Nvidia's Revenue Doubled Year Over Year. The Market Is Still Pricing 2024.

The stock barely beat the S&P 500 in 2026. That is the headline that matters more than the $81.6 billion in quarterly revenue, the 85% year-over-year growth, or the 75% gross margin that held steady through the biggest sequential ramp in the company's history. Nvidia shares are up roughly 12% this year — a number that sits between the +39% from 2025 and the +171% from 2024, and one that has investors quietly asking whether the engine is finally stalling.

The market is still pricing the old story: hyperscaler capital expenditure is peaking, custom silicon is eating Nvidia's lunch, and the margin pressure from a new product cycle is just around the corner. The cash-flow path says something different.

The operating setup is accelerating, not decelerating

Nvidia guided Q2 FY2027 revenue to $91 billion. The consensus estimate was $86.8 billion. That is not a conservative extension of the current quarter — it is an acceleration, set against a base that already grew 85% year over year. And the guidance explicitly assumed zero data center compute revenue from China, which removes what would otherwise be a tailwind. The Q2 number is conservative, not optimistic.

Gross margins are projected at 75.0% (non-GAAP, ±50 basis points), matching Q1 to the decimal. The market was braced for compression as Blackwell transitioned into full production and memory costs climbed. Instead, Nvidia maintained pricing power at a revenue scale that would challenge any other semiconductor company in history. That margin figure matters because it signals that the Blackwell ramp is not a cost event — it is a volume event with system-level pricing discipline.

Trailing-twelve-month free cash flow sits at $119.1 billion, up 65% year over year. Operating cash flow is $125.6 billion against capital expenditures of only $6.6 billion. Nvidia isn't a capital-intensive infrastructure play; it's a cash-generating monopoly wrapped in a product cycle that keeps extending its lead.

The ASIC narrative is a long-term wedge, not a near-term threat

The bear case has a specific shape. Hyperscalers — Google, Meta, Amazon, and Microsoft — are building their own AI accelerators. Google's seventh-generation TPU (Ironwood) is live on Google Cloud. Meta's MTIA 400 completed testing in March 2026 and is deploying. Amazon's Trainium 3 is ramping. Broadcom, the dominant design partner for these custom chips, reported AI semiconductor revenue of $10.8 billion in the most recent quarter, up 143% year over year, with guidance pointing to $16 billion next quarter.

The structural shift is real. Custom ASIC shipments are projected to triple between 2024 and 2027, and by 2028 they may surpass GPU shipments for the first time. For hyperscalers running billions of daily inference requests on stable models, custom silicon offers 3-to-5x better performance per watt at the target workload, with an 18-to-24 month break-even that compounds into massive annual savings.

But the ASIC wedge cuts at the margins, not the core. Custom chips work only for predictable, high-volume inference workloads. They don't replace the flexible compute that Nvidia dominates — training, experimentation, mixed workloads, and the rapid iteration cycle that frontier AI companies like Anthropic and OpenAI actually run on. Nvidia's Q1 results make the distinction visible: data center revenue was $75.2 billion, of which $37 billion came from ACIE (AI clouds, industrial, enterprise), a segment that more than tripled year over year. The company is not just selling to hyperscaler capex anymore. It's selling into the customers that hyperscalers' custom chips don't serve.

Nvidia's own response to the ASIC threat is structural, not rhetorical. The Vera CPU, designed from the ground up (not an Arm off-the-shelf core) for AI agent workloads, targets a $200 billion CPU market currently held by Intel and AMD. CFO Colette Kress guided $20 billion in CPU revenue for the fiscal year, with every major hyperscaler and system maker as a partner. The strategy is vertical integration: sell the full rack — GPU, CPU, networking, software — so that even if a hyperscaler wants a custom accelerator for one workload, they still buy the rest of the infrastructure from Nvidia.

The Vera Rubin cycle is the inflection

Jensen Huang described the Vera Rubin platform as "off to a tremendous start" and expects it to be "even more successful than Grace Blackwell." The system delivers 10x more performance per watt than its predecessor. Huang went further: he expects Nvidia to be "constrained throughout the entire life of Vera Rubin" by demand.

That language matters. During the Blackwell ramp, the market's biggest fear was production delays turning into revenue delays. Rubin has no such history — it was announced at CES in January 2026, returned from TSMC on schedule, and is ramping through H2 2026 without the thermal and yield problems that plagued Blackwell's debut. Huang is telling investors that the next product cycle won't be supply-constrained into revenue, but demand-constrained into supply. The constraint is on the wrong side of the equation for a bear.

Deliveries of the Vera CPU started in June 2026. Early customers include OpenAI, Anthropic, and SpaceX. The Vera CPU isn't a traditional server chip — it's purpose-built for the "agentic AI" workloads that Huang says have gone "parabolic." The agentic AI thesis (autonomous agents performing background tasks at scale) increases CPU demand alongside GPU demand, which is the opposite of the 2022-2024 AI dynamic where GPUs cannibalized CPU relevance. Nvidia is now attacking both sides of the rack.

Why the stock is muted

Nvidia trades at roughly $223, with a market cap of $5.4 trillion. The forward P/E is around 60x, which looks expensive if you're thinking in single-digit growth terms. But at $119 billion in trailing free cash flow and Q2 guidance implying a full-year run rate approaching $360 billion in revenue, the multiple is a function of scale, not hope. The PEG ratio (forward P/E divided by earnings growth rate) sits at 0.31, which is low for a company that grows at this pace — the market would need to see growth slow dramatically for that multiple to look unjustified.

The real reason the stock isn't running is crowd. Bank of America's July survey found that 82% of fund managers view semiconductors as the most crowded trade in the market, with none reporting short positions. When everyone is long, there's no one left to buy. Sector rotation into hyperscalers, software, and healthcare has pulled capital away from pure-play chip names. UBS projects hyperscaler capex growth slowing from 76% in 2024 to 25% in 2027 and 6% in 2028. That deceleration curve is real, and it is the anchor keeping Nvidia's price action below its fundamentals.

But here's the mismatch the crowd hasn't fully processed: Nvidia's Q2 guidance of $91 billion already beats the $86.8 billion consensus. The company is pointing to sequential acceleration in a quarter where the capex slowdown narrative should, if anything, weigh on sentiment. And it's doing so with no assumed revenue from China. If the capex story were actually breaking, guidance would be flat or conservative, not raised. The stock is pricing a future where the hyperscaler spending boom is fading, while the company is pointing to a quarter where that spending is accelerating.

AInvest's aggregate signal still labels Nvidia a Buy, with a composite analysis rating of 0.9 and a fundamental rating of 4.49 out of 5. That's the consensus reflecting the old story, not pricing in the gap between it and the actual operating trajectory.

The bridge

This isn't about sentiment. It's about whether a business that generates $119 billion in free cash flow, guides sequentially higher revenue with stable margins, and is launching a $200 billion CPU platform can justify its scale over the next 12 months.

The Vera Rubin ramp in H2 2026 is the decisive proof point. If Rubin ships on Huang's timeline and maintains the 75% margin profile, the stock's current multiple starts to look like a discount to a $400 billion revenue company — not a stretch, given Q2 guidance at $91 billion and the ACIE segment tripling year over year. A simple 25x forward P/E on conservative $70 billion in diluted net income would imply a price well above the current level. Simple forward multiples beat complex DCF models, and they don't require assuming the ASIC threat materializes faster than the data suggests.

The risk

The setup breaks if the ASIC migration accelerates beyond the current narrow-inference wedge. If hyperscalers announce broad training workloads on custom silicon — or if Nvidia's Q2 results come in flat or below $91 billion with margin compression below 74% — the inflection thesis loses its operating anchor. The Vera CPU is unproven at scale, and convincing hyperscalers to switch from entrenched Intel and AMD CPU stacks is harder than Huang's language implies. Oracle and OpenAI are the only named early partners; a broader adoption story needs more than two.

If Q2 revenue misses guidance or margins slip into the low 70s, cut without ego. Discipline over conviction.

But as things stand, the market is pricing a slowdown that hasn't shown up in the numbers. The cash flow is accelerating. The product cycle is extending, not shrinking. And the bar for the next quarter is already being beaten before the quarter ends.