#NvidiaDefendsAIFunding

$NVDA CEO Jensen Huang says “cybersecurity will likely be the next major use case of AI” and that its “going to run continuously.” If security moves towards always-on AI inference then that could materially expand the workload running through platforms like $CRWD and $PANW.#PPIHotCPINext #OracleAICloudUp121% #BTCSpotETFOutflows

#OracleAdobeToday AI demand is no longer the question. The bill is 👀 Oracle has a massive $638B backlog, but investors want to see how quickly it becomes revenue and whether that cash can outrun AI capex. Adobe faces a similar test with Firefly and GenStudio: can AI lift revenue without eating margins? What caught my attention is the shift. From Oracle's cloud to Adobe's software and Apple's AI hardware, the race is moving from building AI to proving it actually pays.

Oracle, Adobe, Today: AI demand isn’t the issue anymore. The cost is 👀 Oracle has a huge $638B backlog, but investors are keen to see how fast it turns into revenue and if that cash can surpass AI spending. Adobe faces a similar challenge with Firefly and GenStudio: can AI boost revenue without From Oracle’s cloud to Adobe’s software and Apple’s AI hardware, the focus is shifting from creating AI to showing it actually pays off. #SeptHikeOddsHit90%

The harder test for AI funding is revenue quality. Huang's defense points to real customer contracts and Nvidia's small share of deal financing. My read: those details matter, but durable demand would be more convincing if customers can fund repeat GPU purchases from operating cash flow. Infrastructure commitments alone cannot settle that question. #NvidiaDefendsAIFunding

"Short-Lived" Hiking-Cycle, "Neutral" Positioning, 20x Multiples: Goldman's Guide To Buying AI While The Left-Tail Is Still On Fire

I found this really interesting. Apparently, adding cheaper GPUs can help you get more work out of the expensive ones you already own. NVIDIA tested this with four GB200 GPUs and two lower-cost RTX 6000D GPUs. The RTX cards handled the image-processing stage, leaving the GB200s to run the language model. On an image-heavy workload, the combined setup served 70% more traffic while meeting the same response-speed target. They added hardware, so this wasn't a free performance gain. But they increased capacity without buying more GB200s. I think this opens up an interesting opportunity for neoclouds. If access to the most expensive chips limits how much work they can take on, a cheaper chip that frees up those machines becomes worth a lot more to them. That could pull AI demand further down the GPU market. These were professional RTX cards, not consumer gaming cards. But wherever consumer cards can handle those smaller jobs economically, neoclouds could become another buyer competing for the same hardware as gamers and small developers. A consumer GPU wouldn't have to run the whole model to be useful. It would only have to take enough work off a scarce, expensive GPU to justify buying it. If that starts happening at scale, some of the shortages we're watching in data centres could spread into parts of the market that look well supplied today.

For all the AI experts, does this mean the leaders publicly slow down what they release but still internally operate under a prisoners dilemma and maintain pace? Thats kind of the heal turn move, no? Optically this would mean all internal development slows, the market structure remains muddled, FCF is slower to develop and capex becomes more dependent on debt. I think.

#OracleAdobeToday AI demand is no longer the question. The bill is 👀 Oracle has a massive $638B backlog, but investors want to see how quickly it becomes revenue and whether that cash can outrun AI capex. Adobe faces a similar test with Firefly and GenStudio: can AI lift revenue without eating margins? What caught my attention is the shift. From Oracle's cloud to Adobe's software and Apple's AI hardware, the race is moving from building AI to proving it actually pays. #SeptHikeOddsHit90%

10 NAMES MAPPING THE AI CLOUD LANDSCAPE 1. $CRWV combines deep $NVDA integration with software designed to keep massive AI training clusters running efficiently 2. $NBIS combines in-house hardware + software with control over infrastructure and power creating more opportunities to improve efficiency 3. $DOCN makes AI inference and agent development accessible to smaller businesses through a managed cloud platform 4. $IREN combines owned land with a large secured power pipeline giving it a foundation for expanding AI capacity 5. $SHAZ combines regional partnerships with sovereign cloud capabilities to serve customers needing local infrastructure and data control 6. $NET combines AI inference with developer tools, storage and security making it easier to build applications across its edge network 7. Crusoe builds its campuses around available energy aligning data center development with power needed to operate them 8. Lambda specializes in dense GPU clusters with managed software that simplifies training and running AI models 9. $AKAM brings computing, security and content delivery together across an established global network close to customers 10. $FSLY uses lightweight WebAssembly execution to start applications quickly and handle requests close to users

Still surprises me how many think AI compute won’t become the largest market in the world Traditional cloud infra spending grew ~52% annually from 2010–2020, mostly powering the internet AI compute are like "units" of labor. First, white-collar. Then blue-collar (robotics) The internet is huge but not as big as the totality of the labor markets