Seeing Silicon | DeepSeek has democratised the AI field. What now?
DeepSeek shook the global AI markets and showed that everyone from your aunt to your country can create their own AI model. What happens now?
President Donald Trump had just announced Stargate, a $500 billion AI venture to keep US dominance in AI going when DeepSeek hit the world like wildfire. The startup from Hangzhou, a city in eastern China, comprising recent graduates from Chinese universities, had launched R1, its AI model that was much cheaper, used less computing power and performed better than AI models coming from the Silicon Valley billion-dollar companies in some parameters.
On Castro Street, in downtown Mountain View, which is the headquarters of Google, everyone from screen-dazed techies in coffee shops to food app delivery groups milling outside restaurants furiously discussed the new free app, which became the most downloaded free app in the United States this week, overtaking OpenAI’s ChatGPT. The DeepSeek model was open source, allowing anyone to download, copy and build upon them. As the business potential of owning an AI model poofed in the air, so did $600 billion of market cap from NVIDIA's valuation, a whopping 17% of its total net worth.
The cat’s out of the bag
All last week a mix of denial, anger and suspicion played out in Silicon Valley offices as AI-junkies stared at this plummeting stock portfolios. Open AI seemed to have been impacted the most. Its CEO, Sam Altman, has been convincing Washington DC to ramp up data centres and energy requirements that were essential to scale AI models and achieve AGI or artificial general intelligence.
The Chinese startup didn't have access to a lot of infrastructure but had managed to train it at a fraction of computing and energy requirements. As a result, R1 is 50 times cheaper to run than many other US AI models. Overnight OpenAI’s complete business model has vanished (why would you pay for a premium service when you get a model for free?) and its expensive training processes have come into question in its boardroom. “We will obviously deliver much better models and also, it’s legit invigorating to have a new competitor,” tweeted Sam Altman.
The onslaught of China sweeping the AI carpet off the country, which has been touting itself as the AI leader has not gone well with the US government, which likes to talk free markets but can’t take competition from its tech rivals. Already, security concerns have been raised about privacy issues where DeepSeek could potentially send user login information to a Chinese state-owned telecom company. There’s also a hue-and-cry about censorship on the AI model which refuses to give inputs on the Communist Party of China or other topics sensitive to the Chinese government. Somewhat like usage restrictions put on public AI models like Gemini and ChatGPT – but no one is asking that question yet. (Try prompting American models about Palestine and see what you’ll get.)
In the middle of this week, a senator in the US introduced a potential legislative bill that could criminalise DeepSeek download in the US with 20 years in prison. I would’ve thought this was a headline created by a badly made, hallucinating AI model, but unfortunately, the models are yet to catch up with our real lives. The panicked reaction by the US government is not only a reflection of how important the AI boom is for a sagging US economy, it’s also a reflection of the hyperbolic times we live in. In our headline-driven world, no one has asked the question of how will you stop an easy copyable, open-source model which can be run on a single laptop.
China just democratised the AI field
Amid all the frenzied dissing of technical abilities blaming and the tech cold war between the new AI giants China and the United States, one thing is clear. Ironically, China has democratised building AI models. If a small startup, which hired domestic engineering students could make an efficient model, why can’t India do this? Or any other country in the world? For the last two years when OpenAI came out with ChatGPT, companies and governments across the world believed that you need more compute capacity to scale AI models and their performance. DeepSeek’s model shredded that belief to pieces. It’s now possible to build low-cost foundational models.
As competition heats up in the rocky AI wilderness, AI models have become cheaper to make and the world, especially the Global South and its hungry governments and startups, have pounced on the opportunity. For example, IndiaAI Mission, an Indian government initiative, has called for proposals from startups, researchers and entrepreneurs to collaborate on building foundational AI models trained on Indian datasets.
Even as R1 is being banned in countries, DeepSeek’s R1 has levelled the AI field and dramatically lowered the barriers to entry. Whether we ignore this or call out that the king is wearing no clothes, remains to be seen.
Shweta Taneja is an author and journalist based in the Bay Area. Her fortnightly column will reflect on how emerging tech and science are reshaping society in Silicon Valley and beyond. Find her online with @shwetawrites. The views expressed are personal.
