DeepSeek's app topped the US App Store and Nvidia lost about $589 billion in a day
Seven days after R1's release, a free Chinese chatbot topped the US App Store and Nvidia lost about $589 billion of market value in a day, the largest single-day loss for any listed company to that…
- Date
- 27 January 2025
- Who
- DeepSeek, Nvidia, US government and markets
- People
- Marc Andreessen, Jeffrey Emanuel, Satya Nadella, Sam Altman, Donald Trump
- Confidence
- High on the market facts; Medium on causal attribution
- Deep dive
- Reasoning II, from o1 and o3 to DeepSeek-R1 and the labs that replicated them
Tier: Landmark · Significance: 4/5 · Org(s): DeepSeek, Nvidia, US government and markets · People: Marc Andreessen, Jeffrey Emanuel, Satya Nadella, Sam Altman, Donald Trump · Confidence: High on the market facts; Medium on causal attribution Primary sources: Fortune on Andreessen's post · Bloomberg on Nvidia's $589B rout (paywalled; confirmed via search extract) · Fortune on Trump
One-liner. Seven days after R1's release, a free Chinese chatbot topped the US App Store and Nvidia lost about $589 billion of market value in a day, the largest single-day loss for any listed company to that date.
Why it happened. R1 was released on a Monday (2025-01-20), the day of the US inauguration, one day before the $500B "Stargate" announcement (Fortune, 2025-01-21), so its first days were absorbed mostly by specialists. What turned it into a market event is better documented than why, and four things stood out. (1) The DeepSeek assistant reached number one among free iPhone apps in the US by 2025-01-27 (TechNode); (2) a viral essay dated 2025-01-25 by Jeffrey Emanuel, "The Short Case for Nvidia Stock" (the essay, which leans on the "just over $5mm" V3 cost and R1's 79.8% AIME score), widely credited with spreading the bear case after Chamath Palihapitiya and Naval Ravikant shared it (Slashdot summary of MarketWatch); (3) Marc Andreessen's 2025-01-26 post calling R1 "AI's Sputnik moment"; and (4) the "$5.6 million" figure, which was V3's final pre-training run, not R1's total (B08-15). I infer that no single trigger did it and that the combination of a consumer-visible product and a simple, partly wrong cost narrative made it legible to non-specialists.
What happened. On 2025-01-27 Nvidia fell about 17% and lost roughly $589B, eclipsing the previous record of about $279B (a 9% fall in September 2024). Broadcom fell 17.4%, the Philadelphia semiconductor index 9.2%, Alphabet 4.2%, Microsoft 2.1% (Bloomberg put the loss at about $589B (search extract); AFP-based reporting says about $593B (Aaj News), which also gives the sector moves). On Monday 2025-01-27 Sam Altman posted on X that R1 was impressive for its price and that OpenAI would move up some releases (Fortune says "on Monday"); Fortune's report of Donald Trump calling it a "wake-up call" for US industry is dated 2025-01-28, as is The Decoder's piece on Altman's post; Satya Nadella argued efficiency gains raise total demand (Jevons paradox) (Fortune, The Decoder).
How it spread (policy and corporate responses).
| Actor | Response | Date |
|---|---|---|
| US Navy | Bars personnel from using the DeepSeek app "in any capacity" | 2025-01-24 |
| Wiz Research | Finds an unauthenticated ClickHouse database with over a million log lines incl. chat history and keys; DeepSeek secured it (Wiz) | 2025-01-29 |
| Microsoft / OpenAI | Probe whether DeepSeek-linked accounts exfiltrated OpenAI API data (B08-14) | 2025-01-29 |
| Anthropic (Dario Amodei) | Essay arguing that DeepSeek strengthens the case for chip export controls (essay) | 2025-01-29 |
| Italy's Garante, NASA, Taiwan, Australia, South Korea | Data-protection order and government-device bans (Al Jazeera roundup) | 2025-01-30 to 02-05 |
| OpenAI | Altman says OpenAI was on the "wrong side of history" on open weights; later releases gpt-oss (B08-35) | 2025-01-31 |
| US NIST CAISI | Evaluation of R1, R1-0528 and V3.1 finds they lag GPT-5 and Opus 4, especially in software and cyber tasks, with agents built on R1-0528 about 12x likelier to follow malicious hijacking instructions and four times as many CCP-aligned inaccurate narratives (CAISI report, which I read in full; it also reports about 94% compliance with overtly malicious jailbreak requests for R1-0528 versus 8% for US reference models) | 2025-09-30 |
Why it mattered. R1 made reasoning models a US-China policy question, and export-control arguments and government-device bans followed within days. It also changed pricing. R1's $2.19 per million output tokens set the reference for what reasoning "should" cost, and incumbents answered with o3-mini and later price cuts (B08-16). The market effect proved temporary. Nvidia was about 1% below its pre-shock close of $142.62 in premarket trading on 2025-02-18 (Sherwood), crossed $4 trillion on 2025-07-09 and became the first $5 trillion company on 2025-10-29 (Yahoo Finance/AP); the large hyperscalers raised their 2025 capex (B17, B18).
Nuance, controversy and myths. "DeepSeek cost $6M and wiped out Nvidia" compresses three errors. The $5.6M is V3's last run only, efficiency gains tend to raise demand for compute, and the stock recovered. Two other claims are also wrong, "DeepSeek was the first Chinese reasoning model" (R1-Lite, QwQ and k1.5 predate or coincide) and "no one in the US saw it coming" (Google's Flash Thinking was cheaper and a month earlier, but did not trigger the same reaction).
Interview kit.
- 30-second version: R1 combined o1-class reasoning, open weights and a very low price, a mix nobody expected. A viral short thesis on Nvidia and a number-one App Store ranking turned that into a $589B one-day loss, which had nearly reversed within three weeks.
- Likely follow-ups: Was the market right to panic? → The efficiency argument was real but the cost figure was misread; demand rose. Who benefited? → Nvidia shareholders, eventually; OpenAI/Anthropic users via lower prices. What did it change politically? → Export-control rhetoric, government-device bans, and later distillation accusations.
- Common mistake: Calling it a "$1 trillion" loss for Nvidia alone; about $589B was Nvidia, the rest the sector.
- Connect it to: B08-10, B08-15, B17, B18, B20.
Sources. All links above; Bloomberg and OpenAI pages not directly fetchable, so figures come from a search extract of Bloomberg plus Aaj News, Fortune, Sherwood and Yahoo Finance.