Elon Musk predicts AI will surpass humans in 2027! The computing power bottleneck lies in electricity, not chips.

Elon Musk’s Davos prediction: AI will surpass humans next year, and by 2030, the total intelligence will exceed all of humanity. The core argument is that the bottleneck in computing power is electricity, not chips. The solution is space data centers, with 24-hour solar energy and natural cooling at -270°C. SpaceX’s Starship reduces transportation costs by 100 times; building data centers in space is cheaper than in Silicon Valley.

Radical Prediction: AI Surpassing Humans Next Year

While everyone is watching Tesla’s stock fluctuations or Musk’s Twitter spats, he’s actually plotting a big chess game. This time in Davos, Musk didn’t just talk about cars or rockets; he connected AI, energy, robots, and space into a logically rigorous “survival chain.” After hearing this, you’ll realize it’s not just a tech forecast but more like a “Human Civilization Survival Guide.”

To understand what Musk is tinkering with, you first need to grasp his “ultimate anxiety.” In many occasions, Musk positions himself as a “civilization guardian.” In his view, consciousness and life are not guaranteed in the vast universe but are tiny “candles in the dark,” extremely fragile, and could be extinguished at any moment due to war, climate, or AI out of control. Therefore, all his companies serve two purposes: SpaceX (backup civilization)—in case Earth’s light goes out, we need a “backup” on Mars; Tesla (continuation of civilization)—on Earth, to make life better and longer with clean energy and AI.

Under this logic, AI and robots play a critical role. Musk firmly believes that the only way to solve poverty and achieve global prosperity is to reduce production costs to near zero. Once humanoid robots (like Optimus) become widespread, labor will no longer be scarce, and goods and services will be greatly enriched. His prediction for AI evolution is perhaps the most radical: by the end of this year, Tesla’s Optimus robots will be working in factories doing complex tasks; by next year (or at the latest, the year after), the intelligence level of a single AI will surpass any individual human; around 2030, the total intelligence of AI will surpass all of humanity.

“Surpassing humans next year”? I know many will say this is “Elon Musk’s timeline,” just listen. But looking at the speed of AI evolution over the past two years— from GPT-3 to GPT-4, from Claude to o1, from text to video generation— this time, even skeptics dare not dismiss him lightly. Google’s Gemini, Anthropic’s Claude, OpenAI’s o1 are rapidly approaching or surpassing human performance on specific tasks. In Go, protein folding, mathematical proofs, coding, AI has already exceeded human experts in many fields. When Musk talks about “surpassing,” he might be referring to the critical point of Artificial General Intelligence (AGI).

The True Bottleneck in the Computing Power War is Electricity

The climax is here. Since AI is so powerful, we should just go crazy building chips and data centers, right? Musk pours cold water: don’t just focus on chips; the road is blocked. He points out that AI computing power is exploding exponentially, but Earth’s electricity supply simply can’t keep up. “Soon, we will produce far more chips than current electricity can support.” In simple terms: brains are available, but no power.

On Earth, solar energy is a good solution (he cites an example: a 160x160 km solar farm could power the entire US, and China’s capacity in this area is unbeatable), but for the future giant AI monster, it’s not perfect enough. Problems with ground solar include: day-night cycle causing intermittent power, weather impacts (clouds, rain), atmospheric absorption leading to about 30% efficiency loss, and land costs.

Currently, global data centers consume about 1%-2% of total electricity, but with the explosion of AI training and inference demands, this ratio could rise to 10%-20% by 2030. Tech giants like Meta, Google, Microsoft are rapidly expanding data centers, but power supply has become a bottleneck. Some US regions even see “data centers waiting in line for grid expansion,” with new centers waiting 3-5 years for sufficient power quotas.

More seriously, cooling is a major issue. AI chips (like NVIDIA H100) have extremely high power density; a single rack can consume 40-60 kW, equivalent to dozens of households. This high heat density requires massive cooling systems, which in turn consume large amounts of electricity, creating a vicious cycle. On Earth, data centers typically use water cooling or air cooling, but these are inefficient and costly under extreme climates.

Space Data Centers: The Physics Cheat Code

To solve the “power shortage” and “cooling” physics problems, Musk offers a sci-fi-like ultimate solution: move AI data centers to space. Sounds crazy? But in physics, this is a perfect cheat code.

Infinite energy: in space, solar panels have no night or weather, and no atmospheric interference; efficiency is 5 times that of ground-based panels, providing 24/7 full load power. In geostationary orbit, solar panels always face the sun, generating about 1.4 kW per square meter (vs. 0.2-0.3 kW on Earth). A football field-sized solar array in space can produce power equivalent to 5 football fields on the ground.

Natural cooling: AI generates huge heat, but in space shadows, temperatures approach absolute zero (-270°C), providing free and perfect “data center air conditioning.” Data centers only need to design radiators to emit heat into deep space, with no active cooling systems needed. This passive cooling is zero-cost and far more efficient than any ground solution.

Musk asserts: within two or three years, space will become the lowest-cost place to deploy AI. The key is SpaceX’s Starship. As long as rockets can be fully reusable like airplanes, transportation costs can drop 100 times. Then, building a supercomputing center in space could be cheaper than in Silicon Valley.

Three Major Physical Advantages of Space Data Centers

Energy density 5x: Space solar energy 24/7 full load, 5 times the power per unit area of ground panels

Zero-cost cooling: -270°C environment provides passive radiative cooling, no air conditioning needed

Transport costs plummet: Fully reusable SpaceX Starship reduces launch costs from $10,000 per kg to $100

Current Starship tests are ongoing, but SpaceX has successfully completed multiple test flights. If fully reusable by 2027-2028 as planned, launch costs will hit historic lows. Launching 100 tons of AI servers into orbit might cost only $10 million, while building an equivalent data center on the ground could cost over $100 million in land, construction, and power infrastructure.

The Ultimate Closed Loop: SpaceX AI on Mars

This is the closed loop: use SpaceX’s technology to reach space, solve AI’s energy bottleneck; use AI to build robots, establish bases on Mars and Earth; ultimately, let human civilization’s “candle” continue across star systems. This is not four separate projects but a complete systemic engineering.

Tesla’s Optimus robots play the role of “executors” in this loop. Musk reveals that by the end of this year, Optimus will be working in factories doing complex tasks. When robots can build bases on Mars, assemble data centers in space, and produce more robots on Earth, the entire system will enter an exponential phase of self-replication and expansion. Humanity only needs to provide initial commands and supervision; robot armies will complete 99% of physical labor.

At the end of his speech, Musk said something interesting: “I encourage everyone to stay optimistic and excited about the future… being an optimistic person who is proven wrong is better than being a pessimist who is proven right.” In this era of uncertainty, perhaps this is the best attitude we can have toward AI transformation.