Since January 12, we have seen a meaningful step forward in AI utility, led by the emergence of Claude Code, Claude Coworker, and Open Claw. The takeaway is investors are getting more evidence that models can do real work, not just talk about work. Vibecoding is a clear example because it compresses the path from intent to working software, and personalized assistants are starting to feel less like demos. At Deepwater we recently built personalized AI (walled off inside of a third party cloud) assistant to support the firm, and I can attest to the step function improvement in utility.

This shift matters because utility has been the central sticking point for many investors. The market’s posture changes. Over the past three weeks Wall Street has become more alert to second order effects, even if the fundamentals for most companies won’t change for the next year or two. That said, the next 2 to 5 years will likely look different, and that has turned the market into a minefield where relatively obscure AI announcements can move stocks across software, logistics, healthcare, and media.

The software implosion has been well documented. One example in media was the 10 second video from SeeDance, featuring a Tom Cruise versus Brad Pitt scenario, sparked a selloff in media companies. The specifics are less important than the reflex it revealed: investors now have a higher awareness of how quickly model capability is advancing, and they are repricing disruption risk sooner than the fundamentals warrant.

The second recent material shift is the step up in capex guidance, most notably from Amazon and Google. Amazon guided full year 2026 capex outlook roughly 25% above the Street, and Google’s was roughly 50% above. That implies Amazon capex growth from 2025 to 2026 of about 56% and about 98% for Google. Meta is will likely end up about 80% y/y increase.

Some of that step up in spending is related to higher memory pricing. Backing out the increase in memory pricing, I estimate Amazon capex will be up about 45% y/y and Google up just over 85% y/y. The bottom line is the companies closest to AI are either feeling increasing pressure to invest, or seeing an increasingly large opportunity, to invest more aggressively in infrastructure.

The market’s response has been consistent: concern that this is profitless prosperity, with a return profile that is uncertain and likely pushed out. I believe this reaction misses the bigger point. These companies employ some of the most capable teams in the world at understanding where AI can go, and their willingness to aggressively step up investment is noteworthy.

A useful starting point to gauge Nvidia comments on the earnings call is to look back to the company’s January 9th messaging at CES, which boiled down to a simple point: demand is running ahead of expectations.

There were three pieces that stood out. First, during his 90 minute keynote, Jensen used the word “skyrocketing” twice, once to describe the computation required for AI and once to describe demand for Nvidia GPUs. He tied that to a view that the size and number of more robust models is increasing by “an order of magnitude every single year,” and he pushed back on the scaling law debate by saying “the computing law continues to scale,” implying future progress remains compute hungry.

Second, CFO Colette Kress added the company’s prior outlook calling for $500B in Blackwell and Rubin revenue through the end of 2026 has “definitely gotten larger,” describing demand as “tremendous,” and expressing confidence that Nvidia can fulfill that demand.

Third, Jensen’s comments on the H20 relaunch into China were unusually direct. He described demand as “high,” “quite high,” and “very high,” and he suggested the company has prepared capacity, noting that they have “fired up” the supply chain. It sounds like those chips have just began shipping in small quantities over the past few days.

Putting it together, I believe Nvidia will likely give comments that land Street estimates for revenue growth in CY26 at 65% compared to current in print estimates of 55%, up from 50% on January 1st. My sense is the whisper number calls for 60% growth this year.

China. The China narrative is both a headline risk and a potential upside lever. As mentioned, the Street is looking for 55% revenue growth in CY26, which I believe includes roughly $10B of China revenue. Excluding China, that implies closer to 50% growth. As a reference point, $10B is just under 3% of sales, which helps puts into perspective the modest China outlook. Jensen continues to  referenced a $50B opportunity in China, which if they landed all of that revenue in CY26 would bring overall revenue growth to just above 70% growth (Street currently at 55%). They likelihood that all of that China revenue comes in this year is slim to none. The key is that China remains an opportunity, and the Street is not pricing in a China windfall in the next couple of years.

Inference. My belief is AI is not about training, it’s about inference. I believe  inference can be tens, if not hundreds of thousands of times bigger than training. Jensen will likely reiterated has comments over the past three earnings call with a similar message that investors are underestimating how inference can be a long-term growth driver for the company.

Physical AI. At his CES keynote, Jensen opened with a video titled “Physical AI,” which is a tell of the company’s long term message to investors in the upcoming call.  Nvidia appears to be addressing the central question, how can the growth party continue beyond 2026. Physical AI extends beyond chatbots and agentic AI and emphasizes models that understand, reason, and interact with the real world, including computer vision and applied physics. The use cases span from light on device physical AI on a phone or glasses to autonomous driving systems, specialized robots, and humanoid robots. In other words, it’s a massive market.

Today, only about 1% of Nvidia’s sales, reported as Automotive and Robotics, are tied to delivering physical AI. I believe by CY30 this segment can become a $50B plus annual business, up from about $3B today. That implies physical AI could represent about 8% of a $600B revenue base in CY30, compared to my $350B revenue outlook in CY26, with the Street at about $331B.

Stepping back, the market already understands that CY26 should finish ahead of the current Street estimate calling for 55% growth. The real question is what 2027 and 2028 look like, a topic the company is unlikely to address in any meaningful detail on the earnings call.

Investors are left with a basic question: do you believe inference will be significantly bigger than training, that physical AI is a large market, and that Nvidia can maintain its advantage versus custom silicon. If the answer is no, the stock is difficult to own. If the answer is yes, then the Street’s roughly 28% growth view for CY27 will prove too conservative. We believe CY27 can grow closer to 40% over CY26 because we are still in the second inning of the AI buildout, and Nvidia remains one of the clearest beneficiaries of that timeline.

Disclaimer