January 20, 2025, DeepSeek releases the DeepSeek-R1 AI modelIt is immediately clear that is at the level of GPT-4. This alone wouldn't be enough to shake up Open AI… except that it costs about $1 million and 5,58 GPUs to develop, train, and power DeepSeek-R2.000. That means the Chinese AI model is roughly It costs 90% less and wastes 1/8 of the resources. And it is also open-source.
How Chinese AI is disrupting the market world technology at high levels. Different Western nations take cover, banning and indicting DeepSeek.
Content index
DeepSeek Burns NVDIA on the Stock Market
After the release of the Chinese model, Nvidia recorded a loss of about 600 billion dollars in market capitalization. We are talking about an impact that is both heavy and significant. In fact, the possibility of developing advanced artificial intelligence models with limited resources could reduce the dependence on large investments in data centers and especially in GPUs and advanced microchips: NVDIA's core business.
But it could be a flash in the pan
The costly stock market crash, on the other hand, could be just the result of an initial scare. Nvidia may not feel the blow for 3 reasons:
- Big tech companies that already have their own AI models are working on developing more innovative and powerful solutions.
- Lower development costs encourage other companies to create their own generative AI.
- The more players enter the game, the more fragmented the competition becomes.
In several cases, therefore, the demand for GPUs and advanced microchips can increase. Just think that already in the first half of '24 Elon Musk launched Grok-1 under the Apache 2.0 license: also an open-source model, less accessible than DeepSeek but. In the meantime Meta Platforms, Amazon and Alphabet they are investing in the development and adoption of generative artificial intelligence.
The economic threat of DeepSeek, at least in this light, seems to lack real traction.
DeepSeek Under Fire for Ethics and Privacy
January 28, 2025. The Guardian spreads the testimonies of some users. According to the sources reported in the news, if you ask DeepSeek Questions about Tiananmen Square and Taiwan, the model behaves like this: first provides detailed answers, then delete them immediately replacing them with generic or error messages. In parallel, other technical analyses reported on Wired, they claim that DeepSeek implements a two-tier censorship system: the model generates a response and then a filter evaluates the content, eliminating the parts deemed inappropriate. This mechanism is present both in the official application and in the open-source versions run locally, indicating that the censorship is built into the model itself.
On the other hand, Chinese AI censorship can be circumvented in a good measure, asking questions differently, using special characters and other languages. But this is not enough to remove DeepSeek from the crosshairs.
Investigations and blockades of nations
January 30 2025. Il Privacy Guarantor Italian temporarily blocks DeepSeek because there is no transparency on what data is collected, on its sources, on the purposes of the processing and on the possible storage of personal data on servers in China.
Investigations are also underway in North America and Europe.
La Feroot Security (company of cybersecurity Canadian), shares the results of his research with theAssociated Press: the website of DeepSeek contains a code that could send information user login to a state-owned company, China Mobile. No data transmissions were observed during access from North America. However, it cannot be ruled out that this may occur.
3 February 2025. In Italy, the Privacy Guarantor orders DeepSeek to block its chatbot after the company has provided insufficient answers regarding the types of data collected, the sources, the purposes and the location of data storage.
5 February 2025. South Korea's Ministry of Industry Blocks Access to DeepSeek. Entities such as Korea Hydro & Nuclear Power and the Ministry of Defense join.
Australia Bans the use of DeepSeek on all government devices. China responds by accusing Australia of "ideological discrimination" and political actions. The same line is followed Taiwan, while the state of the Texas imposes limitations, precisely in the United States, Where a bill is already being proposed to imitate Australia.
Open AI: “DeepSeek plagiarized us”
On this wave OpenAI accuses DeepSeek of intellectual property infringement. Open AI claims that DeepSeek-R1 trains at the expense of competitors through a technique called “model distillation.” In short, the Chinese are accused of plagiarizing the Americans.
This accusation has yet to be verified.… But The Times is already talking about it.
Should every nation have its own digital sovereignty?
The advent of DeepSeek is therefore worrying at a political level. The fear is that foreign AIs could expose citizens' data to foreign jurisdictions. From this point of view privacy, national security and digital sovereignty of nations are at riski.
If we continue to fuel this dynamic, nations will become increasingly closed off. The issue could lead to riots digital nationalism, with the creation of its own models, or in individual initiatives, as DeepSeek just demonstrated that it costs less to create models…
Or the polarization between Open AI and DeepSeek could intensify, reducing the issue to “USA vs China”, with a greater entrenchment of the centralized monopoly of Western big tech.
Perhaps the danger is not DeepSeek itself, but how it competes
The competition that DeepSeek is triggering is not only worrying on an economic and political level. It is the competition itself, perhaps, that is more worrying than the conflict.. If AI continues to be a force in geopolitical balances, it will be increasingly trusted. Thus, according to critics such as Evgeny Morozov, There is the danger of a techno-utopianism, that is, that thethe belief that technological innovation can solve all problems and improve the human condition.
An example of the illusions we witness is the consumption paradox: massively using AI to improve energy efficiency (therefore reducing consumption) leads to increasing it. A paradox that opens us to the very narrative that individual generative models favor. Excessive trust in their effectiveness makes us believe that the results of their responses are increasingly authoritative and complete. Instead, in addition to there being an inequality in access to technologies, no model distributed on a large scale is perfect, free from cultural biases, limits and censorship: whether Western or Eastern.
The world never changes (?)
Sometimes it may seem to us that the world has changed radically because technology and media have developed at an exponential rate in the space of 20 years, but certain dynamics remain unchanged: East and West do not trust each other, they have a different vision of the world and they challenge each other.
DeepSeek Proposes a Cheap and Efficient Open-Source Model, demonstrating that the AI monopoly can fade into a hybrid between socialism and capitalibertism. Thus it clearly contrasts with the elitist technocracy and algorithmic corporatism represented by Open AI, which is closely linked to the USA. The idea of artificial intelligences being used for propaganda, disinformation and espionage purposes is not that far from what happened between Russia and the United States during the Cold War.
But artificial intelligence is a new and very powerful tool: this is the difference.
Through ancient dynamics, we risk consequences that could truly change the world.
