The day Donald Trump assumed office for his second term, a Chinese startup called DeepSeek quietly released an AI model called 𝗗𝗲𝗲𝗽𝗦𝗲𝗲𝗸–𝗥𝟭. While Trump’s subsequent shake-ups were expected, little did one know that DeepSeek-R1 would send tremors in the AI industry.

Benchmark tests have proved DeepSeek-R1 bests the performance of OpenAI’s best model yet – the 𝗢𝗽𝗲𝗻𝗔𝗜 𝗼𝟭 and accomplishes this at a fraction of the cost o1 and the other similar AI models incur. DeepSeek was founded in 2023 with efficiency and cost as its primary drivers. The company claims it has spent less than $6 million in computing resources for training its model. In comparison, it is estimated that this cost for GPT4 was hundreds of millions of dollars!

Now comes the interesting part. For the users of the DeepSeek-R1 model the 𝗰𝗼𝘀𝘁 𝗼𝗳 𝗶𝗻𝗽𝘂𝘁 𝗮𝗻𝗱 𝗼𝘂𝘁𝗽𝘂𝘁 𝘁𝗼𝗸𝗲𝗻𝘀 𝗶𝘀 𝗮𝗯𝗼𝘂𝘁 𝟯𝟬 𝗧𝗜𝗠𝗘𝗦 𝗹𝗲𝘀𝘀𝗲𝗿 than in OpenAI o1!  Total game changer! DeepSeek has accomplished this feat despite the imposition of constraints and sanctions by the US on China. Incredible story. Reminds one of the Wright brothers’ story. Samuel Pierpont Langley, an American physicist was heavily funded by the U.S. government in his attempts to build a powered, heavier-than-air aircraft. However, through their innovative approach and meticulous experimentation, the Wright brothers succeeded in achieving this with a lower budget. DeepSeek has proved yet again loads of money and resources are not necessary for innovation to thrive. The team at DeepSeek dared to think differently, challenged the basic assumptions in other AI models and the rest is history.

This morning a colleague pointed out that the model has constraints – it does not answer some sensitive questions. Not unexpected, I said to myself. Since DeepSeek has open sourced DeepSeek-R1 (𝘸𝘩𝘪𝘭𝘦 𝘖𝘱𝘦𝘯𝘈𝘐 𝘰1 𝘪𝘴 𝘱𝘳𝘰𝘱𝘳𝘪𝘦𝘵𝘢𝘳𝘺) a simple hack is to download it and use it with, say, Perplexity, and the constraints vanish! No wonder the AI developer community is rejoicing. DeepSeek has also blasted the 𝗦𝗰𝗮𝗹𝗶𝗻𝗴 𝗛𝘆𝗽𝗼𝘁𝗵𝗲𝘀𝗶𝘀 – (that you need more compute and resources to churn out better and more intelligent models) by accomplishing similar results deploying just a fraction of the resources. This has sent shivers down the spine of Nvidia, OpenAI and others who have invested billions of dollars to scale up compute. On Monday the 27th of Jan, NVIDIA shares tumbled and eroded $600 billion in its market value.

Just till a few days back, the world was staring at a massive energy shortfall in the future due to the projected needs of energy-guzzling AI infrastructure. With DeepSeek R-1 consuming much lesser energy, things will look much better on the energy front and climate activists will heave a big sigh of relief.

AI stands democratized.

Thanks to drastically reduced costs we will witness an exponential rise in number of AI-based applications (and of course start ups ) with accelerated adoption of AI in business. We have heard this refrain from CXOs “ I see a lot of AI Proofs-of-Concept being done. Show me one big AI project”. This will now change as the cost of implementing full size projects nosedives. We are stepping into an exciting era of AI Transformation.

Lower infrastructure and running costs for AI will mean more wings to government sponsored learning and skilling initiatives, and other programmes targeted at the masses. I haven’t seen a more significant disruption yet, and am sanguine this will eventually lead to a state of abundance and overall improvement in the happiness, quality of life and health of all of humankind – a realisation of the universal prayer :

𝘰𝘮 𝘴𝘢𝘳𝘷𝘦 𝘣𝘩𝘢𝘷𝘢𝘯𝘵𝘶 𝘴𝘶𝘬𝘩𝘪𝘯𝘢𝘩
𝘴𝘢𝘳𝘷𝘦 𝘴𝘢𝘯𝘵𝘶 𝘯𝘪𝘳𝘢𝘮𝘢𝘺𝘢𝘩 …