China’s cheap AI tokens are a double-edged sword for Asian companies

AI For Business


Background steps such as prompting, validation, reflection, running code, and using other external software tools all consume more tokens, he said.

based on Anthropic estimates that the average AI token cost for enterprise software developers using Claude Code was USD 13 per day, and the monthly cost per developer was approximately USD 150 to USD 250. said Business Insider Report for April.

For a large technology company employing 500 developers, the cost of an AI token would be approximately $75,000 to $125,000 per month, or $900,000 to $1.5 million per year, without discounts or corporate deals.

Most AI providers offer discounts for business customers.

However, these are typically negotiated privately based on usage commitments, contract length, model used, required customer support, and whether cloud services are included.

According to media reports, OpenAI offered some enterprise customers discounts of 10% to 20% on multi-year or bundled contracts.

Verma added that Asia “could be the first region where AI becomes a true mass market on an industrial scale” as countries such as India, Indonesia, Malaysia and the Philippines “have huge service economies, large developer pools and are very price sensitive.”

For the cost of Chinese AI tokens make Because AI is cheaper to run, Asian companies can incorporate it into “call centers, field service, education, logistics and finance operations faster than if they had to pay premium prices in the West,” he said.

However, Wong said companies in India and Southeast Asia should not judge AI costs solely by the sticker price per million tokens.

An AI model trained to efficiently process Chinese or English can be expensive when processing languages ​​such as Tamil, Indonesian, or Vietnamese because it uses more tokens.

Wong said a model that looks 50% cheaper could end up being more expensive in production if it performs poorly in a company’s local operating language, requires repeated trials, or requires more human review.

A more useful metric is the “cost per successful result,” or the total cost of obtaining an accurate, usable result, he added.

Who will win the AI ​​race for business adoption?

China has emerged as a strong challenger to the US in the AI ​​race due to its low token costs, experts told CNA.

The Financial Times reports that the cost advantage of Chinese AI companies comes from cheaper energy and more efficient models, including specialist mixed architectures.

Mixture of Experts (MoE) is an AI architecture popularized last year by DeepSeek’s R1 model. Use multiple specialized submodels within one AI model, but activate only the most relevant submodel for each prompt. This reduces computing costs.

Think of the MoE model as a team of experts, such as doctors, lawyers, and engineers, with only the most relevant experts responding and others doing nothing.



Source link