NVIDIA Feb 25 Earnings: The Most Telegraphed Beat in Years

I don't say this often, but NVIDIA's February 25 earnings might be the most telegraphed beat I've seen in years. Not because Jensen's going to pull a rabbit out of his leather jacket — but because the CFO already showed us the rabbit, the hat, and the entire magic act last quarter.

The setup is almost too clean. When a company gives you visibility to half a trillion dollars in revenue through 2026, tells you demand is "off the charts," and then hyperscalers proceed to announce $600-700 billion in capex plans with 75% earmarked for AI infrastructure — you're not playing poker anymore. You're watching someone count chips they've already won.

The tape is saying one thing: this isn't about if they beat. It's about by how much, and whether the Street has priced in enough upside. Spoiler: they haven't.

The Setup: What the CFO Already Told Us

Let's start with what we know. Last quarter, NVIDIA posted $57 billion in revenue — up 62% year-over-year, with $51.2 billion coming from Data Center alone. That's a business printing money faster than the Treasury.

But here's the kicker: CFO Colette Kress didn't just guide to $65 billion for Q4 (which already beat the $61.66 billion consensus). She gave the Street something most CFOs would never dream of — actual visibility into 2026. "Half a trillion dollars in Blackwell and Rubin revenue through the end of 2026," she said. Not a forecast. Not a hope. Signed contracts.

That $500 billion backlog isn't speculative. It's booked. And it doesn't even include the new contracts that have rolled in since last earnings. This is like sitting down to a poker game and seeing your opponent's cards face-up on the table. The only question is whether you're willing to bet on it.

Blackwell is shipping roughly 1,000 racks per week, sold out until mid-2026. Jensen said in January that demand is still "off the charts," with a 3.6 million GPU backlog. The bottleneck isn't demand — it's how fast they can stamp silicon and ship boxes.

The Hyperscaler Arms Race: $600-700 Billion Says It All

If you want to know why NVIDIA's going to beat, don't look at NVIDIA. Look at their customers.

The Big Five hyperscalers — Amazon, Google, Microsoft, Meta, Oracle — are spending $600-700 billion on infrastructure in 2026. That's a 36% increase from 2025. And 75% of that spend is targeting AI infrastructure directly. Not cloud storage. Not CDN upgrades. AI compute.

NVIDIA holds 85-90% market share in AI accelerators. Do the math. If 75% of $700 billion is going to AI infrastructure, that's $525 billion. NVIDIA's getting the lion's share of that. Not all of it — there's networking, power, cooling, real estate — but when you own the GPU market the way NVIDIA does, you're first in line at the trough.

This isn't a land grab. It's a space race. And NVIDIA is NASA.

New Contracts: The Backlog Gets Fatter

Here's where it gets interesting. The $500 billion backlog Kress cited last quarter? That was before the new deals started rolling in. Since that earnings call, NVIDIA has locked in contracts that would make a defense contractor blush.

Nscale + Microsoft: 200,000 GB300 GPUs. At roughly $30,000 per chip (conservative estimate), that's $6 billion right there.

CoreWeave: NVIDIA invested $2 billion in exchange for 5GW of AI compute capacity. This isn't just a sale — it's a strategic partnership to lock in supply for the next wave of training clusters.

Saudi Arabia: A 600,000-GPU sovereign AI project. The Saudis aren't building a data center — they're building a digital nation. At scale, that's $18 billion in GPUs alone, not counting networking and support infrastructure.

Anthropic: Up to 1 GW of Grace Blackwell and Vera Rubin systems. That's enough power to run a small city, dedicated entirely to training the next generation of Claude models.

OpenAI: At least 10 GW of AI data centers. Sam Altman isn't thinking in terms of model releases anymore — he's thinking in terms of gigawatts. And every watt needs silicon.

AWS-Humane: Up to 150,000 AI accelerators for the next-gen wearable AI push. Even the weird bets are massive.

China: The Priced-In Boogeyman

Let's talk about the elephant in the room: China export restrictions. The H20 chip faces potential bans, and even the approved H200 comes with a 25% revenue-sharing requirement with Chinese partners. Analysts are saying NVIDIA's China revenue could drop from 66% of international sales to just 8%.

Sounds bad, right? It would be — if that $500 billion backlog was dependent on China. It's not.

The bulk of NVIDIA's backlog is U.S. hyperscalers (Amazon, Google, Microsoft, Meta), sovereign AI projects (Saudi Arabia, UAE, Europe), and enterprise customers in friendly jurisdictions. China was gravy. Losing gravy hurts the margins, but it doesn't kill the meal.

And here's the thing about regulatory risk: the market already knows. This isn't new information. The H20 ban threats have been circulating for months. The Street has had plenty of time to price it in. If anything, the resilience of NVIDIA's stock in the face of China headwinds tells you the big money isn't worried.

Back when I was running the book, we had a saying: "If everyone's talking about the risk, it's already in the price." China is the risk everyone's talking about. Which means it's probably the risk that matters least.

The Tape: What the Charts Say

All right, let's see what the technicals are telling us. Our engine gives NVDA a 6/10 NEUTRAL score — "Wait for setup." Not exactly a ringing endorsement, but let's dig into the details.

The stock is sitting right at resistance ($184.46), which is why the algo is saying "wait." But the patterns tell a different story. The engine detected a Bull Flag with 70.2% confidence and a Falling Wedge at 68.7%. Both are continuation patterns — meaning the next move is likely up, not down.

The Fibonacci 0.618 retracement sits at $185.53, just 0.63% above current price. That's a key level. If NVDA can punch through that on volume, the next stop is $200 — which just happens to be where the call wall sits. Coincidence? I don't think so.

RSI at 48 means we're not overbought. MACD is positive at +0.18, showing upward momentum. The put wall at $175 is your downside backstop — that's where the big money is defending. The setup isn't screaming "buy right now," but it's coiled. Pre-earnings consolidation before the breakout.

The Flow: Follow the Money

If you want to know what the smart money is thinking, don't listen to CNBC. Look at the flow.

$7.2 million in call premium versus $137,000 in put premium. That's not a bet — that's a conviction trade. 98% bullish flow. The only puts showing up are likely hedges or short-dated scalps.

The top trade is a $160 call expiring September 2026 — $417,500 premium. That's not a lottery ticket. That's someone with a six-month horizon betting NVDA is going significantly higher. And they're not alone. Multiple $190 calls expiring March 20 came through at $300-405K each. Those expire after earnings, which tells you what they're expecting.

Dark pool VWAP at $184.38 is right in line with the current price. Institutions aren't front-running — they're accumulating quietly. When the dark pool and the tape agree, you pay attention.

Only one sweep detected in recent flow. That's interesting. Sweeps usually signal urgency — someone willing to pay up across multiple exchanges to get filled immediately. The absence of sweeps tells me the big money isn't panicking to get in. They're patient. They know the setup. They're waiting for confirmation, just like you should be.

Valuation: Still Cheap for the Best Seat in Town

Here's where it gets absurd. NVIDIA is trading at a trailing P/E of 46.39. For a stock growing at 62% year-over-year, that's not expensive — it's a gift. The five-year average P/E is 65.25. We're trading below the historical mean for a company that's accelerating, not decelerating.

Forward P/E for FY2027 is 24.24x, based on an EPS estimate of $7.66 — which assumes 63% earnings growth. That gives you a PEG ratio of 0.38. Anything under 1.0 is considered undervalued. NVIDIA's PEG is 0.38. That's not just cheap — that's mispriced.

The Street consensus from 39 analysts is "Strong Buy," with an average price target of $255.82. That's 37% upside from current levels. Goldman Sachs expects $67.3 billion in revenue this quarter versus the $65.5 billion consensus — a $2 billion beat. UBS is even more aggressive at $76 billion. Someone's smoking something, but the direction is clear: upward revisions.

You're getting 63% earnings growth for less than 25x forward earnings. Find me another company with a $500 billion backlog, 85% market share in the fastest-growing sector in tech, and a PEG ratio under 0.4. I'll wait.

The Risks (Because I'm Not Reckless)

Let's not pretend this is a slam dunk. There are real risks here, and you'd be a fool to ignore them.

Tariffs and Trade Policy: The current administration's unpredictable stance on China could escalate. New tariffs, supply chain disruptions, or retaliatory measures could hit NVIDIA harder than expected. Risk level: medium-high.

Competition from Custom ASICs: Google's TPU, Amazon's Trainium, and other custom chips are gaining ground in inference workloads. Analysts estimate ASICs could capture 15-25% market share by late 2026. That's not an existential threat for training workloads (where NVIDIA dominates), but it could compress margins in inference. Risk level: low-medium near term.

AMD's MI300X: AMD's latest chip has a 32% hardware advantage in some benchmarks. But here's the dirty secret: hardware specs don't matter if the software sucks. CUDA delivers 10-30% better real-world performance because the ecosystem is mature. AMD is five years behind on software, and that gap isn't closing anytime soon. Risk level: low.

Margin Compression: As competition heats up and hyperscalers negotiate harder, NVIDIA's gross margins could come under pressure. The company has guided to stable margins, but that's always the first thing to crack when supply meets demand. Risk level: medium.

Supply Chain (CoWoS, HBM): NVIDIA's chips rely on advanced packaging (CoWoS) and high-bandwidth memory (HBM), both of which have limited suppliers. TSMC and SK Hynix are expanding capacity, but bottlenecks could constrain shipments. Risk level: medium.

Hyperscaler Cash Flow Constraints: If one or more hyperscalers hit a wall on capex — whether due to shareholder pressure, debt concerns, or slower cloud revenue growth — NVIDIA's backlog could start to slip. So far, no signs of that happening. Risk level: low-medium.

The Play: How I'd Approach It (Not Advice, Just My Read)

If I were still running the book, here's how I'd play it. This is not financial advice — I'm not your advisor, I'm a guy with opinions and a keyboard. But if you asked me over a bourbon what I'd do, this is what I'd tell you.

The Setup: NVDA is sitting at $184.35, just below resistance at $184.46. The Fib 0.618 level is $185.53. The call wall is at $200. Put wall at $175. Pre-earnings consolidation with bullish continuation patterns (Bull Flag, Falling Wedge).

The Trade: I'd be looking at the March 20, 2026 $190 calls (expiring after earnings on Feb 25). Current premium is in the $300-405K range for block trades, so retail would be paying less for smaller size. The strike is 3% out of the money, giving you leverage without being reckless.

The Thesis: If NVDA beats and raises (which I expect), the stock breaks through $185.53, clears the $190 strike, and runs toward the $200 call wall. That's a 8.5% move from current levels — easily achievable on an earnings beat with a guidance raise.

Risk Management: Don't bet the farm. Earnings are binary events. Allocate 2-5% of your options portfolio, max. Set a mental stop at -30% if the trade goes against you. If NVDA gaps down on a miss (unlikely but possible), you want to cut losses before theta decay kills you.

Alternate Play (Lower Risk): If you're not comfortable with naked calls into earnings, consider a call debit spread: buy the $190 call, sell the $200 call. You cap your upside at $200, but you reduce your cost basis and define your max loss. Less sexy, but smarter for most people.

Longer-Term Play: The $160 September 2026 calls are where the real money is hiding. Someone paid $417.5K for that position. That's a six-month bet on NVDA being significantly higher by fall. If you've got the capital and the patience, LEAPs are the way to play secular trends, not quarterly earnings lotteries.

Marcus's Read: The Bottom Line

I've been trading long enough to know that "sure things" don't exist. But I also know when the odds are stacked in your favor. And right now, the odds are screaming NVIDIA beats on Feb 25.

The CFO gave you half a trillion dollars in visibility. The hyperscalers announced $700 billion in capex. The backlog grew after last quarter's guidance. The flow is 98% bullish. The dark pool is aligned. The valuation is cheap (PEG 0.38). The patterns are bullish (Bull Flag, Falling Wedge). The Street consensus is already pricing in a beat, but not a big beat.

China risk? Priced in. Competition? Years away from mattering. Supply chain? They're shipping 1,000 Blackwell racks a week. Margin compression? Not when you're sold out until mid-2026.

The only question is whether you believe the CFO. I do. When someone tells you they have $500 billion in signed contracts, you don't fade that. You lean into it.

This isn't a "buy and hope" situation. This is a "buy because the math works" situation. The tape is telling you what's coming. The flow is telling you what the big money thinks. The fundamentals are telling you the backlog is real. The technicals are telling you the breakout is setting up.

I'm not saying mortgage the house. I'm saying pay attention. Because when a company this dominant, in a sector this explosive, with a backlog this large, trades at a PEG ratio this low — you don't overthink it. You act.

Feb 25 is the day the Street finds out whether they priced in enough upside. My guess? They didn't.