Member-only story
DeepSeek rains on the AI hype parade
Uh-Oh. Who could have predicted “AI” is a dotcom bubble?
As I write this, Wall Street traders are in meltdown, running around in circles with their hair on fire. Panicked individuals, especially those who had invested heavily into AI, were shouting into phones: “Sell! Sell! Sell! AaaAA!!” These live scenes reveal that awareness is dawning that the sure-thing AI boom maybe a bust.
As a Reuters report puts it:
“In the last session, Nvidia lost $593 billion in market value — a record one-day loss for any company, while shares of companies in semiconductor, power and infrastructure companies exposed to AI collectively shed more than $1 trillion.”
Around 20th January, 2025, Chinese-made artificial intelligence (AI) model DeepSeek released its latest version and the performance quickly impressed AI experts. With word spreading, the attention of the entire tech industry and the world on the app, it has now shot to the top of Apple Store’s downloads, stunning investors and sinking some tech stocks.
What makes DeepSeek special is the company’s claim that it was built at a fraction of the cost of industry-leading models like OpenAI (and much more scarily if you are an NVIDIA stock investor): it uses fewer advanced chips. The core problem (pardon the pun) as James Vincent in The Guardian points out:
The system is thought to be 95% cheaper than OpenAI’s o1 and uses one tenth of the computing power of another comparable LLM, Meta’s Llama 3.1 model. To achieve equivalent performance at a fraction of the budget is what’s truly shocking about R1, and it’s this that has made its launch so impactful.
Tech investors aren’t the only ones now in hot water. What is emerging is that there’s no real case for gigantic sums of money to be sunk to create AI’s “critical infrastructure” (for example, “Stargate” — the Trump administration’s $500 billion AI-fund).
