Introduction

Nvidia’s stock market valuation dropped by six hundred billion dollars on 26^th January 2025. This is twice the GDP of Portugal and nearly the GDP of Agrentina. The business of the global leader in AI chip design was shaken because the weekend before a generative AI app that rivals ChatGPT from an — until then — unknown Chinese research company under the name of DeepSeek had become the most downloaded free app in the United States. Nvidia’s market capitalisation was not the only one to plummet that Monday due to the Chinese breakthrough; Amazon, Alphabet (Google’s parent), Microsoft and Meta suffered significant, though less dramatic, losses, resulting in a total drop of one trillion dollars in a single day.

This was a market overreaction — partially corrected over the following days — and one which cannot be explained by DeepSeek’s R1 model being better than existing ones. In fact, the model does not differ significantly from other cutting-edge generative AI models, particularly OpenAI’s ChatGPT. But the mere fact that a Chinese model is considered as good as the forerunner and other US models is a major development.  It is also a geopolitical reason strong enough to shake the stock value of the world’s largest companies in market capitalisation, especially that of Nvidia.  

DeepSeek used a training method that extracts learning from other models reducing the training rounds and thus the processing power needed for training. Needing fewer AI processors per model means that Nvidia will be selling potentially less AI chips.

The DeepSeek episode can be interpreted through several lenses. One is that China caught up to the US, validating their stamp as the second place tech superpower. Indeed, China challenges the US leadership in digital technologies. According to a Feifan Research report, half of the world’s 1500 active AI companies were based in China by June 2024.^[1] And, although patents are not the main knowledge appropriation mechanism in this sector, Chinese large tech companies lead in the number of generative AI patents.^[2]

Another interpretation is that there is no serious threat to US leadership because the model is not as performant as ChatGPT. The US dominates the generative AI global market with a share of forty per cent and is also the undisputed leader in number of frontier AI models, including the largest ones in number of parameters.^[3] This leadership is unlikely to be undermined because a Chinese company has released a model on par with the best US offerings, two years after American companies pioneered such releases.

A third interpretation —  the one explored here — posits that Chinese Big Tech companies and AI startups are not aiming to leapfrog the US. Rather than pursuing global leadership, they appear to favour rapid imitation, replicating with adaptations US innovations at speed without necessarily seeking to surpass them. If this interpretation is correct, the narrative promoted by US Big Tech that holds that China’s catching-up in digital technologies is the real threat for the US economy and not their own concentrated power — which has been a recurrent topic at least since the US Big Tech congressional hearing in 2020 — would be proven exaggerated. Regardless of how close Chinese rivals may seem to be, the global and national policy and sectoral organisation contexts encourage them to prefer copying rather than taking the risks associated with pushing the global frontier.  

Moreover, on this view, the emergence of companies like DeepSeek, which compete with Western AI startups, actually favours Amazon, Microsoft and Google. As I have previously shown for Common Wealth, these three US giants control the AI value chain from their clouds.^[4] Their power is not only exercised on those offering AI-powered apps that run on their servers but also on those conceiving and developing frontier models, such as Open AI, Mistral and Anthropic. Even if these companies were never formally acquired, as Google did with DeepMind, they all rely on cloud giants — both as investors and providers of infrastructure and other technologies.  

For Amazon, Microsoft and Google, DeepSeek’s models are, therefore, good news for at least three interrelated reasons. First, by validating the hypothesis that China is getting closer to the US, they contribute to convincing regulators that they need US Big Tech to battle against China and thus should not attempt to curb their power.  

Second, DeepSeek means more competition among companies that primarily develop AI models. This pushes Western AI startups to develop even more advanced AI models, thus keeping their focus on that segment of the AI value chain without attempting to overcome their dependence on Big Tech clouds for training and afterwards running and selling their models as computing services.

Third, DeepSeek’s release expands the AI market both by fostering individual and business users’ AI adoption and by pushing Western AI startups to accelerate their pace. This further favours cloud giants because more AI means more consumption of their datacentres for data storage and processing power for developing and running new — even larger — models. Yet their clouds are much more than infrastructure offered as a rented service. Startups, app developers and companies from the most diverse sectors rent AI models, software, platforms and other digital technologies as a service on Amazon, Microsoft and Google clouds. Even DeepSeek offers its models as a rented service on these digital technologies’ supermarkets, paying a fee to cloud giants every time someone rents its models. So even if one day, DeepSeek or another Chinese company developed a model that truly outperformed Western competitors, it would likely rely on Amazon, Microsoft and Google clouds to sell it. Either way, these three US leaders always win.

In the rest of this article, I provide reasons that support the idea that Chinese Big Tech has preferred fast imitation in AI and explore how this has affected Chinese AI startups like DeepSeek. This is followed by a more detailed assessment of how the success of Chinese AI companies, both large and small, ultimately favour Amazon, Microsoft and Google.

Why Chinese Big Tech Prefers Imitation

In the economic literature, one reason for preferring imitation is insufficient capabilities.^[5] If Chinese companies and China’s innovation system remain laggards, imitation may be an alternative in the process of developing the capabilities needed to leapfrog. A way to identify this potential lack of expertise or capabilities is to examine AI scientific articles before the release of ChatGPT, which signalled the market opening of generative AI models.

To identify where Chinese organisations stood in relation to generative AI models before the release of ChatGPT, I analysed the 37,058 scientific articles presented at the top 14 AI conferences between 2018 and 2020.^[6] The starting year was chosen because the seminal piece by Google’s DeepMind and OpenAI entitled “Deep Reinforcement Learning from Human Preferences” was published in 2017. This was among the first precedents of reinforcement learning systems interacting with real-world environments. It is a building block of today’s generative AI models since they are partly trained with reinforcement learning. Unlike supervised learning, in which the model is trained with preclassified datasets, in reinforcement learning the model answers prompts and it is provided with a binary feedback stating whether the answer was good or bad. This feedback is further processed by the model to improve future results. Generative AI models combine these two ways of training.

By looking at the most frequent coauthorships of research presented at those leading AI conferences between 2018 and 2020, I found that US Big Tech, in particular Microsoft and Google followed by Meta and Amazon are the most central actors of the frontier AI research network. In network analysis, they occupy a position that is defined as being the main bridges between different parts of the network, which means that they have the largest capacity to influence on the whole network’s research agenda. Chinese Big Tech also integrates the network but mostly coauthoring with other Chinese organisations and Microsoft, which has had a strong research foothold in China for decades.^[7]

While Amazon, Microsoft and Google were already frequently presenting research on generative AI at these events, Chinese corporations remained focused on pre-existent AI within the family of deep learning models applied to their main business areas. Nonetheless, on a closer look, it is possible to identify that some Chinese Big Tech companies were working on techniques needed for developing generative AI models.

Because the network methodology used for this analysis is a relative measure that links organisations with the topics that mostly represent them in relation to all the possible links between organisations and topics, a closer look at the Chinese sample of presentations can provide insights on who was leading generative AI research inside China.

Among the 8614 presentations out of the original sample that had at least one coauthor based in China, I extracted the 1000 most frequent terms in their titles, abstracts and keywords which included terms related to generative AI. This means that there was research on this type of model in China between 2018 and 2020 even if not as prominent as for Amazon, Microsoft and Google. In particular, it is relevant that in this smaller sample, those primarily connected to generative AI terms were not only leading Chinese universities but also Alibaba, Tencent, Huawei and ByteDance (see Table 1, generative AI terms are bolded). By contrast, Baidu’s research remained characterised by topics that are not directly related to generative AI. This does not mean that Baidu was not researching on generative AI but that other Chinese organisations were more actively working on these topics (see Table 1).

[.fig]Table 1: Chinese Big Tech most frequent terms in presentations at the top AI Conferences (2018-2020)[.fig]

[.notes]Source: Author's analysis based on Scopus data. Generative AI-related terms are bolded.[.notes]

Alibaba is the Chinese large tech company associated with more topics related to generative AI, particularly language models. Huawei’s publications dealt with “reinforcement learning”. Tencent and ByteDance were presenting research on “relation extraction”, a term that refers to predicting and labelling semantic relationships between entities in text analysis. Only Baidu’s presentations were not dealing with research on topics related to generative AI.

Seeing the global and Chinese analyses of scientific presentations together, it can be argued that at least some Chinese giants were interested in frontier AI but not to the same extent as US Big Tech and other organisations in the global network. This is because, in relative terms, they were less frequently presenting on frontier AI topics than others in the global network; Chinese companies were not connected to generative AI multi-terms. This changes once the sample of presentations is narrowed to proxy China’s AI research network. In the China-based sample, Alibaba and Huawei, followed by Tencent and Bytedance appeared among those presenting more often on topics related to generative AI than other organisations.

These results suggest that Chinese leaders had generative AI research capabilities and were presenting results on this area at the world’s leading conferences years before ChatGPT was released. Even if they were not presenting as many papers as their US counterparts, they were still working on the topic and achieving results that were good enough to be presented at the world’s most important AI conferences. Hence, it was not a lack of technological capabilities behind the fact that they did not first come up with an interface like ChatGPT.

Chinese large tech companies have been enjoying what the economics of innovation literature describes as state-driven and demand-driven windows of opportunity. The former is generated by state policies protecting and encouraging AI in China. The latter is explained by the size of the Chinese market and the demand for AI products.^[8] Given that Chinese large players had frontier AI capabilities before ChatGPT was released and that they had a secured a large market for releasing such products, they could have released AI solutions ahead of the US. Why was it, then, that they did not come up with an interface like ChatGPT ahead of OpenAI and that it was DeepSeek and not Chinese Big Tech that eventually offered a model? While it is impossible to completely rule out bad luck, there are several reasons to think that they actually prefer fast imitation.

Motivations for imitation

There is a somewhat entrenched assumption among AI scientists and engineers that China excels at cloning and scaling digital technologies but struggles to create something completely new. An example is how Alibaba adapted Amazon’s business to the Chinese market.^[9] This view is often partly attributed to Chinese culture, particularly its education system, which, on this view, “emphasises respect for and attention to existing knowledge and doctrine, rather than fostering critical thinking and challenging existing limits”, thereby curbing creativity.^[10] However, this is only one possible explanation for why large Chinese tech companies have tended to imitate the US in the field of generative AI instead of turning their generative AI research into marketable solutions before their American rivals.

Another is that Chinese Big Tech does not appear to be pursuing global AI leadership, but rather a national strategy. With the exception of TikTok’s global success, the internationalisation of Chinese tech giants has so far been limited. In AI, their efforts remain largely focused on economies connected through China’s Digital Silk Road, which do not fall under US tech giants’ priorities.^[11] While the international expansion of the cloud could potentially encourage broader global adoption of Chinese AI models offered as cloud services, even Alibaba — the leading cloud provider in China — has found little to no foothold in the global cloud market. Key barriers include limited access to international talent and ongoing geopolitical tensions.^[12]

Since Barack Obama’s presidency and, more aggressively, since Trump’s first presidency, the US has implemented a range of measures that have limited the global expansion of Chinese tech companies. The US has weaponised interdependence in global value chains related to AI, particularly for semiconductors, by strengthening trade restrictions. Weaponisation could in future be expanded to undersea internet cables, over half of which are owned by US Big Tech companies, who also consume nearly seventy per cent of global bandwidth.^[13]

The US government has also put pressure on the rest of the world to follow suit. A major effect has been that, to secure its large US clients, leading chips foundry TSMC cut ties with Huawei, which was until then among its top clients. This has raised a barrier to any Chinese company aiming to develop frontier AI models given that they need huge numbers of interconnected frontier AI chips that only TSMC can manufacture and that, so far, only US companies can design, with Nvidia as the global leader.  

This could further explain why Chinese tech giants have preferred fast imitation in AI in order to make a more efficient use of their available processing power, training only models with a proven economic value. Huawei has reacted by designing AI chips manufactured by Semiconductor Manufacturing International Corporation (SMIC).^[14] They have been improving the share of successful chips per batch — the industry terms this “semiconductor yield” — but they are still far from being as performant as the tandem Nvidia-TSMC.

Growing geopolitical tensions have also led Chinese tech companies to close their research and development (R&D) labs in the US. By late 2023, the pressures to decouple even pushed Microsoft to encourage the relocation of its AI scientists and engineers from China to Canada.^[15]

Geopolitical tensions have made it increasingly likely that achieving technological leadership will not translate into global market leadership for Chinese tech companies, while it would require higher — and riskier — investments. This seems even more likely given the Trump administration’s announcements regarding trade tariffs. Even though these measures focus on goods rather than digital services — and the two countries are already significantly decoupled in that area — there is reason to believe that the global turbulence caused by these announcements further discourages Chinese tech firms from attempting to outpace the US.  

The experience of Huawei serves as a cautionary tale. Although it forged ahead in 5G, geopolitical responses curtailed its global market expansion. Huawei was the first to develop the technology, but Washington systematically shut down its potential and signed agreements for installing 5G networks in Western economies, allowing Ericsson and Nokia time to replicate the technology and restrict the Chinese telecom giant’s market growth.

Further evidence along these lines comes from Chinese companies’ involvement in international standard-setting. Chinese companies, particularly Huawei, are the most active contributors to international standardisation in information and communication technologies. However, the acceptance rate of Chinese technical proposals is lower than for the US and Europe.^[16] This reinforces the perception that geopolitical constraints hinder Chinese tech companies’ global returns in case of technological leadership.

These geopolitical limitations have translated into different strategies inside the tech sector. For instance, Huawei’s strategy in smartphones differs from its own and other Chinese Big Tech strategies in AI. After the US limited Chinese companies from purchasing foreign key technology, Huawei dedicated efforts to design its own — advanced but still not frontier — smartphone chips to retain market share even inside China. The smartphone sector has not faced the same degree of protectionism as AI, with Apple’s iPhones maintaining a dominant position in the high-end segment of the market. Apple’s market share in China has recently decreased partly due to Huawei’s competitive strength since it released smartphones with its own designed semiconductors.^[17]

Huawei’s strategy to catch-up in semiconductors was not matched by a similar strategy in AI, both for Huawei and other Chinese Big Tech. In the case of AI, foreign solutions have virtually no space in China’s domestic market mainly due to the government’s industrial policy, which limits foreign competition in AI. The Chinese digital technologies market is dominated by Chinese companies, notably Big Tech and, more recently, DeepSeek, because the public and private sectors prefer Chinese digital technologies even when they internationalise. Thus, it is unlikely that outpacing the US giants would significantly alter AI market share inside China. If anything, the race is internal, among domestic players, and that race can still be won by being the first to launch local versions of foreign technologies inside China. This is why a secured internal market can be added as a reason for disincentivising Chinese companies to outpace US companies in the development of AI-based solutions and instead prefer fast imitation.  

A complementary reason for the preference for fast imitation is that the ease of access to data within China reduces incentives to internationalise, thereby indirectly discouraging efforts to become the global AI leader. Chinese Big Tech and selected AI startups have benefited from privileged access to public sector data and have operated under historically lax data privacy regulations.^[18] Additionally, the sheer size of China’s connected population, especially those using Alibaba and Tencent platforms, makes harvested datasets sufficiently large. The domestic market is also a sufficiently large attractive business in itself. It could be argued that resulting Chinese AI models will be good predictors only to the extent of the data that they have, which further justifies remaining mostly Chinese since they were trained with Chinese data.  

Overall, the paradox of the Chinese state’s AI policy is that the same instruments that create a space for Chinese companies to develop and deploy their own digital solutions, enable them to profit from fast imitation. Once a new AI model or other digital technologies’ product is introduced and proves its economic value in the West, making a Chinese version relying on all their accumulated data is easier, less risky and protected from foreign competition. Simply put, China’s “Great Firewall” is a two-sided barrier.

Seen together, the limited opportunities to expand abroad and a secured internal market disincentivise Chinese companies from taking the risk to introduce AI innovations that push the global frontier. Meanwhile, remaining fast followers enables Chinese companies developing AI to make businesses out of the US breakthroughs. By privileging imitation, Chinese tech giants and startups reduce the economic and innovation risks of AI while competing for the local market, including Chinese government contracts.  

DeepSeek: threat or opportunity for Big Tech?

US and Chinese Big Tech not only differ in their approach to cutting edge innovation. Another relevant difference is the aim of their investment in startups. Although they all seek to control startups and profit from integrating them into their respective networks in a subordinated position, only US Big Tech has invested in AI startups working on foundation AI models. The case of Microsoft and OpenAI is only one of the dozens of agreements signed by US Big Tech with cutting-edge AI startups aimed at accessing and steering their technologies while locking them into their systems.^[19]  

Chinese Big Tech also invests prominently in startups, but its motivations differ. It was found that corporate venture capital in China is primarily driven by domestic market diversification aims.^[20] When Chinese tech giants invest in AI startups, they prioritise applications instead of foundation models. Baidu’s CEO openly advocated for AI real-world applications, seen as more important than competing for models.^[21]

Alibaba only invested in DeepSeek after it released models that could rival ChatGPT, thus once it had proven it market potential. Before, investments in AI startups were not aimed at encouraging the development of foundation models. For instance, the AI startup Megvii received venture capital from Alibaba but the aim was not to promote the company’s ongoing research on AI models but to use its face unlock technology in Alipay until Alibaba replaced it with an internal solution. Megvii was also hired to customise Huawei and Xiaomi’s AI applications. Both at the national and provincial level, the Chinese government has been one of Megvii’s main clients too. Megvii offers facial recognition with surveillance camaras for these governments that in turn facilitate access to data.  

Overall, large Chinese tech companies and the state have influenced Chinese AI startups’ priorities, pushing them to focus on narrow applications instead of more fundamental and path-breaking AI. Against this backdrop, it should not be surprising that DeepSeek was funded and founded by a tech outlier: the hedge fund High-Flyer.^[22] DeepSeek, originally called Fire-Flyer, was High-Flyer’s research branch. It initially focused on deep learning research for analysing financial data, which required stockpiling GPUs. High-Flyer accumulated them before and as the US restrictions against China took hold.

Accumulated GPUs were still far less than those used to train generative AI following the business model that US Big Tech and their satellite startups promote, which is extremely intensive in computing power. To achieve a similar model, DeepSeek used another learning technique called distillation. In this context, “distil” means automatically extracting the learning from already existing models. The technique had been used before, but for smaller-scale models that were not as good as the originals. DeepSeek is smaller in parametre count (which also makes it cheaper to use), and yet it plays in the big leagues.  

Starting from distillation, the training that remains is shorter and each round of training requires less investment because it does not start from scratch, but from learning already carried out by other models, that in the case of DeepSeek had been done by other companies. This is another reason why DeepSeek-R1 is cheaper. Although we cannot make a direct comparison to the cost of training models like OpenAI’s GPT or Meta’s LLaMA — as that would require knowing DeepSeek-R1’s total production cost, which remains secret — it is clear that DeepSeek produces a lower-cost AI. Its latest training round required only 5.6 million dollars.

The fact that DeepSeek is Chinese is, thus, only part of the reasons why Nvidia, Amazon, Microsoft, Meta and Google stocks dropped. DeepSeek has made an economic version of US leading AI models, directly impacting on the US-led global business model for generative AI. As I anticipated in the introduction, the development of frontier models without concentrating massive amounts of cutting-edge AI chips for training undermines Nvidia’s business since it is almost entirely dedicated to designing those chips. Hence its drop in market capitalisation when DeepSeek R1 came out.

DeepSeek R1 can also be downloaded and run locally, unlike every other cutting-edge model. And it is offered at a lower price than that which Meta charges for its LLaMA model, which is accessed through Amazon, Microsoft and Google clouds. Unlike Nvidia, for these three US cloud giants DeepSeek is more an opportunity to further expand their cloud businesses rather than a threat.

Amazon, Microsoft and Google Always Win

Virtually all generative AI startups today rely on at least one cloud giant, including foreign companies like DeepSeek or France’s Mistral. Although DeepSeek models can be downloaded, they are also available as a service on these giants’ clouds. This illustrates that the cloud’s dominance goes way beyond infrastructure as a service, since even companies that have not used the cloud for training their models, like Meta and DeepSeek or more recently xAI, depend on the cloud for offering them. This is why the success of the Chinese DeepSeek does not pose a serious threat to them. DeepSeek does not aim to compete with cloud giants but with companies like OpenAI and Anthropic whose business is exclusively selling models. Cloud giants also sell their own models, but this is only a tiny part of their business.

In fact, the expansion of competition in AI models furthers cloud giants’ overall dominance since it deters leading AI startups from diversifying and eventually competing by developing an independent cloud ecosystem. More competition at the AI modelling layer makes Western AI startups even more dependent. Amazon, Microsoft and Google are often their main investors and host most of the demand for paid AI usage outside China. Companies, governments, universities and more see the cloud as the one-stop shop for all digital services. No wonder DeepSeek also offers its models as a service there.

For OpenAI and others, it seems pointless to leave that ecosystem, especially when competing models are offered. Developing models without the backing of Amazon, Microsoft or Google — especially if doing so involves using from-scratch techniques that are costlier than DeepSeek’s — is not a viable business for Western AI startups. Meanwhile, in China, the business is still one that privileges imitation. Sometimes, as in the case of DeepSeek, the strategy pays off.

A Geopolitical Chessboard

For China, this outcome is problematic. The Chinese government committed to making China the world’s AI innovation centre by 2030 in the New Generation Artificial Intelligence Development Plan (AIDP) of 2017.^[23] This goal assumes that Chinese tech firms aim to become technology forerunners since China cannot become the world’s AI leader without Chinese companies forging ahead. Meanwhile, Chinese corporate powers remain dependent on the Chinese state’s dexterity to sustain their combined position of fast imitators and China’s market leaders.

In turn, US giants have mobilised a narrative based on the threat of China forging ahead. This has helped them to gather more allies in the US and other governments and is used to justify their calls for state support for AI innovation, given what is presented as a national threat.^[24] The possibility of Chinese expansion also serves US Big Tech strategy to advocate against antitrust. Yet all this looks exaggerated considering that the overall context leads Chinese companies to prefer fast imitation rather than outpacing the US.

So far, the scenario described here represents a win-win situation for US tech giants. They use geopolitical criticality as a weapon to reassure that Chinese giants remain mostly inside China while potentially enjoying not only a more favourable regulation but also additional public investments in AI. Given their current control of the AI value chain, pouring more money into it would favour them.^[25]

The context on both sides of the Pacific has at least two commonalities. First, it underscores the deep interdependence between corporate and political power.^[26] Second, neither the US nor China’s AI development puts social and ecological challenges ahead of private profits.

Digital Sovereignty

Whatever happens in the showdown between the two global powers, DeepSeek-R1’s emergence contributes to convincing policymakers that it is possible to develop AI in the rest of the world even if with smaller budget than the billions of dollars invested by Big Tech. But beyond DeepSeek, the centrality of these technologies for shaping every dimension of our lives from social relations, to how we think, as well as their intrinsic economies of scale, economies of scope and network effects, means we must push for the development of such AI models as public solutions. This means models funded by the public sector and that are truly open source, not only disclosing the final parametres of the model as Meta does with Llama, but including disclosure of the whole development and data used for training.

Yet the models to be developed should not be aimed replicating or chasing those promoted by US and Chinese companies. Instead, AI models should be developed in interdisciplinary teams that take into account their ecological, social and ethical implications. These teams should include specialists in AI impacts on society, labour, the ecology, human rights and more from academia and civil society. Such a public, yet autonomous institution could assess and recommend to governments which AI models are desirable and for which purposes, thus limiting the push for solving everything with AI models. However, once new foundation models are conceived and developed, requirements at the level of processing power for further iterations and adaptation of those models will be lower by following distillation techniques along the lines of DeepSeek.

The chance to build cheaper frontier models is an invitation to work internationally on technological developments that prioritise social needs and respects planetary boundaries by only developing AI models for needed uses where other — less resource-intensive — alternatives are not available. Such models should integrate a broader public-led international, democratic, people-centered and ecological value chain for digital technologies where the fundamental segments of the chain are offered as public goods, including physical infrastructure and fundamental platforms for the development of specific solutions.^[27]

There is no time to waste. We are already witnessing the effects of the loss of public deliberation spaces. Social media’s AI algorithms decide what messages are shown, where, when, and, indeed, which ones are not shown at all. From there, a handful of companies influence public opinion (and common sense). Musk has gone as far as announcing that his platform X will be the only channel used for the Social Security Administration’s public communication.^[28] Even if this will most likely not happen, the mere possibility speaks the degree of the problem.

Social media is but the tip of the iceberg. Digital dependency is expanding to every aspect of life as AI and other digital technologies offered as cloud services penetrate every business, organisation and even states. The capacity to govern, to be sovereign, is hijacked when the technologies essential to the economic and political functioning of a country are controlled by a handful of companies from the world’s superpower. I have argued here that they will not be outpaced by Chinese companies any time soon. Even if they were, this would not solve the deep global dependencies generated once essential information technologies are under private control of a few companies from two countries.

Overcoming this dependency is urgent not only because of the resulting economic losses. Big Tech is defining what technology we get and therefore what developments are neglected — which tech we do not get. Without redirection that creates and allows for public, open, auditable models, we risk continued tech development that favours uses ranging from military or surveillance control (in and out of workplaces) to AI that replaces creative work rather than making work more fulfilling. Calling instead for a public yet international and open solutions^[29] that should at least be used in schools, hospitals and by the public administration is today, more than ever, extremely urgent.