The “DeepSeek Shock” of January 2025
The year began with a massive disruption from an unexpected source. In early 2025, the Chinese startup DeepSeekreleased its R1 model, sending a shockwave through Silicon Valley. What made R1 terrifying to US tech giants wasn’t just its performance—which rivaled GPT-4o—but its efficiency.
DeepSeek proved that a world-class reasoning model could be trained for a fraction of the cost (estimated at under $6 million) compared to the billions being spent by OpenAI and Google. This “efficiency gap” caused US tech stocks to tumble in late January 2025, as investors began to wonder if the massive $100 billion “compute clusters” being planned by Microsoft and Meta were actually necessary.
The $500 Billion “Stargate” Project
In response to the growing AI rivalry with China, the US government—led by the incoming Trump administration—launched the Stargate Project in the spring of 2025. This $500 billion public-private partnership aimed to build a series of “AI superfactories” across the United States over four years.
Backed by Microsoft, OpenAI, and NVIDIA, Stargate was designed to secure American AI supremacy by building massive data centers with their own dedicated nuclear power sources. However, the project immediately hit “The Energy Wall.” By mid-2025, several of these projects were delayed because the US power grid simply couldn’t handle the load. This forced a massive pivot toward Small Modular Reactors (SMRs), making nuclear energy the hottest “tech” investment of 2025.
OpenAI’s “Mid-Life Crisis”
While the infrastructure was being built, the software hit a plateau. OpenAI released GPT-5 in August 2025, but the reception was uncharacteristically lukewarm. While the model was objectively better at coding and math, it didn’t provide the “magic” leap in intelligence that ChatGPT once did.
This led to a major internal restructuring, with OpenAI officially converting into a Public Benefit Corporation to attract more stable, long-term investment. By the end of 2025, the narrative had shifted: the “scaling laws” (the idea that just adding more data and power makes AI smarter) were showing diminishing returns. The industry realized that to get to the next level, AI needed to move out of the data center and into the physical world.
2025 Year-in-Review:
-
The Winner: NVIDIA, which became the first $4 trillion company as every government on Earth raced to build sovereign AI clouds.
-
The Trend: Agentic AI. We stopped talking to chatbots and started letting “agents” handle our emails, calendars, and shopping.
-
The Reality Check: The “Stargate” delays proved that AI progress is now limited by copper wires and transformers, not just clever code.