As 2025 has kicked off and the political landscape is well into high gear, an unexpected model launch sent the American AI landscape into turmoil. It is well documented that much of the Western world’s invested wealth lies on the performance of seven American tech stocks, better known as the magnificent 7: Apple, Google, Amazon, Mircosoft, Meta, Tesla, and Nvidia.
When a Chinese hedge-fund-owned AI called DeepSeek r1 caught investors’ attention, with reports that training it cost less than $6 million, it sent the market into a tailspin. Nvidia lost $600 billion in market cap, and reports of tech billionaires losing up to $100 billion are still pouring in. But while users flocked to the free-to-use model, which performs on par with American technology that costs $200, the result wasn’t unexpected.
Nvidia’s secret chip & Chat GPT’s role in DeepSeek

What are some of the costs of running a Large Language Model (LLM,) be that in the US or China? – Large data servers, several engineers constantly teaching, evaluating, and moderating responses from the model – this is important as large data servers often run on the top-of-the-line AI Chips made by the likes of Nvidia – often costing between $32000 and $40000 per chip.
These chips are proprietary and by law not to be sold to China. But the Chinese AI Model Deepseek r1, is a model that mostly runs on the last generation of Nvidia chips, but through proxy importation by the likes of Singapore, China still has access to the same chipset as the United States. Now – although trained on mostly Chat GPT outputs, the technology is available for free, and competes with OpenAI’s $200 offering, at no cost.
The Cost of LLMs & The Political Class’s Reaction
The economic costs of maintaining such large data servers aside, the number of engineers meant to work the back end of such systems to make sure they run smoothly. The new administration is closely associated with Elon Musk, CEO of X, which saw large layoffs to boost “efficiency,” when he took over the social media company in 2022. And what took a hit as these layoffs came into effect? Moderation and fact-checking took a downturn resulting in all advertisers pulling out.
While LLM and AI systems do not have advertisements (yet?) that entire endeavor was indicative of a loss in legitimacy and trust. Should the new administration follow in Musk’s footsteps in an effort to cut costs rapidly to reassure investors that the model’s Chinese counterpart’s efficiency is achievable, it may lead to the same loss in legitimacy and trust that users have in these AI models.
While touring the country and on a pit stop in Florida, President Donald Trump remarked, “The release of DeepSeek AI from a Chinese company should be a wake-up call for our industries that we need to be laser focused on competing.” However, did not comment on the substantial questions around the platform, such as American users’ data protection, and competition to the domestic platforms. This is following the President’s launch of the $500 Billion Stargate project to further America’s standing in the AI industry.
Less than a month after American users flocked from TikTok to RedNote to protest the government’s potential ban on TikTok, the more advanced and free-to-use Chinese AI has attracted a lot of attention. Not all of which has been good for it.
What makes a Chinese product as cheap as it is?
Be it a one-to-one copy of a designer clothing piece or replicas of technology of computer parts, the Chinese manufacturing industry delivers products at a fraction of the price. But what are the hidden costs?
This week the world’s leading cloud security service platform, wiz.io, released a report where their engineers uncovered users’ sensitive data available readily – “Within minutes, we found a publicly accessible ClickHouse database linked to DeepSeek, completely open and unauthenticated, exposing sensitive data… This database contained a significant volume of chat history, backend data and sensitive information, including log streams, API Secrets, and operational details.”
Without getting into the weeds of the technical aspects of LLM technology, OpenAI ’sChat GPT API, or the code that lets a user access the company’s servers to use the AI, is a closed source. Meaning, it is OpenAI’s private property. On the other hand, DeepSeek’s API is open source, meaning it is in the public domain. Anyone can access it, at no cost. However, when using the domestic Chinese servers to access DeepSeek’s Chinese servers, domestic Chinese content guidelines are enforced, in other words, censorship. Users across the internet have tried asking the model questions about Taiwan, Hong Kong, and the ongoing Indo-Chinese border disputes, to no avail.
The effect of AI on the environment
The Washington Post reported that asking the latest LLM to generate 100 words of text requires just over a full bottle of water. That is exponentially more than the amount of water required for several google searches on the same topic. By 2030, AI servers will account for just under 4% of the world’s electricity consumption. Increasing energy demands in the AI sector have led giant tech corporations like Nvidia and Microsoft to begin procuring their nuclear energy facilities. This doesn’t account for the present. Chat GPT and other large data servers that power models like it emitted nearly 4% of the world’s greenhouse gases, per a 2023 report from Columbia University.
So – much like how more people started traveling more when electric cars made fuel cheaper, more people will continue to use more AI when electricity for these models is readily available. This doesn’t align with any realistic goals regarding preserving environmental health. This does however align with the new President’s executive order, pulling America out of the 2015 Paris Climate Agreement, the threshold of which the world inches closer by the day.
America and the world lie in a delicate balance, but especially in the world of AI, because it is no longer the world with AI. Each day it seems closer to an AI’s world, of which we are a part. Proper regulations, with users’ data protection, and transparent model training practices will make or break the technology that is all set to shape the coming years and the way in which future generations will work, learn, and live.