Businesses Are Switching to Cheaper Chinese AI Models — How Japanese Firms in the Philippines Can Balance Cost and Security

Businesses Are Switching to Cheaper Chinese AI Models — How Japanese Firms in the Philippines Can Balance Cost and Security

Using the move by companies like DoorDash and Airbnb to test cheaper Chinese-built AI models as a starting point, this guide takes a practical look at how Japanese firms in the Philippines can cut costs while keeping customer data safe.

---

Part 1: Why This Matters

Step 1: The Philippine Business Context (3 min)

AI-powered work is billed by the volume of text processed (a unit called tokens). As that cost rises, a growing number of companies are testing cheaper AI models built in China rather than by U.S. firms — and that is the theme of the source article.

Why does this matter in the Philippines? Because the many BPO operations there (businesses that handle work outsourced by other companies) and the local sites that support a Japanese head office are highly sensitive to AI usage costs. When you run large volumes of AI for customer service, outsourced accounting, or internal help desks, the cost eats directly into profit. High interest in cheaper models is only natural.

At the same time, choosing on price alone raises a separate problem: information security. In the Philippines, the handling of personal data is overseen by the NPC (National Privacy Commission, the government body that supervises data protection). Before sending customer data to an overseas AI, a Japanese company has to make a careful judgment. Both the Japanese head office and the local site need to understand the balance between cheap and safe.

Picture yourself as a Japanese manager walking into the morning meeting at your Manila office. Your local IT lead proposes: "To lower our monthly AI bill, let's switch to a cheaper Chinese-built model." You think, "I do want to cut costs — but will customer data handling be all right?" This guide is built to support how you decide in that moment and how you explain it to your colleagues.

Step 2: Key Points from the Source Article (5 min)

Here are the facts described in the source article, organized point by point.

Source: Fortune — "Businesses are experimenting with cheaper Chinese AI models as U.S. rivals get more expensive" (July 17, 2026)

This table was compiled from publicly available facts for educational purposes. For details, please see the original article linked above.

Related: see How AI APIs Help Philippine SMEs Cut Costs and Look Beyond ChatGPT.

Step 3: Comprehension Check (5 min)

These questions check your understanding of the source article. Try to recall the answers as you read on.

Q1. What is the single biggest reason companies have started testing cheaper AI models built in China?

Hint: It's a money-related factor that grows more significant the more you use.

Q2. What two advantages did DoorDash's Andy Fang cite about Moonshot AI's model?

Hint: They are expressed in two words — one about quality, one about money.

Q3. Yasir Atalan of CSIS organized the reasons for switching into three factors. What are they?

Hint: Cost, plus two more. One of them is "the accessibility of open source."

Q4. When running an AI model in-house, what cost burden did Atalan say was involved?

Hint: An amount on the order of $30,000, needed for computing hardware such as GPUs.

Q5. What danger did Snehal Antani of Horizon3.ai warn about?

Hint: The key is where your own code and users' data get exposed.

---

Related: see How AI Integration Helps Philippine Businesses Modernize Legacy Systems.

Part 2: Putting It Into Practice

Step 4: Rollout Steps for the Philippines (10 min)

Here is how to proceed so your Philippine site doesn't stumble when considering a cheaper AI model.

Atalan in the source article also describes this as using different models for different tasks rather than replacing everything with Chinese-built AI. This "mix-and-match" mindset is the practical backbone in the Philippines, too.

Step 5: Common Mistakes and Fixes (5 min)

Here are three mistakes that tend to happen when your Philippine site takes on this theme.

Mistake 1: "Choosing a model on cost alone"

Dazzled by the low price, you switch without checking how customer data is handled. The source article, too, has experts warning of the danger of exposing data to foreign surveillance.

Bad: The local IT lead reports to head office only that "the monthly cost will be halved," gets approval, and switches.

Good: Alongside the cost comparison, you share with head office a list of exactly which data would be sent overseas. You decide only after confirming security.

Mistake 2: "Overlooking the cost of running it in-house"

To keep data from leaving, you try to run the model in-house — and overlook the required hardware cost. The source article states that investment on the order of $30,000 is needed.

Bad: You think, "It's free open source, so there's no cost," and decide to adopt it without budgeting.

Good: You estimate the purchase cost of computing hardware such as GPUs, plus electricity, in advance and build it into the budget in pesos before deciding.

Mistake 3: "Handing everything to one model"

Because the cheaper model worked well, you replace even hard-judgment tasks all at once. Quality drops and problems arise in customer service.

Bad: Without verifying quality, you switch even important customer inquiries over to the cheaper model.

Good: You first give only simple work to the cheaper model, and keep a proven model for tasks that need difficult judgment — mixing the two.

---

Part 3: Going Deeper

Step 6: Related Technical Terms (5 min)

Here are five important terms that appear in the source article, with their meaning and how they apply in the Philippines.

Open-source models are AI whose inner workings — the equivalent of a blueprint — are published, so anyone can obtain them and run them in their own environment. When a Manila BPO site wants to run AI in-house without letting customer data leave, choosing such a public model keeps the information inside the company.

Tokens are the chunks of text an AI processes, and the billing unit by which cost is set according to how much you use. If your Philippine site has AI process large volumes of inquiries every day, this token volume drives the monthly cost, so estimating it in advance is essential.

Data sovereignty is the idea of which country's laws and control govern the information you handle. When a Japanese company in the Philippines sends customer data to an overseas AI, you need to confirm which country's control that data falls under, from the standpoint of NPC data protection.

Local deployment is running AI entirely within your own computers, without relying on an outside company. When a Manila site does not want confidential information to leave, choosing this in-house operation with powerful hardware keeps the information inside the company.

AI agents are AIs that advance toward a goal on their own, without a person giving detailed instructions at each step. As with DoorDash in the source article, Philippine sites are increasingly having such AI agents handle sequences of work like order intake and internal procedures.

Step 7: Consider How It Applies to Your Company (10 min)

Bring it back to your own situation and discuss the following three themes.

Make visible where your AI costs arise

Discuss whether you have a clear picture of which tasks incur how much AI cost today.

A hint for thinking: Breaking costs down by task — customer service, outsourced accounting, internal help desk — reveals where you could shift to a cheaper model.

Draw the line between data you may send abroad and data you may not

Sort out which information may be sent to an overseas AI and which must stay in-house.

A hint for thinking: Customers' personal data and your own confidential code need careful handling from the standpoint of NPC data protection. Putting the line in writing keeps the front line from hesitating.

How to balance quality and cost by mixing models

Discuss whether a policy of using different models for different tasks — rather than handing everything to one — fits your company.

A hint for thinking: Keeping a proven model for hard-judgment tasks and shifting only simple work to a cheaper model lets you cut costs while protecting quality.

Next action: Pick one task where your company already uses AI, and write its monthly cost and the types of data it handles on a single sheet of paper. That one sheet becomes the starting point when you review your models.

---

Part 4: FAQ

Q1. Can our Philippine site just use cheaper Chinese-built AI models as-is?

They're attractive on cost, but before using them as-is, check the types of data involved. For tasks that include customers' personal data, handling must follow the Philippines' NPC (National Privacy Commission) Data Privacy Act. The safe approach is to start with tasks that don't involve personal data.

Q2. How should we explain this to get head-office approval in Japan?

Show not just the cost comparison but also a list of exactly which data would be sent overseas. Japanese head offices tend to weigh information security heavily. Presenting both cheap and safe side by side makes approval easier to obtain.

Q3. Is "in-house operation" that keeps data internal realistic in the Philippines?

It has the benefit of keeping information from leaving, but it requires investment on the order of $30,000 in powerful hardware. On top of that, electricity is relatively expensive in the Philippines, so factor that into your decision as well. Realistically, consider it only for tasks with especially high confidentiality.

Q4. How do we get local staff to understand it?

Explain clearly, in both English and Tagalog, the reason for using the new model and the scope of data that may be handled. In the Philippines, verbal agreement tends to come first. Don't stop at a verbal explanation — leave a simple written procedure so the front line doesn't hesitate.

Q5. If we switch everything to cheaper models, will costs drop the most?

Not necessarily. If quality drops and problems arise in customer service, you end up paying more in rework. The source article, too, describes the trend of using different models for different tasks. Keeping a proven model for hard tasks and shifting only simple work to a cheaper model is what ultimately balances cost and quality.

---

Tips for Getting Value (3 Tips)

First, put your costs and data on a single sheet of paper. Writing out which tasks cost how much and what data they handle on one sheet makes it obvious at a glance which tasks you can shift to a cheaper model. Because it becomes the starting point for your decisions, it's worth doing first.

Start small with tasks that don't involve customer data. Switching everything at once makes quality and security problems erupt all together. Testing with a small team on tasks that don't include personal data lets you confirm quality while keeping the impact of any failure contained.

Don't hand everything to one model — use different models for different tasks. Keep a proven model for tasks that need difficult judgment and shift only simple work to a cheaper model, so you protect quality while cutting cost. Try applying the "mix-and-match" mindset the source article shows to your own operations.

---

Bonus: How to Make the Most of PH AI Works

PH AI Works is a solutions company that supports AI and technology adoption in the Philippines. Today's theme — "switching to cheaper AI models" and "balancing information security" — is exactly the kind of area where local sites often struggle to decide.

As a next step, you can consult us on things like:

• Help making your AI costs visible by task and sorting out where you can shift to a cheaper model
• Drawing the line on which data may be sent to an overseas AI, and building an operating policy that reflects Philippine data-protection considerations
• Designing an approach for mixing models by task, and preparing explanatory materials for local staff

Feel free to reach out. Consultations are free.

---

References & Sources

• Fortune — "Businesses are experimenting with cheaper Chinese AI models as U.S. rivals get more expensive" (July 17, 2026)