Need proof that the generative AI market has become a bloodbath? Look no further than Alibaba Cloud.
The cloud computing arm of Chinese tech giant Alibaba just announced eye-watering price cuts of up to 85% across its AI offerings. The steepest discount applies to their visual language model Qwen-VL, according to CNBC.
It's just the latest salvo in an intensifying battle among China's tech titans. Alibaba, Tencent, Baidu, JD.com, Huawei, and TikTok's parent ByteDance have all launched large language models in the past year and a half. With little to differentiate their products, they're locked in a price war that mirrors what's happening in the West.
The Western Front
The price war kicked into high gear in August when OpenAI announced steep cuts, with Google following suit two days later, slashing prices for its Gemini 1.5 Flash model by up to 78%. Both companies also rolled out cheaper, stripped-down models for basic tasks.
Anthropic's strategy has been more nuanced. While they raised prices on their new small-scale Haiku model, betting on superior performance, they also introduced a new Sonnet 3.5 model at a fraction of their flagship Opus model's price. Since Sonnet matches or outperforms Opus on many tasks, it effectively amounts to a steep price cut - few customers still see a reason to pay premium prices for Opus.
To justify higher prices, AI models need clear competitive advantages - what investors call "moats." But since GPT-4's debut, improvements have been mostly incremental. Making matters worse, open-source models like Meta's Llama are getting increasingly capable and computationally efficient.
Chinese AI startup Deepseek recently demonstrated this reality by matching GPT-4 and Claude's performance with relatively modest investment. They're not just offering competitive API prices - they're making their model available as open source.
The Premium Play
OpenAI is testing the waters for premium pricing with a more capable o1 model, available through ChatGPT Pro subscriptions, though it still needs to prove its worth. Google has stated it isn't planning a similar premium offering, at least for now.
Meanwhile, OpenAI appears to be plotting a different strategy for its standard ChatGPT service - gradual price increases that could double costs over five years. The goal? To reach $100 billion in revenue by 2030, which, according to its contract with Microsoft, would allow OpenAI to claim it has invented artificial general intelligence (AGI).
According to The Information, future, significantly more powerful OpenAI models might command monthly subscriptions as high as $2,000 - potentially accelerating the path to that revenue milestone and the AGI declaration that comes with it. OpenAI recently introduced its more capable o3 model, which might justitfy higher prices, but it also will increase costs for OpenAI.
These premium prices might make sense if AI models eventually replace human labor. Until then, it looks like a war of attrition among model providers - one where only the strongest (or best-funded) will survive.
