CNBC Today on NVIDIA Stock Ahead of Earnings — NVDA Analysis

FinVid. CNBC Today On NVIDIA Stock Ahead of NVIDIA Earnings. Detailed Summary. February 14th, 2026.

This is an 18-minute video from FinVid compiling CNBC commentary and original analysis on Nvidia ahead of their February 25th earnings report. The video covers capex justification, chip diversification concerns, margin guidance expectations, and why the creator remains bullish long-term.

Section 1. The AI Capex Debate on CNBC.

The video opens with Bank of America's flow show noting that the most obvious catalyst to reverse the current "AI awe to AI poor" sentiment would be a capex cut. But the analyst thinks the chances of that are slim and that it wouldn't be well received by the market anyway.

Jensen Huang told CNBC last week in California that the $660 billion being spent by hyperscalers this year is completely justified. His argument: all of these companies' cash flows are going to start rising. People are comparing capex to current cash flows, and Jensen says one of those numbers is wrong. It's the cash flow number that's wrong because it hasn't caught up yet. Every single company sees the same inflection point, which is why everybody is leaning in so hard.

CNBC's response was measured. One commentator noted that Jensen would be expected to say that, but the underlying logic has merit. The counter-argument: investors were already pencilling in double-digit increases in free cash flow for companies like Meta and Microsoft before the AI spending surge. There's risk in fronting this much spend. But Amazon and others are seeing utilization go up as fast as they can add capacity, so they're acting rationally. The problem, as one host noted, is that "a lot of rational actors investing toward the same goal creates winners and losers. That's the whole point of capitalism."

They also noted Anthropic raising $30 billion on a Series G at a $380 billion post-money valuation. Revenue running at about $14 billion annually. These numbers are getting very interesting.

Section 2. NVIDIA Leasing a Data Centre Funded by Junk Bonds.

Bloomberg reported Thursday evening that Nvidia will be leasing a data centre in Nevada, with construction funded by $3.8 billion in junk bonds. A special purpose vehicle backed by Track Capital is issuing high-yield debt to fund a 200-megawatt data centre and substation. Nvidia will then lease that facility.

FinVid's interpretation: Nvidia wants more capacity for their own internal workloads. Jensen told the world at CES in January that Nvidia has built up massive DGX capacity to develop open-source models that drive the entire industry forward. Nvidia wants to be the leader in open-source models, and this new Nevada facility is likely part of that strategy.

The strategic logic follows Jensen's "five-layer cake" analogy. At the top is the application layer, supported by the model layer underneath. By being the leader in open-source models, Nvidia ensures that many future AI applications will be powered by Nvidia models. Multiple popular AI applications built on Nvidia's own models creates a very advantageous flywheel position.

The CNBC panel noted this adds another layer to the conversation about project finance in AI. Unlike Alphabet going out and getting the best cost of capital with 100-year bonds, this is high-yield debt. The comforting piece, one host argued, is that so many people are watching closely. "We really are braced up in advance for the big one to hit," suggesting the market is more aware of potential malinvestment than in previous bubbles like the late 1990s or the pre-financial-crisis housing bubble.

Section 3. Chip Diversification and the OpenAI Move.

OpenAI unveiled its first model to run entirely on chips from the startup Cerebras, GPT 5.3 Codex Spark, a small model optimised for fast low-latency inference. This is part of a larger trend. Google shipped Gemini 3 trained and served entirely on its own TPUs. Chinese AI lab GLM released a model trained on Huawei chips. Meta and Microsoft are both shipping their own versions of AI chips.

CNBC's analysis was sharp. The move matters more financially than people realise because while training is a one-time cost, inference, actually running the model for millions of users, is a recurring cost. As products scale, inference is where the majority of compute dollars go. The flagship model is the prestige play, but the workhorse model is the revenue play. Nvidia can continue to win the former, but it's ceding ground in the volume game.

The diplomatic framing from OpenAI: "Nvidia is still foundational" while "deliberately expanding the ecosystem." CNBC's translation: "They're saying we love Jensen, but we're dating other people."

There's also real friction. Reuters reported earlier in February that OpenAI is unsatisfied with some of Nvidia's latest chips for inference. Every major AI lab still wants Blackwell allocation, the gold standard, but behind closed doors nobody wants to be this dependent on a single supplier. They're smiling at Nvidia with one hand and signing deals with Cerebras, Broadcom, and AMD with the other.

Arista Networks reported earnings and revealed that a year ago 99% of their deployments were Nvidia, but now AMD is the preferred accelerator in 20 to 25% of deployments. FinVid's important caveat: Arista's deployments don't represent all deployments in the AI ecosystem, and this is not zero sum. The total addressable market is growing so fast that even if Nvidia's percentage share decreases slightly, the absolute revenue grows significantly. "Let's not miss the forest for the trees."

Section 4. The Investment Thesis for Infrastructure.

One CNBC guest offered a compelling framework. If you want 10x returns or solid returns for the next decade and you've already seen huge gains on the hardware side, the key insight is: "Inference IS AI. AI is thinking. That's inference. Think of those two together, not training. Training is not AI, inference is AI. This could be tens, hundreds of thousands of times bigger."

For software stocks, he expects a bounce but prefers usage-based models like Snowflake and Datadog over seat-based models. For infrastructure, he thinks it continues higher for longer.

On Nvidia specifically: still worth owning, but diversify your portfolio beyond just Nvidia.

Section 5. Earnings Preview and Key Catalysts.

FinVid lays out the bull case for Nvidia's February 25th earnings.

On margin guidance, which will be the key focus: Nvidia previously guided margins down into the low 70s percent range during the Blackwell manufacturing ramp, then promised a return to mid-70s by late in the fiscal year. They delivered on that promise. If they guide gross margins notably above 75%, that's very positive. If they guide low 70s, the market will see it as a disappointment. Input costs are rising and leadership has said they're working to hold margins in the mid-70s range.

The $500 billion cumulative revenue figure Jensen shared at GTC in October is the big number to watch. That was cumulative revenue from Blackwell and early Rubin ramps through 2026. Jensen had said 6 million chips had already shipped, implying $350 billion in cumulative revenue over the next five financial quarters. Nvidia's CFO said in January that "the $500 billion has definitely gotten larger." If we get a sizable update to this number at earnings, it could serve as a catalyst, just as the original $500 billion figure caused the stock to hit an all-time high back in October 29th.

FinVid notes that Nvidia stock is exactly where it traded 6 months ago. It's trading at a forward PE in the mid to low 20s, basically a market multiple, while delivering substantially greater earnings growth than the broader market. His view: it's only a matter of time before the stock moves higher based purely on fundamental growth. He can't tell you when, but he says it's inevitable.

Section 6. The Product Roadmap and Long-Term Runway.

The production ramp timeline: Blackwell Ultra is ramping very quickly right now. Rubin launches in 2026. Rubin CPX at end of 2026. Rubin Ultra in 2027. Feynman after that in 2028. A clear data centre product roadmap stretching to 2028.

Jensen recently told CNBC it will take 7 to 8 years to build up to the level needed, and that we have several years of buildout ahead. If the total buildout takes 7 to 8 years and we're 3 to 4 years in, that leaves several more years where demand continues to outpace supply. Jensen has previously said we won't hit a glut for a few years, and new use cases in physical AI will increase demand even further, potentially extending the runway beyond what's currently projected.

Physical AI is the sleeper thesis. Nvidia's CFO called it "a multi-trillion dollar opportunity and the next leg of growth." Nvidia sells hardware for data centres where models are trained, offers Omniverse where models are taught and tested, and sells on-device hardware like AGX for real-time inference in robots. Over 2 million developers are already building on the Nvidia robotic stack, and FinVid says this is not getting enough attention.

Jensen believes there will be $3 to $4 trillion of global AI factory buildout between now and 2030. FinVid's conclusion: he doesn't think we're anywhere near a bubble-bursting event. Nvidia still has plenty of runway and will be worth substantially more in future years than today. Stay calm, maintain a long-term perspective, and don't make hasty decisions.