Enterprise AI inference costs hit 2026 low amid global price war, DeepSeek surge
Synopsis
Enterprise AI inference costs have crashed to a 2026 low of just US$1.16 per million tokens — less than half the May peak — as OpenAI's 80% price slash and a surge in Chinese open-source adoption from firms like DeepSeek rewrite the economics of AI deployment for businesses worldwide.
Key Takeaways
AI inference prices fell to a 2026 low of US$1.16–US$1.18 per million tokens during August 6–8 , per Jefferies citing Silicon Data .
Costs have dropped more than 43% since May 31 , when the average stood at US$2.04 per million tokens .
OpenAI cut rates for its GPT-5.6 model series by up to 80% last month, escalating the global AI price war.
Anthropic 's Claude Opus 5 reportedly matches its flagship Fable 5 model's performance at half the price.
Chinese open-source firms — including DeepSeek , Zhipu AI , Moonshot AI , and MiniMax — are driving low-cost alternatives that pressure Western API providers.
Jefferies analysts led by Thomas Chong attribute the decline to an 'increasing emphasis on cost efficiencies' across both US and Chinese tech ecosystems.
Enterprise AI inference costs have dropped to their lowest point of 2026, driven by an intensifying global price war and rapid adoption of low-cost Chinese open-source models, according to research by investment bank Jefferies. The findings, citing data from US-based research firm Silicon Data, signal a structural shift in how businesses budget for AI workloads.
Prices at a record low for 2026
Average inference prices — measured per million tokens, or chunks of data processed by a model — ranged between US$1.16 and US$1.18 from August 6 to 8, marking the lowest level recorded this year, Jefferies said on Monday. The decline is steep: unit costs stood at US$2.04 on May 31 and had already fallen to US$1.45 in late July before dropping further. Silicon Data's index tracks pricing across business API providers and open-weight inference platforms used by software developers.
Why it matters: the price war accelerates
The price decline coincided with an 'increasing emphasis on cost efficiencies' across both the US and Chinese tech ecosystems, Jefferies analysts led by Thomas Chong noted. OpenAI escalated the price war last month by slashing rates for its latest GPT-5.6 model series by up to 80 per cent. Meanwhile, Anthropic's Claude Opus 5 reportedly delivered performance comparable to its flagship Fable 5 model at half the price, according to the Jefferies report.
The competitive backdrop: Chinese open-source pushes costs lower
On the open-source front, Chinese firms are aggressively expanding the boundaries of affordable computing. DeepSeek, alongside peers such as Zhipu AI, Moonshot AI, and MiniMax, has contributed to a wave of low-cost, high-performance models that enterprises can deploy without paying premium API rates. Platforms tracked by Silicon Data — including OpenRouter — aggregate these options, giving developers direct price visibility and intensifying competitive pressure on incumbent providers like Google and Tencent Holdings.
What's next: further compression likely
The convergence of Western price cuts and Chinese open-source momentum suggests inference costs may compress further through the second half of 2026. Enterprises that locked in higher-cost AI contracts earlier this year face the most immediate exposure, while cloud-native developers stand to benefit from dramatically lower build costs. Analysts and industry observers will be watching whether OpenAI, Anthropic, and major cloud hyperscalers respond with another round of cuts — or shift competition to capability and reliability rather than price alone.
Point of View
Firms like DeepSeek have effectively exported deflationary pressure onto Western API incumbents, forcing OpenAI into an 80% price cut that would have seemed implausible twelve months ago. What mainstream coverage underplays is the feedback loop: cheaper inference expands enterprise adoption, which grows token volumes, which funds further model development — a dynamic that benefits well-capitalised open-source ecosystems disproportionately. The companies most exposed are mid-tier cloud AI providers who lack both the scale to absorb margin compression and the open-source credibility to attract cost-sensitive developers.
NationPress
10 Aug 2026
Frequently Asked Questions
Why have enterprise AI inference costs fallen in 2026?
Enterprise AI inference costs have fallen to a 2026 low due to an intensifying global price war and growing adoption of low-cost Chinese open-source models such as those from DeepSeek , according to Jefferies . OpenAI cut prices for its GPT-5.6 series by up to 80% , while Chinese firms continue releasing competitive open-weight models that drive down market rates.
What is the current AI inference price per million tokens?
As of August 6–8, 2026 , average AI inference prices ranged between US$1.16 and US$1.18 per million tokens , the lowest level recorded this year, per Jefferies citing Silicon Data . This compares to US$2.04 on May 31 and US$1.45 in late July .
How does DeepSeek affect global AI pricing?
DeepSeek and other Chinese open-source AI firms have introduced high-performance, low-cost models that enterprises can deploy without paying premium API fees. This has intensified competitive pressure on Western providers like OpenAI , Anthropic , and Google , pushing the entire market toward lower price points.
What did OpenAI do to compete on AI pricing?
OpenAI escalated the AI price war last month by cutting rates for its GPT-5.6 model series by up to 80% . The move came as rivals like Anthropic also lowered costs, with Claude Opus 5 reportedly offering performance comparable to the flagship Fable 5 model at half the price.
Which companies are tracked in AI inference pricing research?
Silicon Data 's index tracks pricing across business API providers and open-weight inference platforms used by software developers, including OpenRouter . Companies referenced in the broader competitive landscape include OpenAI , Anthropic , DeepSeek , Google , Tencent Holdings , Zhipu AI , Moonshot AI , and MiniMax .
