Nvidia: the pre-earnings debate
MARCUS WEBB (Macro-Focused Value Manager, Multi-Asset Fund): So we're sitting here the night before what's probably the most watched earnings print of the year, and I think the first question we have to ask ourselves is whether this is actually an interesting setup or whether the market has already done all the work for us. Consensus is at $65.5 billion in revenue, EPS around $1.50 to $1.52, and the stock is sitting at $190 — up 47% over the last year but essentially flat for six months. Sarah, you're long and comfortable. Help me understand why tonight you're sleeping well.
SARAH CHEN (Long-Only Growth Fund Manager, Technology Specialist): Because the pattern is too consistent to ignore, Marcus. Thirteen consecutive quarters of beating revenue consensus. Thirteen. And every single time we sit in a room like this, there's a credible-sounding reason why this quarter is the one where it finally falls apart. A year ago it was DeepSeek — the narrative was that more efficient models would crater GPU demand. The stock sold off going in, NVIDIA reported, and it wasn't even close. Today the story is ASICs and hyperscaler custom silicon. Same fear, different wrapper. The underlying demand signal hasn't changed — Meta is spending $60 to $65 billion in capex this year, Amazon, Google and Microsoft have all maintained or accelerated their data centre commitments. That money flows to Blackwell. It's that simple.
MARCUS: It's not that simple though, and I think you know that. The beat-and-raise dynamic that drove explosive moves in 2023 and 2024 has fundamentally changed character. Management has gotten very good at guiding to numbers they can clear — so what you're really pricing in now is a modest beat on already elevated expectations. That's a very different risk-reward to what early holders had. And the ASIC story isn't just a narrative, Sarah. Google's TPUs, Amazon's Trainium, Broadcom's custom silicon — these are real programmes with real engineering talent and real roadmaps. The consensus range on FY2027 revenue is extraordinarily wide, and that width is telling you something about how much genuine uncertainty sits underneath the headline number.
SARAH: The ASIC threat is real at the margin, I'll grant you that. But margin is the operative word. There's a supply chain reality here that people keep underweighting — NVDA is expected to control roughly 60% of TSMC's CoWoS advanced packaging capacity by late 2026. AMD simply cannot compete at that level of integration, and the hyperscalers building custom chips are doing so for specific narrow workloads, not as a wholesale replacement for Nvidia's stack. CUDA is 17 years of developer entrenchment. You don't unwind that in a product cycle. And here's something I find genuinely interesting going into this print — one analyst has made the case that the market is actually underpricing NVDA relative to the AI cycle right now, because sentiment has rotated toward Broadcom, the optical plays, memory names. The argument is that NVDA's share price is discounting a 2026 peak in demand that simply isn't supported by the capex data.
MARCUS: I've seen that argument and I think it has some merit on a relative basis, but it doesn't address my core concern which is the absolute valuation at $190 heading into a binary event. You're at a $4.6 trillion market cap. Even on a forward FY2027 P/E — if EPS growth comes through as expected at around $7.66 per share — you're still at roughly 25 times. That's not obviously cheap for a business that, underneath the platform narrative, still derives the overwhelming majority of its gross profit from selling hardware into a buildout cycle. Morningstar has a fair value of $240, which implies the stock is moderately undervalued here, and I respect their work — but that number embeds assumptions about moat durability that I think deserve more scrutiny than the market is currently giving them.
SARAH: And yet on a ten-year average P/E basis the stock is trading at a significant discount to its own history. Look — I'm not dismissing the valuation conversation, but I think the margin story is more durable than you're giving it credit for. Blackwell's gross margins have been a concern on the ramp, consensus is modelling around 76% for this quarter, which is a slight compression — but the software layer is building. AI Enterprise at $4,500 per GPU per year is real recurring revenue, and the Rubin platform coming in the second half of this year resets the benchmark again. The product cadence is relentless.
MARCUS: The software revenue is real but it's still small as a percentage of the overall business. The margin story is mostly hardware-dependent and that's the honest answer. And look — I'm not arguing this company is going away. I'm arguing that the risk-reward at $190 the night before earnings is asymmetric in the wrong direction. A strong quarter with strong guidance gets you 4 or 5% upside. Any hesitation on the Q1 guide — which consensus has at $66 to $67 billion — and you're looking at an 8 to 10% drawdown. Several large institutional holders have been disclosed as trimming positions ahead of this report. That's not noise.
SARAH: Those are traders, Marcus. Long-duration holders don't move on one quarter. And the guidance question — that's where I actually feel most comfortable. If Jensen walks out and guides to $68, $70 billion for Q1, the conversation about ASIC risk and margin compression gets shelved for another three months, and the stock re-rates toward the $210 to $220 range. The path of least resistance here is higher, and I think the market knows it.
MARCUS: Or Jensen guides conservatively because he's managing expectations into a year where China sales are still largely excluded and the second half product transition to B300 creates genuine visibility risk — and suddenly the bear case has numbers attached to it. I'd rather have dry powder and add on weakness Thursday morning than be sitting fully committed at these levels tonight. The quality of the business isn't the debate. The entry point is.
SARAH: Fair enough. But I'd remind you that the people who waited for weakness on NVDA over the last three years are still waiting.
MARCUS: (laughs) And the people who bought at $140 thinking it couldn't go lower found out it could. We'll know tomorrow night.
(Note this is a fictional debate between AI-generated characters, obviously!)