-related services, such as Microsoft 365 Copilot, have either remained stable or effectively increased.

Reading this as simply “the cloud has become cheaper” can lead to misguided judgments.

It is essential to look at what has become cheaper and what has become more expensive at the same time. This perspective is what small business owners need right now.

What Has Become Cheaper and What Has Become More Expensive?

First, let’s clarify the facts.

Cheaper: Cloud Desktop (Virtual PC Environment)

• The usage fee for Azure Virtual Desktop has been reduced by approximately 20%.
• From about $100 per employee per month to about $80.
• A service that provides a cloud-based PC environment for remote work.
• This area is experiencing intense price competition with AWS (Amazon WorkSpaces) and Google Cloud.

More Expensive: AI-Related Computing Resources

• The demand for GPU instances (servers necessary for AI processing) significantly exceeds supply.
• NVIDIA’s H100 chip costs about 4 million yen each, and hundreds of these are bundled together to run AI models.
• Microsoft’s investment in AI infrastructure has ballooned to around $50 billion annually (approximately 7.5 trillion yen).
• Naturally, these costs will be passed on to service prices.

In other words, the “cost of using a PC for people” has decreased, while the “cost of using a PC for AI” has increased. Without understanding this structure, it is easy to misprioritize IT investments.

Why Is This “Twisting” Happening?

The reason is simple.

Cloud desktops have become commoditized. With intense price competition among AWS, Google, and Microsoft, differentiation has become difficult. Therefore, companies are lowering prices to retain customers. This has entered a phase of “low-margin, high-volume” sales.

On the other hand, AI infrastructure is in short supply. Companies worldwide are competing for GPUs. Microsoft has invested over $13 billion in OpenAI to build its own AI service infrastructure, but the return on that investment is yet to come. There is no reason for AI service prices to drop; in fact, they are likely to rise.

Another crucial factor is the cost of electricity. AI inference processing consumes more than ten times the power of standard cloud processing. The power consumption of data centers is surging, and in the U.S., there are even cases where new data centers are being built alongside natural gas power plants. According to the IEA, global power consumption by data centers could double by 2026.

This rise in electricity costs will undoubtedly be reflected in AI service prices. Those who are optimistic that “AI will become cheaper” are overlooking this cost structure.

What Changes for Small Businesses?

Now, let’s consider a specific calculation for a local small business with 30 employees.

Scenario 1: Using Only Cloud Desktops

• 30 employees × $80 per month = $36,000 per month
• Before the price drop, it was $45,000 per month, resulting in annual savings of $108,000.
• Considering the costs of managing on-premises PCs (server maintenance, Windows updates, handling failures), this is more than sufficient to cover expenses.

Scenario 2: Cloud Desktop + Basic AI Usage

• Cloud Desktop: $36,000 per month
• Implementing Microsoft 365 Copilot ($30 per person per month = about $4,500) for 10 people: $4,500 per month
• Total: $40,500 per month
• Automating meeting minutes, drafting emails, and Excel analysis with Copilot could save about 3 hours per week per person, totaling 120 hours per month for 10 people. At an hourly wage of $20, this equates to $24,000 in productivity gains per month.

Scenario 3: Fully Utilizing AI

• Cloud Desktop: $36,000 per month
• Company-wide implementation of Copilot (30 people): $13,500 per month
• Operating a dedicated chatbot using Azure OpenAI Service: $5,000 to $15,000 per month
• Total: $54,500 to $64,500 per month
• However, automating inquiry responses could potentially save the equivalent of one employee’s salary (about $25,000 to $30,000 per month).

The key point here is that the savings from Scenario 1 can be directly allocated to AI investments in Scenarios 2 and 3. By applying the annual savings of $108,000 from the cloud price reduction to the annual cost of Copilot for 10 people ($54,000), AI implementation can begin with zero additional budget.

This is the practical significance of looking at what has become cheaper and what has become more expensive at the same time.

Is “AI Expensive” True?—The Comparison Should Be Labor Costs

I want to shift the perspective once more.

When hearing that “AI services are increasing in price,” it can be alarming, but it is crucial not to misidentify the comparison target. The competition for AI is not with other AI services, but with human labor costs.

Consider this example:

• Data entry for invoices: A part-time worker spends 40 hours a month on this → Hourly wage of $12 × 40 hours = $4,800 per month
• Automating the same process with Azure AI Document Intelligence → $1,000 to $2,000 per month
• Savings: $2,800 to $3,800 per month, or $33,000 to $45,000 annually

Even if AI costs rise slightly, the gap compared to labor costs remains significant. The issue is not whether “AI is expensive,” but rather “what it is being compared to.”

However, there is a caveat. AI prices will continue to fluctuate. Current uses that barely break even could become unprofitable with a single price increase. It is essential to simulate whether “AI costs can still be recouped even if they increase by 50%” at the time of implementation.

Three Things Small Businesses Should Do Now

1. Reallocate Savings from Cloud Price Reductions to “AI Experiment Budgets”

It would be a waste to simply record the savings from price reductions as profit. Just a few thousand dollars per month is sufficient. Allocate this to a “test budget” to try out Copilot or ChatGPT Team with a small group. If successful, it can be expanded; if not, it can be stopped. A $500 experiment won’t be a fatal blow even if it fails.

2. Understand the “Cost Structure of AI”

GPU prices, electricity costs, and model sizes determine the cost structure of AI services. Currently, GPU shortages keep prices high, but after late 2025, as competing chips from AMD and Intel become available, price competition may begin. Now is the phase to “start small even if it is expensive.” It is better to make larger bets once prices stabilize.

3. Prioritize What to Automate

It is not necessary to automate everything. Start by selecting from the three categories: “repetitive tasks,” “tasks that are dependent on specific individuals,” and “tasks prone to errors.” Meeting minutes, invoice processing, and inquiry responses are three areas where AI can quickly show effectiveness across industries.

Conclusion: The “Twisting” of Prices Is an Opportunity for Small Businesses

The cloud has become cheaper, and AI has become more expensive. Within this seemingly contradictory movement lies a rational strategy for small businesses.

Use the savings from what has become cheaper to experiment with what has become more expensive.

That’s all there is to it.

Large corporations make AI investments in the hundreds of millions all at once. Small businesses do not need to mimic that. Conduct a $500 experiment for three months, and keep only what proves effective. This cycle of “testing small and deciding quickly” is the greatest weapon for small businesses.

Simply staring at Microsoft’s price list won’t change anything. Compare the “things that have become cheaper” with the “things that have become more expensive,” and apply them to your own numbers. The view that emerges from this will likely be entirely different from the headlines in the news.