AI Race Intensifies: DeepSeek Aims for USD 70 Billion, OpenAI Cuts Prices

Chinese artificial intelligence start-up DeepSeek reportedly seeks approximately USD 70 billion in a new financing round, targeting a pre-investment valuation. This follows a Series A round that secured around USD 7 billion, valuing the firm at nearly USD 60 billion. DeepSeek is also expanding its workforce.

Credit: UNSPLASH

OpenAI Chief Executive Sam Altman has indicated the company's willingness to significantly reduce pricing for its latest AI models. This move responds to increasing competition from US rival Anthropic and a range of Chinese alternatives. Mr Altman noted via social media that OpenAI's GPT-5.6 Sol currently stands at half the cost of Anthropic's Claude Fable 5. He further stated that OpenAI would be content to provide its services at one-quarter of the price. OpenAI released GPT-5.6 Sol with pricing at USD 5 per million input tokens and USD 30 per million output tokens. Anthropic's Fable 5, the consumer variant of its Mythos 5 model, is priced at USD 10 per million input tokens and USD 50 per million output tokens.

DeepSeek's new fundraising efforts stem from investor demand. The earlier Series A round was restricted to a select group of backers. Other investors eagerly sought to support a national AI champion, according to sources cited by the South China Morning Post. And the previous round attracted notable Chinese tech organisations, including Tencent Holdings, NetEase, and JD.com. Contemporary Amperex Technology Ltd, a major electric-vehicle battery manufacturer, also participated, alongside venture capital firms such as Monolith, Loyal Valley Capital, and Shixiang. Sources cautioned that these fundraising discussions remain at a preliminary stage, meaning the potential valuation could change.

DeepSeek also has intentions for a domestic initial public offering on Shanghai's Star Market. This listing would align with Beijing's desire for an AI model start-up on the exchange, which currently lacks one. Zhipu AI and Minimax, both Hong Kong-listed entities, are similarly pursuing listings on mainland Chinese exchanges. DeepSeek, however, is not prioritising an immediate listing, focusing instead on developing more advanced and capable models amid intense domestic and global competition within the sector.

DeepSeek's rivals include Zhipu AI, whose GLM-5.2 model has gained global acclaim for its coding capabilities. DeepSeek itself commenced a hiring initiative in late June, announcing plans to at least double the size of each department. The organisation listed 33 open positions across seven main categories, encompassing full-stack development and algorithms, AI core system research and development, deep learning research, and model data strategy product management and engineering. But this pricing pressure extends beyond US shores, with Chinese devs actively driving down costs.

China represents a significant source of this pricing competition. Domestic AI devs spent a considerable portion of this year reducing the cost of frontier models. While OpenAI has lowered prices for its GPT-5.6 series compared with previous products, its flagship and mid-tier models continue to be more expensive than prominent Chinese counterparts. For instance, Zhipu AI's GLM-5.2 model charges approximately USD 1.40 per million input tokens and USD 4.40 per million output tokens. In comparison, DeepSeek's V4-Pro model has standard rates of USD 0.44 per million input tokens and USD 0.87 per million output tokens.

ByteDance has also introduced pricing incentives. Its Volcano Engine cloud division announced a promotion offering enterprise users of its Trae coding platform a 50 per cent reduction on bundled AI model usage until the close of 2026. This action aligns with a broader trend among Chinese AI providers, who seek to attract enterprise customers through lower inference costs. Alibaba, Tencent, and MiniMax have all introduced discounts or promotional offers for their flagship models in recent months. So, the market sees continued adjustments.

The simultaneous reduction in prices and the rapid advancements in the capabilities of both US proprietary and Chinese open-weight models have made cost an increasingly crucial factor for users selecting AI services. Proprietary models require users to pay for cloud-based access. In contrast, open-weight alternatives allow users to download the underlying code at no charge and deploy it on their own local hardware. A UBS report published last month stated that increasing AI token costs were a "real concern for most organisations," leading to greater interest in open-weight models among larger global enterprises.

A Goldman Sachs research report, published last week, indicated that Chinese open-weight models are "reaching a critical point of intelligence performance" when compared to proprietary models. Despite this, GPT-5.6 demonstrates efficiency in inference. Artificial Analysis, an independent platform for AI benchmarking, reported that the model completed coding agent tasks using roughly one-ninth of the output tokens consumed by DeepSeek V4. The GPT-5.6 model also achieved higher overall scores in these tasks.

• DeepSeek aims for a USD 70 billion pre-investment valuation in its latest funding round, following a USD 60 billion valuation from its Series A.

• OpenAI's Chief Executive has indicated potential significant price reductions for its AI models, citing competition from Anthropic and Chinese providers.

• Chinese AI developers, including DeepSeek and Zhipu AI, offer models at lower costs than OpenAI's current rates, driving market price competition.

• Rising AI token costs and the increasing capabilities of open-weight models have amplified interest from global organisations.

• DeepSeek is also pursuing a domestic initial public offering on Shanghai's Star Market while prioritising model development and a hiring expansion.

Source: SCMP