I didn't plan to spend my Thursday night digging through semiconductor supply chain data. But here I am, coffee in hand, staring at TSMC's CoWoS capacity numbers like they're the final episode of a show I can't stop watching.
The charts collapsed earlier this week. And I didn't blink. Because this isn't about the market's mood swings—it's about what happens when a company's own success becomes the heaviest weight it carries.
NVIDIA reports FY2027 Q2 earnings in days. And the spread between consensus and guidance is thinner than a microchip's oxide layer.
Wall Street expects $92.18 billion in revenue. NVIDIA guided to $91 billion. The difference? 1.3%. That's not a beat gap. That's a coin flip dressed up as an earnings call.
Let's talk about what's really going on.
The Streak That Became a Trap
Seventeen quarters. That's how long NVIDIA has been smashing through analyst expectations like a wrecking ball through a glass factory.
But here's the thing nobody's saying out loud: that streak has become a cage. Every single quarter, the bar rises. Analysts see the streak, and they price it in. The market sees the streak, and it demands it. NVIDIA sees the streak, and it has to keep feeding the machine.
Community buzz isn't about whether they'll beat anymore. It's about what happens when they can't.
The numbers are so big now that the rate of surprise has to shrink. You can't keep doubling a base that's already the size of a small country's GDP. And when the growth percentages start coming down—when the "+97% YoY" becomes "+50%"—the market might react not to the health of the business, but to the psychological shift.
I didn't build this forecast model. I just read the tea leaves: NVIDIA's product cadence has been relentless—Ampere to Hopper to Blackwell to Rubin. That rhythm has been their hidden superpower. Thirteen quarters of "surprises" doesn't happen by accident. It happens because management has mastered the art of underpromising and overdelivering.
But here's what I've learned from watching this company through more cycles than I care to count: that kind of guidance discipline works until it doesn't.
Blackwell Ultra: The $300,000 Question
Let me take you inside the technical layers.
The Blackwell architecture (B200/GB200) uses TSMC's custom 4NP process—a beefed-up 5nm node that's been in production for over two years. Yield rates have stabilized above 90%. The teething problems are gone.
But Blackwell Ultra—the B300 series—is the one actually entering the market in this quarter. And that's where the interesting stuff happens.
B300 uses a dual-die design with CoWoS-L packaging. That's TSMC's 2.5D advanced packaging technology. If you don't know what CoWoS is, let me break it down simply: it's the bridge that connects all the chips together so they can talk to each other at insane speeds. Without it, Blackwell doesn't exist.
Here's the thing that keeps me up at night: NVIDIA is consuming over 60% of TSMC's total CoWoS capacity. When NVIDIA breathes, TSMC's packaging lines hold their breath. And when a competitor wants to launch a chip, they're fighting for whatever CoWoS crumbs remain.
The supply chain isn't just a manufacturing story. It's a moat. You can design the best GPU in the world, but if you can't package it, it's a beautiful paperweight.
Based on my own audit experience tracking TSMC's monthly revenue reports and packaging yields, the CoWoS expansion plan is on track. TSMC has committed to doubling monthly capacity to 80,000 wafers by end of 2025. Equipment delivery cycles for traditional packaging tools are shorter than EUV tools—roughly 6-9 months from install to production.
But here's the catch I keep coming back to: if B300 shipments ramp faster than expected, that signals CoWoS bottleneck relief. And that's good news for NVIDIA. But it's also good news for everyone else waiting behind them.
The Margin Puzzle That Nobody's Solving
Let me get into the financials now, because this is where it gets spicy.
Analysts expect adjusted EPS of $2.09. That's a 99% year-over-year increase. But wait. That's higher than the revenue growth rate of 97%.
How do you grow EPS faster than revenue?
Margin expansion. That's the only math that works. The market is betting that NVIDIA's gross margins will improve with the Blackwell Ultra ramp. Better product mix. Better pricing power. Better cost efficiency from CoWoS scale.
I love this assumption. It's clean. It's logical. And it could be completely wrong.
The margin equation for NVIDIA right now is: HBM costs are rising. HBM3E is in short supply. HBM4 is coming in 2026 at even higher prices. And NVIDIA has to buy HBM from SK Hynix, Samsung, and Micron—it can't make its own.
So the margin story comes down to one question: can NVIDIA's pricing power outrun its input costs?
In a supply-constrained market, the answer has historically been yes. NVIDIA's B200 GPU sells for $30,000-$40,000 per card. The GB200 NVL72 system goes for around $3 million. And they're selling everything they can make.
But the "everything they can make" is the issue. If CoWoS capacity is tight, NVIDIA can't ship more units. And if they can't ship more units, the revenue growth has to come from price—which is already at dizzying levels.
The margin question is the biggest tell in this earnings report. If gross margin surprises to the upside, NVIDIA's pricing power is still expanding. If it comes in below expectations, the cost pressure is real.
The China Factor: A Political Football That Won't Stop Bouncing
Everyone talks about AI demand, but the China angle is the quiet storm.
NVIDIA's China revenue has collapsed from about 25% of total revenue in 2022 to under 10% now. Export controls have throttled the flow of high-end GPUs to Chinese customers. H100. H200. B200. All off the table without a license.
But here's what I keep circling back to: China still represents 20-30% of global AI chip demand. That's a massive market that NVIDIA can't fully serve.
The "H20" workaround—a downscaled chip that meets export requirements—was NVIDIA's attempt to keep a foothold. It worked. It kept revenue flowing. But it's not a long-term solution. It's a patch. A temporary bridge.
And the Chinese ecosystem is building its own alternatives. Huawei's Ascend chips are getting better. They're not at NVIDIA's level yet—the software ecosystem gap is still 2-3 years—but the gap is closing.
The geopolitical risk isn't a binary event. It's a slow bleed.
If NVIDIA loses China entirely, they lose about 10% of revenue. The market has already priced that in, mostly. But if they gain access back—if the export controls get relaxed—that's upside that the market hasn't fully priced in.
The "China market update" in this earnings call will be a critical tell.
The Competitive Landscape: The Mountain Nobody's Climbed
NVIDIA's market share in AI training GPUs is somewhere between 80-90%. That's not a lead. That's a dynasty.
AMD's MI300/MI350 series is the closest competitor on paper. But here's the thing: hardware specs don't win markets. Software ecosystems do.
CUDA is NVIDIA's moat. And it's not just about the code. It's about the millions of developers who've learned CUDA, built libraries, and trained their models on NVIDIA's stack. Switching to AMD's ROCm or Intel's OneAPI isn't a technical decision. It's a cultural shift.
Based on my experience talking with developers in the space, the general sentiment is: "It's not worth the headache to switch." And that's what you call a durable competitive advantage.
Google TPU and AWS Trainium are interesting in specific scenarios. They're cheaper, more power-efficient, and optimized for specific workloads. But they lack the general-purpose flexibility that NVIDIA offers.
The threat isn't AMD. It's the cloud giants building their own chips. If Google decides to go all-in on TPUs for its internal workloads, that's less demand for NVIDIA. But that's a slow-burning threat, not an imminent one.
The Valuation Question: Are We Paying for the Future or the Present?
Now let's talk about the elephant in the room. The valuation.
NVIDIA trades at roughly 50-60x trailing earnings. That's not cheap. And it's not a value stock. But in the context of a company growing revenue 97% year-over-year, the forward PE doesn't look as scary.
Here's what I tell people who ask me if they should buy NVIDIA stock: you're not paying for the company. You're paying for the monopoly.
The question is whether that monopoly will last.
If AI demand continues to explode—and all signs point to yes—NVIDIA will continue to print money. The CSPs (Microsoft, Meta, Google, Amazon) are spending $300 billion+ on AI infrastructure in 2025-2026. Most of that flows to NVIDIA.
But if AI demand hits a speed bump? If the CSPs decide to pull back on capital expenditures? If the AI bubble bursts?
Then the 50-60x P/E becomes a 30x P/E. And that's a painful adjustment.
The market is a momentum machine. It loves winners. But it turns on them faster than you can blink.
The Real Threat Nobody's Talking About
Let me take you to the contrarian angle that nobody in the mainstream is discussing.
Everyone's focused on the supply chain. The CoWoS. The HBM. The TSMC capacity.
But I keep thinking about the demand-side concentration risk.
NVIDIA's top five customers—Microsoft, Meta, Google, Amazon, Oracle—account for 60-70% of revenue. That's massive concentration. And these companies are subject to their own cyclicality.
If Microsoft or Meta decides to cut their AI infrastructure spending—if their AI products fail to generate returns—NVIDIA's revenue could tank in a hurry.
The AI infrastructure buildout is a bet on the future. But it's a borrowed bet. CSPs are spending now on the promise of future returns. If those returns don't materialize, the spending stops.
The "AI bubble" debate isn't about whether AI is real. It's about whether the spending is justified by the returns.
And that's the question that will determine NVIDIA's fate.
What I'm Actually Watching This Earnings Call
I've been through enough cycles to know what matters. So here's my shortlist:
1. Blackwell Ultra shipment speed. If B300 is ramping faster than expected, that's a signal that CoWoS bottlenecks are easing. That's good for NVIDIA and good for the entire supply chain.
2. Gross margin. If margins are expanding, NVIDIA's pricing power is intact. If they're contracting, the cost pressure is real.
3. China update. Any signal on export control changes would be a big deal. Even a hint of relaxed restrictions would be a major positive.
4. Next quarter's guidance. If the guidance is over $100 billion, that's a statement of confidence. If it's below expectations, that's a red flag.
5. The tone. The vibes. The body language. The confidence in the voice. I've seen a management team that's confident versus one that's defensive. And the difference is palpable.
The Bottom Line
NVIDIA is at a pivotal moment. The company's products are still the best in the world. The demand is still there. The supply is getting better.
But the stakes are getting higher. The streak of "beats" is a burden. The expectations are so high that a "slight miss" could be punished severely.
The story isn't about whether NVIDIA is a great company. It's about whether the market's expectations have become unmoored from the fundamentals.
The market is a creature of momentum. And momentum creates euphoria. And euphoria creates the possibility of a hard landing.
Here's what I'll be watching: not the revenue number. Not the earnings per share. But the change in the narrative.
If NVIDIA delivers a solid beat and the stock drops, that's a sign the market is looking for reasons to sell. That's the real signal.
Speed isn't about the fastest earnings call coverage. It's about understanding what's actually happening under the surface. And right now, the surface is calm. But the currents underneath are shifting.
The AI story isn't ending. But the days of easy, surprising beats are probably numbered. And the market is going to have to learn to live with NVIDIA being "just" a great company—not a miracle machine.
When the chart collapses, I don't panic. I look for the signal. And the signal here is clear: NVIDIA's next chapter is about execution, not surprise. And that's a completely different story for the stock price.