The Myth of “Cheap” AI Training Just Got Busted (DeepSeek Case)

[cryptoenthusiast]

Remember back in early 2025 when Chinese company DeepSeek dropped that wild claim — saying they trained a massive AI model for under $6 million? Everyone from investors to AI engineers had their jaws on the floor.

It didn’t just stir the AI scene — it rocked the global stock market:

NVIDIA lost nearly 17% in one day (about $600 billion wiped from its value 🤯).

Tech giants like AMD, Google, ASML and Siemens Energy followed with heavy drops.

The NASDAQ took a -3% hit, and the U.S. tech sector saw around $1 trillion in market cap vanish.

Why the chaos? Because if DeepSeek really trained a competitive model for that cheap, it meant all the billions going into chips and AI infrastructure might’ve been wasted.

But here’s the kicker — it turns out that "$6M" number? Totally misleading.
SemiAnalysis Debunks the Fantasy

Independent research firm SemiAnalysis dug deep into DeepSeek’s actual operations and revealed:

🧾 Real costs are over $1.3 BILLION, not $6M.

Here's what DeepSeek didn't count in their $6M claim:

$6M was only for one training run, not the full training lifecycle (which includes dozens of runs, testing, tuning, etc.)

They didn’t include:

    R&D costs

    Architecture experiments

    Synthetic data generation

    Electricity, cooling, and operational costs

    Salaries of top-tier AI researchers (some earning over $1.3M/year)

    Cost of buying and running tens of thousands of GPUs

Basically, the $6M number was cherry-picked to create a narrative — and it worked for a while. But not anymore.
🧠 Big Picture: AI Still Costs Billions

SemiAnalysis compared this to companies like Anthropic (Claude 3.5 Sonnet), where even just pre-training costs tens of millions. Full stack costs? Billions.

So yeah — the myth of “cheap” AI is fun for headlines, but not rooted in reality.

If you’re into AI, startups, or just watching tech bubbles form and pop — this whole DeepSeek drama is a wild case study in perception vs. reality.

Show some love if this breakdown helped — and follow me for more deep dives like this!
#WEB3 #Crypto #Coin #AI #DeepSeek