How DeepSeek is Preventing AI from Replacing Humanity

Wil Smith in a still from I, ROBOT 2004.

This dystopian nightmare—where a billionaire monopolizes 99% of Earth’s land, water, and even air, while billions of impoverished people are confined to tiny homes surviving on their own waste, and anyone who resists will be killed by robots and drones—comes from 刘慈欣(Cixin Liu), China’s most renowned sci-fi writer. Best known for The Three-Body Problem, which earned him the Hugo Award and was praised by Obama. However, few Western media have paid attention to Liu’s earlier short story, For the Benefit of Mankind, which paints an chilling vision.

Unlike Western sci-fi, where robots are bound by the rule “do not harm humans,” Liu’s creations prioritize “the inviolability of private property,” granting them license to kill the poor to protect resources hoarded by the wealthy. This grim outlook stems from a pre-DeepSeek era when Silicon Valley dominated AI development—a trajectory already unfolding in the U.S. In 2021, Google disclosed that its U.S. data centres consumed 4.34 billion gallons of water—freshwater effectively diverted from American communities.

Take Arizona, one of the driest states, with annual rainfall just half of the national average. Yet following Apple, Microsoft and Meta, Google planning a massive data centre in Mesa, demanding 1-4 million gallons of water daily for cooling—equivalent to the daily usage of 27,000 local residents. In 2023, Business Insider revealed Arizona’s groundwater faces a 4% annual deficit, projected to take a century to recover. Despite this, the state didn’t restrict tech giants’ “water-guzzling beasts,” but the development of Phoenix.

Aerial view of the Apple Data Center in Mesa near Phoenix, Arizona

Silicon Valley giants not only threaten the American people’s right to development but also their right to survival.

In 2019, during a severe drought in South Carolina, Google’s Berkeley County data centre drained 1.5 million gallons of groundwater daily while siphoning 5 million gallons from municipal supplies—tripling the region’s safe water consumption limit. Local residents protested, but the South Carolina Department of Health and Environmental Control approved Google’s request. Ironically, the same department ordered the utility company Mount Pleasant Waterworks, responsible for supplying water to 45,000 residents, to reduce its usage by 57% over the next four years.

These data centres could originally be relocated to underwater or high-latitude regions for natural cooling. Instead, they cluster near urban areas to shave milliseconds off-network latency, evaporating billions of gallons as collateral damage.

Chinese strategist, Prof. Wang Xiangsui warns of an emerging “quasi-human society” dominated by AI and robots. If governed by profit-obsessed corporations or individuals, clashes over resources between humans and this artificial society could escalate into full-scale conflict. While water scarcity now grabs headlines, energy poses a longer-term crisis.

Elon Musk’s Grok3, powered by 100,000 NVIDIA H100 chips, consumes up to 28% more annual electricity than New York City. This electricity power is enough for 8.2 million people, yet its daily U.S. user base remains under 1 million, largely due to its $40/month subscription fee – According to the US Federal Reserve System, 37% of Americans are not even able to cover $400 unexpected expenses. Let alone OpenAl’s ChatGPT-4.5, which charges $150 per million tokens.

Elon Musk powers new ‘World’s Fastest AI Data Center” with gargantuan portable power generators, rising air pollution concerns in Memphis

In contrast, DeepSeek R1 costs just $2.19 for the same output—democratizing AI access for all. Prof. Wang points out; the more important point is DeepSeek offers humanity an alternative to self-destruction. Autonomous AI could evolve into an equal partner or a lethal adversary, akin to Oedipus in Greek myth. Its trajectory hinges on its foundational code: profit-maximization for elites risks creating a “quasi-human” society that overthrows humanity.

Fortunately, DeepSeek recently open-sourced 7 technologies addressing a core challenge—training powerful AI without expending more resources. This could drastically reduce the water and energy conflicts between humans and quasi-human societies.

Traditional AI training wastes energy by processing data in batches. DeepSeek’s DualPipe allows simultaneous data intake and training, cutting runtime by half and saving 30% energy and water. Innovations like DeepEP, DeepGEMM, and 3FS boost server efficiency, enabling clusters to support 4-7 times more users. FlashMLA democratizes access by empowering low-cost chips to rival premium hardware, narrowing the global digital divide. Even scheduling tweaks—prioritizing user services by day and AI training at night—shrink infrastructure demands.

To sway profit-driven U.S. firms, DeepSeek revealed the staggering profitability of these open-sourced technologies: a 545% margin, earning $562,027 daily, theoretically $200 million annually. On the other hand, according to CNBC, OpenAI lost $5 billion in 2024.

Overnight, OpenAI and NVIDIA stocks plunged over 10%. European banks now hail Chinese tech as a historic investment. Prof. Wang believes this is “humanity’s greatest self-rescue mission,” DeepSeek need not top every single AI performance chart. Its mere existence as an open-source rival blocks tech oligarchs from realizing Cixin Liu’s nightmare. Originally, Silicon Valley envisioned AI as a first-class ticket for elites to soar above the masses. DeepSeek, however, adds economy seats—allowing ordinary people to ascend alongside billionaires.

The battle between open-source collaboration and closed-profit models may never end, resembling quantum tides in perpetual flux. Yet this dynamic equilibrium sustains hope. So long as 99% of humanity can still afford open-sourced AI like DeepSeek, it marks a victory for China’s millennia-old ethos of “deep seeking.”

In March, 2025, as global capital fuels a new future, the “quasi-human society” remains unborn, but the crown of computational hegemony is already trampled underfoot. The open-source president ascends the palace, steps onto the balcony, and in his face, every person sees their own reflection.

Editor: Charriot Zhai