The Singularity Is Near: How AI Brainpower Surpasses Humanity’s Brainpower in 2025
By: David Johnston, crypto, AI and Web3 technologist
As we enter the era of artificial intelligence (AI), it is becoming increasingly important to quantify the computational capabilities of these systems. Just as “horsepower” or “HP” measures the mechanical power output of engines, we need a term to measure digital machine intelligence, particularly an AI’s ability to process and generate language.
A year ago, these models surpassed the proficiency of high school students, then university graduates, and now PhDs. Regardless of whether it’s true today, or if it becomes true in a few months with the release of Llama 4 or Grok 3, it is becoming increasingly clear that Artificial General Intelligence (AGI) level models are going to be widely available as an open source resource to everyone in 2025.
Just as a person might pause to carefully consider multiple angles of a challenging situation, language models are demonstrating markedly improved reasoning capabilities when given the latitude to generate extended responses and explore various potential solutions. The increase in computational capability starts to become less important than how the capability is applied. The surge in computational capacity will necessitate new ways to manage and direct this power.
But how do we do this?
Also Read: AiThority Interview with Erin LeDell, Chief Scientist at Distributional AI
Efficiently distributing the AI “Brainpower” boom with Smart Agents
The easiest way to do this will be with Smart Agents. These agents will act as intermediaries, translating human intent into actionable tasks for AI systems, namely other third-party agents. The AIs will speak to each other and humans will interact with their representative in this system, the personal Smart Agent, by setting goals, objectives, and preferences, effectively steering the vast AI computational resources available to them towards meaningful work.
The next generation of GPU chips, the Blackwell series, are expected to quadruple this capability for Smart Agents, offering 1,200 “Brainpower” each. By the end of 2025, NVIDIA anticipates producing around 6 million of these GPUs, potentially offering a combined 7.2 billion “Brainpower” if all were operational at 100% efficiency. However, assuming only 70% efficiency, this would still yield around 5 billion “Brainpower” from new hardware alone.
Adding in the existing 3.5 million H100 GPUs from previous years, we estimate an additional 1 billion “Brainpower,” leading to a total of about 6 billion “Brainpower” from NVIDIA’s hardware alone by the end of 2025.
Globally, there are approximately 8 billion people, with 3.5 billion in the labor force. Humans typically provide cognitive labor for about 2,000 hours per year (eight hours/day, five days/week, 50 weeks/year), leading to a total of around 7 trillion hours of human cognitive labor annually.
In contrast, if we consider AI’s “Brainpower” at 6 billion BP, operating continuously, this would translate to around 52 trillion hours of computational work annually. This comparison highlights a vast disparity in available computational versus human cognitive hours starting in 2025.
By the end of 2025, AI will likely provide seven times more computational hours for language tasks than human labor. This isn’t just about quantity, it’s about the potential to augment human efforts rather than replace them.
If technological advancements continue at the current pace, by 2026, we might see another quadrupling of “Brainpower” per chip, leading to even more dramatic figures, and 1,000 times more AI Brainpower than Human Brainpower by 2029.
Of course we don’t have to wait until then. For $30 per Brainpower, times 3.5 billion, we can have the whole of humanities intellectual output for $105 Billion USD, which just happens to be about NVIDIA revenue projection for 2025 for their sold out new GPU series.
In this case, the Singularity is very near indeed.
Also Read: Why Quantum AI is the Next Big Thing for the Future of AI
If the computer was the bicycle for your mind, AI will be the high-speed car
The term “Brainpower” helps us conceptualize the computational capabilities of AI, particularly in language processing, and as LLMs generalize to ever more tasks, this becomes a broader measure of intelligence. The real challenge and opportunity lie in how we harness this power through smart agents to enhance human productivity, creativity, and well-being.
In the 20th century we replaced physical labor with machine-generated horsepower. This is perhaps best exemplified by the freedom provided by owning an automobile, a rite of passage most teenagers now associate with independence and becoming an adult. With 1.5 billion cars produced, and an average of five seats per car, there is basically now a seat in a car available for every human on earth. A car that can carry them 100 to 1,000 times further than they could walk with their own two feet.
The 21st century equivalent is going to be having access to a personal Smart Agent with 1,000 Brainpower available to fuel Agents to complete nearly any task that can be imagined. So our brains will benefit from this digital car, ready to carry us further than our organic brain could unassisted, in the same way physical transportation gave us the ability to cross the physical world with ease.
How we use these new Super Agents is up to us. The path toward the best outcomes is surely to decentralize this awesome power to as many people as possible as quickly as possible.
Comments are closed.