China's Cheaper AI Could Change What Australian Businesses Pay

The price gap is no longer small

Chinese AI companies have spent 2026 closing the capability gap with the leading US labs while opening an enormous gap on price.

Moonshot AI released Kimi K3, currently the largest open source model available, on 16 July. Z.ai's GLM-5.2 model, released in June, has been benchmarked at roughly 1.92 US dollars per million output tokens, against Anthropic's Opus 4.8 at around 25 dollars for the same unit, a saving of more than 90 per cent. DeepSeek has gone further again, cutting token prices by 75 per cent and claiming its agents now cost as little as one thirty fourth of comparable OpenAI or Anthropic setups for certain agentic tasks.

That is not a rounding difference. It is a fundamentally different cost structure for running AI at scale.

Australian businesses are already moving

This is not a hypothetical for Australian companies watching from a distance. Sydney based Relevance AI, which builds autonomous AI agents, told reporters that traffic to open weight models has shifted from around 5 to 7.5 per cent at the start of the year to between 20 and 25 per cent today. Amplify AI, a Sydney advisory firm, said some clients are consuming 20,000 dollars worth of tokens a day, and that switching to hosted open weight models can cut that bill by close to 80 per cent.

Their client base spans real, recognisable names, including Canva, KPMG and Autodesk, alongside healthcare and finance firms. This is happening inside serious, established businesses, not just startups chasing the cheapest option available.

IG Markets analyst Tony Sycamore summed up the shift bluntly: "Chinese AI, Chinese infrastructure, are as good as some of the US frontier models." Azeem Azhar of Exponential View made the economic case just as plainly: "At the scale agents operate, even small cost differences compound into meaningful budget gaps."

The catch that gets less airtime

Price is not the only consideration, and it would be dishonest to write this story without the other side.

Kendra Schaefer of Trivium China raised a genuine concern that goes beyond commercial preference: "a locally installed model is a vector for compromise." Running a Chinese developed model inside sensitive business infrastructure is a different risk profile to using a US hosted API, particularly for businesses handling customer data, financial information or anything with regulatory sensitivity.

This is not a reason to dismiss the cost argument. It is a reason to treat the decision as more than a simple price comparison.

Why this matters for Australian business owners specifically

Australia is a price taker in this market, not a price setter. Nothing here is decided in Canberra or Sydney. But the consequences land directly on Australian balance sheets, and three of them matter now.

The AI price war is a genuine opportunity, not just a curiosity. If your business is paying for AI capability today, whether directly through an API or indirectly through a vendor's pricing, the gap between frontier US pricing and competitive Chinese pricing is now large enough to be worth actively questioning, not simply accepting as the going rate.

Cheaper does not automatically mean the right choice for your business. Data residency, security posture and what a vendor can tell you about where information goes are real questions, not formalities. A business handling customer or financial data has more at stake here than one generating marketing copy.

Expect your AI vendors to shift underneath you. Vendors chasing margin will feel this cost pressure directly, and the smart ones will already be evaluating cheaper model options behind the scenes. Ask what model actually powers the tool you are paying for, and whether that has changed recently, or is likely to.

---

Sources

• China's Moonshot, Z.AI and DeepSeek are challenging U.S. AI labs and beating them on cost, Fortune, 26 Jul 2026
• Chinese models are on track to win the agentic AI price war, The Strategist (ASPI), 21 Jul 2026
• Chinese AI Models Drive Adoption Among Australian Businesses, 31 Jul 2026
• China's Moonshot AI unveils Kimi K3 that rivals OpenAI, Anthropic, CNBC, 17 Jul 2026