Jahangir Raina
So DeepSeek has triggered a trillion dollar rethink. Folks are questioning whether we really need tonnes of GPUs and energy to run the infra for AI.
I am especially curious about the sentiment around that $500 billion AI investment announced a few days ago with authority and swagger. The investment is based on the hypotheses that we will need lots of computing power and electricity.
Billions have been invested on those hypotheses which could potentially be wrong. DeepSeek is therefore a significant moment in the AI trajectory.
Despite the urge to politicise the tech rivalry between China and the US, China hasn’t actually produced many tech companies with dominant market share at global level.
There are Baidu, ByteDance, Alibaba, Xiaomi, SenseTime: Major players in their own right but we are not talking any serious challenge to US competitors as yet. Perhaps DeepSeek is the moment. Or perhaps it is too early to draw conclusions.
I want to draw some parallels between DeepSeek and another Chinese success story from the past and highlight what a cost advantage can potentially achieve if executed with experience.
Among the tech competitors emerging from China, telecom industry is familiar with the story of Huawei. Not quite a ‘Sputnik moment’ or a thunderous entry but a more sustainable impact that the telecom industry has adjusted to.
Let me talk about Huawei a bit.
Huawei has been around since the 80s, started outreach out of China in late nineties, and within a decade thereafter established itself as one of the incumbent telecom vendors internationally.
Huawei started as a PBX manufacturer and had its first US partnership deal with 3Com (later acquired by HP) for datacom products. That was the first time that telecom companies outside China actually heard about Huawei.
As Huawei was scaling its business, China was getting ready for the WTO entry. China was about to open its telecom market and before it did so it went wild with network expansion building as much footprint as possible before the expected foreign investors made entry into its domestic market. The expansion was government driven and created entrepreneurship opportunities on the equipment side as well. Huawei was encouraged and grew as a result.
That growth at Huawei fuelled international ambitions. We started to see the vendor developing carrier grade communications equipment. For the voice equipment market the announcement of high tier softswitch by Huawei was a wake up call.
Like DeepSeek, the USP of Huawei was its cost advantage, an advantage it openly flaunted. Prices per telecom port/line came crashing down from $30 to $3. Competitors had to bite the dust. Lucent and Nortel fell in the US. Alcatel and Marconi in Europe. Ericsson and Nokia survived by virtue of their focus on wireless. If you ask those surviving vendors today, Huawei would be the competitor that keeps them up at night.
The Huawei generation of Chinese tech companies were rightly challenged for being actively promoted by the Chinese government support in scaling their business. But one cannot say the same about DeepSeek generation startups. There is no government subsidy or support to speak of.
To achieve a substantially low cost foundation model that rivals OpenAI is truly a wake up call with a much bigger impact than that achieved by a generation of companies belonging to Huawei era. However the experience of companies from preceding tech generation can potentially complement the strengths of new AI startup ecosystem in China.
Chinese AI companies are challenged by the lack of access to AI optimised chips till the ecosystem evolves domestically. Huawei is one of the companies that has a chip offering – optimised for the telecom apps. DeepSeek actually already runs inference on Huawei chips. What are the odds of two challengers to global incumbents from two different generations collaborating on technology?
DeepSeek can significantly lower down the development and operational costs for conversational AI apps developed in telecoms. As in other industry use cases, telcos will have more choice in terms of LLM partnerships. Where the customer scale does not exist – such as the scenarios involving local languages of smaller population groups – low cost models can help bridge the AI divide.
Another area in telecom where DeepSeek might have an impact is Voice AI. The company has not announced voice mode or speech recognition features yet. Once it does, the Voice AI industry will celebrate, I am sure.
The emergence of DeepSeek upon the AI landscape echoes the trajectory of Huawei in the telecom industry, underscoring the transformative power of cost efficiency in challenging established global players.
Just as Huawei disrupted the telecom market by slashing costs, DeepSeek seems poised to redefine the AI industry by demonstrating that AI models can be developed and deployed without huge amounts of computing power or energy consumption.
Article has been published by ai@telecom Newsletter
