Why Nvidia Fell 17% on Monday, January 27, 2025 — and How Long It Stayed Down

Archive note: this retrospective was researched and first published in August 2026. The event time marks MoveSurge's Monday, January 27 U.S. market-wrap event.

Selloff-day reference: Monday, January 27, 2025 is day zero in this timeline—the session when Nvidia fell 16.9%, commonly rounded to 17%.

Nvidia closed at $118.42 on Monday, January 27, down 16.9% from $142.62 on Friday, January 24, three calendar days earlier. Roughly $589 billion of market value disappeared in one session. The Nasdaq fell 3.1% and the S&P 500 lost 1.5%, while the Dow gained 0.7%. That split identifies the episode as a concentrated repricing of the AI infrastructure trade rather than an equal selloff across every sector.

What investors priced on Monday, January 27

The weekend narrative combined four observations: DeepSeek had released a capable open-weight reasoning model seven calendar days before the drop; an earlier paper documented efficient V3 training; the app had become the top U.S. free download one day before the drop; and U.S. technology companies were preparing very large data-center budgets. The market converted those facts into a stronger conclusion that cheaper models could strand planned spending and reduce demand for high-end accelerators.

The conclusion rested partly on a faulty comparison. DeepSeek's $5.576 million figure measured the official V3 training work at an assumed compute price and excluded prior research and experiments. It was not the all-in cost of R1, the company or its infrastructure. The primary-source cost audit documents the boundary.

The price path after the historic drop

Did DeepSeek keep Nvidia under pressure for months?

The price statement is accurate with an important qualification. Nvidia remained below its Friday, January 24 close through Friday, June 6—130 calendar days after the drop—and first exceeded it on Monday, June 9, 133 days after the drop. The path was not continuous: the shares nearly recovered by Tuesday, February 18, 22 days after the selloff, then entered another decline.

Assigning the entire February-to-April weakness to DeepSeek would be incorrect. Later sessions incorporated Nvidia's earnings and Blackwell transition, new U.S. controls on H20 sales to China, tariff shocks and a broad contraction in growth-stock valuations. DeepSeek remained part of the debate about returns on AI capital spending, but it was one factor in a changing set of risks.

What the spending data showed by summer

Nvidia reported first-quarter fiscal 2026 revenue of $44.1 billion on Wednesday, May 28, 121 calendar days after the drop, up 69% from a year earlier. Data Center revenue was $39.1 billion, up 73%. On Wednesday, July 23, Alphabet raised its expected 2025 capital spending from $75 billion to $85 billion and cited strong cloud demand plus investment in servers and data centers. Nvidia became the first public company to reach a $4 trillion market value on Wednesday, July 9.

Those outcomes did not prove that every AI project would earn an adequate return. They did contradict the weekend's most extreme version of the thesis that efficient models had made infrastructure spending obsolete. By summer, investors were separating lower unit costs from total demand for training, inference, networking and power.

What can be claimed about causality

Contemporaneous reports consistently identified DeepSeek as the catalyst for the Monday, January 27 repricing, and the sector pattern matched that explanation. Public evidence cannot allocate the 16.9% move among social-media reach, television coverage, analyst notes, crowded positions and automated deleveraging. The responsible conclusion is that DeepSeek supplied the trigger and a simplified cost narrative increased its force.

The post-by-post amplification audit shows which claims were supported, misleading, unsupported or opinion.

How to read the normalized price-and-volume chart

The main chart rebases both price series to 100 at Nvidia's Friday, January 24 cash-session close—three calendar days before the drop—then plots exact common five-minute regular-session timestamps. Overnight gaps are removed from the horizontal axis, so the chart compares relative movement instead of absolute price levels. The red series is NVDA from a validated trade-bar series. The blue series is a bid-priced US100 cash-CFD proxy. It is not CME NQ, the official NDX index, QQQ, the Nasdaq Composite or exchange volume.

In the validated broker series, Nvidia's last one-minute regular-session price fell from $142.60 on Friday, January 24—three calendar days before the drop—to $118.51 on Monday, January 27, the selloff day: −16.8934%. The supplied normalized NVDA share-volume series for Monday was 530,229,066 shares, 3.85× the average of the prior five sessions. The volume panel uses NVDA shares only. US100 activity is broker tick count and is deliberately excluded from a shares comparison.

The official daily-close history used in the recovery table reports $142.62 and $118.42 for those two dates. That small difference is expected: one series uses classified one-minute broker bars and the other uses an official-style daily close history. The article keeps the two datasets labeled instead of splicing them into one false precision.

Qualitative events aligned with the market window

The arrows establish chronology, not causality. A social post can be market-relevant because it gives investors a shared frame; the chart cannot infer which account caused which trade.

What the event-study statistics add

Across Monday's exact common five-minute regular-session bars, the Pearson correlation between NVDA and the US100 cash-CFD proxy was 0.5747 using log returns. A pre-event daily beta of 1.2683 produces an estimated Monday abnormal return of approximately −12.48%. The calculation says Nvidia underperformed the technology proxy by far more than its earlier co-movement would imply. It does not prove that DeepSeek explains the entire residual.

The proxy contains Nvidia exposure, making it an imperfect and non-independent control. Correlation also changes with window, session classification and return definition. These statistics are descriptive triangulation alongside the sector pattern and contemporaneous reporting.

Why “pressure for months” needs qualification

The official-style price history stayed below the Friday, January 24 close through Friday, June 6—130 calendar days after the drop—and first exceeded it on Monday, June 9, 133 calendar days after the drop. That statement describes the price path. It does not assign every intervening decline to DeepSeek. Nvidia was already within 2.3% of the baseline on Tuesday, February 18—22 calendar days after the drop—before earnings, Blackwell margins, H20 export controls, tariffs and a wider growth-stock reset changed the risk set.

A clean aftermath analysis therefore separates three questions: whether DeepSeek triggered the Monday repricing; whether efficient models changed investor beliefs about AI capital intensity; and which later events drove the February-to-April path. The first has strong contemporaneous support. The second remained an active thesis. The third requires a multi-catalyst timeline and cannot be answered by extending Monday's label across every later bar.