AI Costs in SaaS: What Samsung Has to Do With It

Your customer service tool went up 30% at renewal. The reason in the email was "new artificial intelligence features". You never asked for those features. But you're paying for them. Today I want to explain where the AI costs in your SaaS come from. And why a factory in South Korea has more to do with your bill than it seems.

What Samsung announced, in plain English

News came out that Samsung plans to more than double production of HBM4 and HBM4E next year. The acronym sounds scary, but the idea is simple.

HBM is a type of very fast memory. It sits right next to the chips that run AI models, like the Nvidia cards in data centers. Without this memory, the chip sits around waiting for data. It's like a Formula 1 engine fed through a drinking straw.

Over the last two years, this memory was one of AI's bottlenecks. Few companies make it, demand exploded, and whoever could buy it paid a lot. That even spilled over into regular computer and phone memory, which also got more expensive.

When the world's largest memory maker says it will double production, the message is clear: the shortage is likely to ease.

Why this shows up on your software bill

You don't buy chips. You subscribe to a CRM, a chatbot, an email marketing tool. So what's the connection?

It's a chain with four links:

• Samsung sells memory to the companies that make AI chips.
• Those chips go into data centers run by Amazon, Google, Microsoft and the like.
• Companies like OpenAI and Anthropic rent that computing power and charge for model usage.
• The SaaS you subscribe to pays for those models and passes the cost on to you, with a margin.

Every time the tool summarizes a ticket, suggests a reply or writes an email, someone down the line is paying for processing. And AI processing got expensive partly because there wasn't enough hardware.

That's why so many software companies pushed "AI-powered" plans in recent months. Part of it is real cost. Part of it is opportunity: the topic is hot, and customers accept price increases more easily.

Will more chip supply lower AI costs in SaaS?

It will lower costs for the companies that build the software. Whether it lowers costs for you is another story.

The per-use price of AI models has already dropped a lot since 2023. Tasks that cost a few dollars now cost cents. With more memory on the market starting in 2027, the trend is for prices to keep falling.

Now tell me: how many times have you seen a SaaS lower its price on its own?

Vendor costs drop quietly. Subscription prices only go up with an announcement.

This is my most direct opinion here: lower infrastructure costs don't reach your invoice automatically. They reach you when you ask, compare and threaten to switch. If you don't negotiate, you'll keep paying 2025 prices on 2027 costs.

There's one detail working in your favor. New competitors enter the market exactly when infrastructure gets cheaper. More chip supply means more startups offering the same thing for less. That boosts your bargaining power.

What I'd do in your shoes before the next renewal

You don't need to understand semiconductors to act. You need a spreadsheet and half an hour.

1. List what you pay for "AI". Open the invoices for your tools. See which plans went up because of artificial intelligence features. For many of the clients I work with, this adds up to 15% to 25% of their software spend.

2. Find out what actually gets used. Most tools have a usage dashboard. If your CRM's "AI assistant" was used 12 times last month by a team of eight, you're paying for a decorative feature.

3. Avoid long contracts at today's price. An annual plan with a discount looks great. But locking in for two or three years a price based on scarce hardware is betting against the trend. If you do sign an annual plan, ask for a review clause.

4. Ask for a plan without AI. It sounds obvious, but few people ask. Many vendors have one, they just don't show it on the pricing page. The worst that can happen is they say "no".

5. Ask which model they use. This isn't technical curiosity. If the tool uses an expensive model for simple tasks, like sorting an email, you're paying for a Formula 1 car to go to the bakery. A serious vendor knows how to answer.

What if you're using AI directly yourself?

Some companies have moved past just subscribing to tools. They use a model's API to automate customer service, generate proposals or read invoices. Here the Samsung news matters even more.

A real example, with rounded numbers. A client of mine, a distributor with about 40 employees, spent around R$ 1,800 a month on an off-the-shelf tool for sorting orders via WhatsApp. We built an automation that calls an AI model directly through the API, only when it's needed. Processing costs dropped below R$ 200 a month. Including maintenance, the savings topped 70%.

This only works well if the automation is built without tying you to one vendor. Models change prices all the time. The best value in March might be in third place by September. If everything is built around a single vendor, you can't switch without redoing the project.

The rule I follow: the automation must let you swap the AI model like you'd swap coffee brands. Change one setting and you're done. When prices drop, and they will, you benefit the same day.

What doesn't change, cheap chips or not

Let's keep our feet on the ground. Doubling production isn't flipping a switch. Semiconductor factories take time to scale, and demand for AI keeps growing. The surplus may take a while to show up. Big tech may buy everything first, as it always does.

So don't make decisions counting on low prices tomorrow. Make decisions that leave you free to take advantage when they arrive.

Some things hold true in any scenario:

• Paying for features nobody uses is waste, with expensive chips or cheap ones.
• Good automation starts with the process, not the trendy tool.
• A flexible contract is worth more than a loyalty discount.

In the end, the Samsung news isn't about Samsung. It's about a cost that's shifting under your feet. Those who pay attention now will reach 2027 paying less for the same thing. Those who ignore it will keep funding the vendor's margin, and smiling at the renewal meeting.

If you want to review your tools calmly and find where you're paying for AI you don't need, or where a custom automation would cost less, tell me how your operation works.

Your tool went up 30% at renewal "because of AI". And you never asked for those features.

The cost of running AI is falling. Samsung is set to more than double production of the memory that powers these chips.

But vendor costs drop quietly. Subscription prices only go up with an announcement.

A client of mine was paying R$ 1,800 a month for an off-the-shelf tool. With a custom automation that calls the model only when needed, they now spend less than R$ 200.

Before your next renewal: check what actually gets used, ask for a plan without AI, and avoid long contracts locked at today's price.

I wrote about this on the blog. If you want to review your tools with me, reach out.

#ArtificialIntelligence #SaaS #Automation #Costs #SmallBusiness