Artificial intelligence company DeepSeek has unveiled two groundbreaking developments through its new model R1, quietly redefining the economics of artificial intelligence. The model delivered top-tier performance at just 1/40th the cost of previous models and cut training costs by more than 90% in December.
DeepSeek’s advancements have improved both accuracy and efficiency by using chain-of-thought prompting, a technique that allows AI models to narrate their reasoning. Additionally, DeepSeek used AI to generate its own data sets without human labeling, reducing the need for expensive data annotation.
The impact of these advancements is massive. Skyrocketing performance per dollar has significant implications for startups and enterprise adoption, potentially upending market forces in favor of scrappy startups that can now compete with larger companies. Tech giants have collectively invested over $100 billion in infrastructure development but must now determine how to generate returns on their investments.
The landscape has shifted significantly since DeepSeek’s announcement. Startups once struggled to compete with tech giants due to infrastructure spending, but now smaller models can offer comparable capabilities at a fraction of the cost. This shift also reduces staffing costs, allowing startups to develop and refine models more efficiently.
As demand shifts from training-focused hardware to more efficient inference solutions, chip developers face heightened risks. The rise of consumer-grade neural processing units (NPUs) could further accelerate this shift, enabling AI models to run locally on devices like smartphones and laptops.
The implications for the big-tech overlords are bearish. While they have pointed to national-security concerns to justify their spending, domestic researchers have already replicated or outmatched DeepSeek’s standards. Enterprises may now ask whether their massive investments in AI infrastructure have been a waste of money if cheaper alternatives work just as well.
Historical trends suggest most AI advances have come from scale and excessive spending. However, with DeepSeek’s efficiencies, hyper-scalers will need to innovate on new models without ballooning inference costs. Tech giants are already racing to replicate and surpass DeepSeek’s achievements, with open-source models playing a pivotal role in the competition.
AI startups gain leverage by competing with both open-source and closed-source models, allowing them to generate better price-performance ratios while improving their margins. The message is clear: move quickly to harness advancements before market dynamics or compute talent become outdated.
Source: https://fortune.com/2025/02/24/deepseek-shows-ai-startups-can-now-outpace-the-tech-giants-who-may-have-wasted-billions