NVIDIA reports tomorrow and three things changed since last week

NVIDIA drops Q4 earnings on February 25. Wall Street expects $65.5 billion in revenue, $1.52 EPS, and roughly 75% gross margins. Polymarket puts the odds of beating consensus at 94.5%. None of that is the interesting part.

The interesting part is what happened in the last five days that reframes the entire setup. The Supreme Court struck down IEEPA-based tariffs. China's AI labs matched Western frontier capabilities on multiple benchmarks. And DeepSeek just expanded to 1M+ token context windows, proving you can build competitive models on fewer chips than anyone thought possible.

If you hold NVIDIA stock or build on GPU infrastructure, tomorrow's call isn't about whether they beat. It's about what the next twelve months look like after the ground shifted.

The tariff ruling changes NVIDIA's cost structure

On February 20, the Supreme Court ruled 6-3 that Trump's IEEPA-based tariffs were unconstitutional. Chief Justice Roberts was joined by Gorsuch and Barrett from the conservative wing. The ruling means up to $175 billion in already-collected tariffs may need to be refunded to importers.

Trump immediately invoked Section 122 of the Trade Act of 1974, slapping a 15% global tariff as a workaround. But Section 122 has a 150-day limit and narrower legal footing. The legal uncertainty is the point.

For NVIDIA specifically, this matters in two directions. Lower tariffs reduce hardware costs for US data center buildouts. The five largest cloud providers are spending $660-690 billion on AI infrastructure in 2026. Even a few percentage points on hardware import costs moves billions in capex efficiency. But the workaround tariff and ongoing legal battles mean procurement teams can't plan with confidence, which could delay purchase orders.

Watch Jensen Huang's language on the earnings call about supply chain costs and customer order visibility. That's where the tariff impact shows up first.

China closed the gap faster than the models predicted

Three things happened in Chinese AI this month that deserve attention:

DeepSeek expanded to 1M+ tokens. Their context window now matches Claude and Gemini. Six months ago, they were at 128K. The gap went from "generation behind" to "parity" in one update cycle.

Alibaba released Qwen3.5-397B. It's a mixture-of-experts model with 397 billion total parameters but only 17 billion active at inference time. That architecture means competitive output quality at a fraction of the compute cost. Less demand for top-tier GPUs per inference request.

ByteDance's Seedance 2.0 is generating near-commercial quality video, alarming Hollywood studios. Video generation is one of the most GPU-intensive AI workloads. If Chinese labs are achieving this on domestically available hardware, the export control thesis weakens.

The bull case for NVIDIA has always included "China can't compete without our chips." That's looking shakier each month. NVIDIA is still the infrastructure backbone for US and European buildouts. But the total addressable market narrative needs updating, and I don't think Wall Street has fully priced in how quickly Chinese labs are iterating around export controls.

What to actually watch on the earnings call

The headline numbers will be fine. Polymarket doesn't put 94.5% odds on something without reason. Here's where I'm actually focusing:

Q1 FY2027 guidance. Consensus is $70.96 billion. If Huang guides above $75 billion, that's a strong signal that hyperscaler spending isn't decelerating. Below $70 billion would be the first real crack in the AI capex thesis.

Blackwell pipeline commentary. NVIDIA has $350 billion in remaining visibility through 2026. How much of that is firm orders versus soft commitments? The distinction matters because of the tariff uncertainty.

Gross margin trajectory. 75% gross margins are extraordinary. Any compression signals increased competition from AMD's MI350 or custom silicon from Google (TPUs) and Amazon (Trainium). Margin guidance tells you more about competitive pressure than revenue does.

China-related revenue. NVIDIA has been selling cut-down chips (H20, etc.) to Chinese customers under export controls. Any change in that revenue line tells you whether tighter restrictions or Chinese self-sufficiency is winning.

The macro picture isn't helping

GDP at 1.4%. PCE inflation at 2.9%. Mag Seven down 8.8% YTD. Bitcoin cratered to $60K in February's worst crash since 2022.

NVIDIA needs to justify not just its own valuation but the spending decisions of every company buying its chips. If the AI infrastructure buildout is real and durable, NVIDIA's numbers prove it tomorrow. If this is a prisoner's dilemma where nobody can afford to stop spending first, well, someone always blinks eventually. I'm watching this one closely.

Key takeaways

  • NVIDIA reports Feb 25. Q1 guidance above $75B = bullish signal; below $70B = first crack in the AI capex thesis
  • Supreme Court struck down IEEPA tariffs, but Trump's Section 122 workaround creates a 150-day uncertainty window for hardware procurement
  • Chinese AI labs (DeepSeek, Alibaba, ByteDance) are closing the capability gap faster than expected, which complicates NVIDIA's total addressable market story
  • Watch gross margins and Blackwell pipeline specifics, not headline revenue
  • The broader macro environment (weak GDP, sticky inflation, tech rotation) makes this earnings call a referendum on whether the AI spending cycle has legs

References:

  1. NVIDIA Q4 2026 earnings preview: key metrics and expectations -- TradingView
  2. Supreme Court strikes down Trump's IEEPA tariffs in 6-3 ruling -- Reuters
  3. DeepSeek expands context window to 1M+ tokens -- The Verge
  4. Big Tech set to spend $650 billion in 2026 as AI investments soar -- Yahoo Finance