Anthropic’s latest data shows that Fable 5.1 outperformed its previous flagship Fable 5 and OpenAI’s GPT-5.6 Sol across multiple benchmarks, with especially strong improvements in coding and scientific research. That tells me AI capabilities are still advancing rapidly at the software layer. As these models become more capable, they can move deeper into real enterprise workflows and reshape how software is actually used. I think we are still early in that transition. AI is beginning to drive a much larger transformation across the software industry. $CRM

Everyone talks about public AI models. I think private AI is the quieter story. For governments and large enterprises, the problem is simple: Sensitive data cannot just leave the internal network. That gives $PLTR a pretty natural advantage. A lot of its customers already operate in environments where data control, security and permissions matter more than having the cheapest model. $MSTR is approaching the same theme from a different direction, packaging compute and AI infrastructure for enterprise deployments. I still think private AI is underappreciated. The more companies worry about where their data goes, the more valuable controlled, internal AI systems become. Both names are volatile, so I’m not treating either like a low-risk core position. But this branch of the AI trade deserves more attention.

Enterprise AI is finally being forced to do the math. Customers are cutting projects that look impressive but cannot show a real return. That’s the part I like about this phase. $NOW already sits inside workflows companies use every day, so adding AI can turn into measurable subscription revenue without rebuilding everything from scratch. $C3AI has a harder setup. The story sounds good, but customized projects take longer to deploy and the valuation still assumes a lot goes right. A few months ago, almost anything with “AI” in the pitch could get attention. Now customers want savings, productivity and renewals. That should separate the real software businesses from the stories pretty quickly.

Enterprise AI budgets are starting to get a lot more disciplined. Microsoft and Meta are reportedly cutting back on internal Claude spending, which fits the shift I’ve been watching: Companies are no longer buying third-party AI just because it looks impressive. They want measurable ROI. That’s why I still prefer $NOW here. Its AI products sit inside workflows customers already use, and the subscription revenue can actually be tracked. $C3AI is a tougher setup. It relies much more on large customized projects, so when budgets tighten, those deals can take longer to close. This is where the AI software market starts separating. The winners will be the companies that make existing businesses more valuable. The pure “AI story” names are going to have a much harder time.

Silver still feels like two different trades happening at the same time. AI hardware is adding real industrial demand, but ETF money keeps leaving. That disconnect is what makes the setup interesting. $AG gives you much more direct exposure to silver itself, so if industrial demand starts showing up in the price, the upside can move fast. $SSRM is different. It has silver exposure, but the stock still trades much more like a gold name, so the silver upside gets diluted. That’s why I’m not treating them the same. $AG is the higher-beta way to play industrial demand. $SSRM is the steadier precious-metals exposure. Right now the market is still trading Fed policy harder than physical consumption. I’m watching for the moment that starts to change.

Gold is bouncing again, but I still don’t think you can buy miners as one basket. Weaker payroll data is bringing rate-cut expectations back into the trade. The bigger difference for me is still cost. $CDE has better operating leverage when gold moves because its cash-cost structure gives more of that price increase a chance to reach the bottom line. $GOLD has some higher-cost operations and still has to keep spending heavily across parts of the portfolio. So even if gold goes higher, the earnings response can look very different. That’s why I keep coming back to the cost curve. Gold gives you the direction. Costs decide how much of the move the miner actually keeps. I’d rather buy the lower-cost operators on weakness than chase the whole sector after a gold spike.

Enterprise AI is getting a lot less forgiving. Companies are starting to cut projects that sound impressive but cannot show a clear ROI. That is the phase I have been waiting for. The software names that win from here probably will not be the ones with the best demos. They will be the ones already sitting inside real workflows. $NOW fits that pretty well. Its AI automation plugs directly into IT operations customers already use, which makes the path to paid adoption much cleaner. $WDAY has the same opportunity in HR and finance, but adoption still looks slower and more trial-heavy. So for me, the gap is starting to widen. $NOW has the stronger monetization setup today. $WDAY still has to prove that pilots can turn into meaningful paid deployment. AI software is moving from story time to cash flow time.

I’m starting to care less about how much gold a miner produces and more about how much margin it actually keeps. A lot of miners are hitting production targets. The problem is energy, labor and maintenance costs are rising right alongside the gold price. That changes how I look at names like $CDE and $HL. $CDE has better upside when metals run because the operating leverage can show up quickly. $HL has a broader operating footprint, but that also means more places for cost inflation to show up. So even with gold near strong levels, I’m not buying miners as one basket. Production growth looks good in a headline. Margin expansion is the number that actually pays me.

Gold miners are reminding everyone of something pretty simple: Higher gold prices do not automatically mean higher profits. Q3 production looks fine across a lot of the sector, but energy, labor and maintenance costs are still climbing. That can eat a surprising amount of the upside. $CDE is the name I like more when metals are moving because the operating leverage can show up faster. $HL has more exposure to cost pressure across a broader mine base, so stronger gold prices do not flow through as cleanly. That’s why I’m still selective with miners. I’d rather own the operators that can actually keep the margin expansion. Gold going up is only half the story. The other half is how much of that price increase the company gets to keep.

Enterprise AI is entering a more interesting phase now. Customers are not just asking what the Agent can do. They are asking what the ROI looks like. That sounds obvious, but it changes the whole software trade. $NOW has a big advantage because its Agents can be layered into workflows customers already use. The problem is that enterprise buying cycles are getting longer as companies become more selective. $WDAY has the same issue in HR and finance. The use cases are real. The customer base is sticky. But the revenue ramp is still much slower than the headlines make it look. So I’m watching AI-related ARR and contract growth much more closely than product launches now. The B2B Agent story is still alive. It just has to start proving the economics.

Silver is doing two jobs at once. It trades like a precious metal, but it also gets pulled directly into industrial demand. AI hardware and grid buildout are adding one source of demand. Solar is more complicated because manufacturers keep trying to reduce silver usage. That’s why I don’t like treating silver as just “high-beta gold.” $SVM gives you much more direct exposure to silver prices and mine execution. $RGLD is almost the opposite. It leans more toward gold and earns through royalties and streams instead of running the mines itself. So I’m watching two very different setups: $SVM for upside if silver really moves. $RGLD for cleaner downside protection. The supply story is interesting. But a high gold-silver ratio does not mean silver has to catch up tomorrow.

Gold is getting pulled in two directions again. U.S. employment data is weakening, which pushes rate-cut expectations higher. Central banks are still buying. But ETF money is still leaving. That split is why I’m not treating every gold miner the same. $SSRM gives me a steadier mix of gold and silver exposure with better cost control. $CDE is the higher-beta version. More upside if metals move, but the downside gets ugly fast when speculative money leaves. I still like the long-term setup for precious metals. Short term, ETF flows can matter more than the macro story people want to believe. So I’m not chasing spikes here. I’d rather let the market give me a better entry.

Enterprise AI is finally giving us something better than demos. $NOW says its AI Agent products have crossed $1B in annual contract value. That gets my attention because B2B AI is not about downloads. It’s about whether companies are actually paying to automate real workflows. ServiceNow has a huge advantage here. Its Agents sit inside systems customers already use for IT operations, so adoption can happen without rebuilding the whole stack. $WDAY is doing something similar in HR and finance with Sana. The tradeoff is speed. Enterprise software moves slowly. Pilots take time, budgets take time, and full deployments take even longer. So I’m not expecting one quarter to suddenly change everything. The number I care about now is AI-related ACV growth. That tells me a lot more than another product launch.

This is a pretty interesting way to look at the AI buildout. Amazon reportedly wants to move around 8 billion dollars worth of Nvidia Grace Blackwell chips into a separate special purpose vehicle and bring outside investors in. But Amazon is not getting rid of the chips. It would lease them right back and keep using them in its data centers. That part caught me. Everyone keeps asking if AI demand is slowing. Meanwhile one of the biggest cloud companies is trying to figure out a new way to finance thousands of Nvidia chips it still wants to use. These systems are just getting insanely expensive. Amazon is already planning around 220 billion dollars of capex for 2026. The hidden wrinkle here is that they are reportedly only offering up to a 10 percent equity stake to outside investors, using their high credit rating to raise the rest in debt. And honestly this looks like a way to move some of that ownership risk somewhere else. These chips lose value fast as newer generations come out. If Amazon can keep using the hardware without having 8 billion dollars of it sitting on its own balance sheet, I can see why that is attractive. It is also why Nvidia stock hit a fresh all time high today running past 237 dollars with a market cap crossing 5.7 trillion dollars. The market is starting to realize that the hyperscale buyers are not cutting back on orders at all. They are just changing the corporate accounting to buy even more. So maybe the next bottleneck is not demand for compute. It is how much of this buildout companies actually want sitting on their own balance sheets. I think we are going to see a lot more creative financial engineering around AI infrastructure if spending keeps going like this $AMZN $NVDA #AIInfrastructure #Hyperscalers #CapEx #AICompute

Precious metals are trading like two different markets right now. Central banks are still buying gold, while ETF money keeps moving the other way. And silver has something gold doesn’t: Real industrial demand from solar and AI hardware. That’s why I’m looking at $CDE and $TFPM very differently. $CDE gives me much more upside if gold and silver both move, but the mine-cost risk is real. When metal prices pull back, earnings can get hit fast. $TFPM is the cleaner setup for me. It doesn’t run the mines. It collects royalties and streams, so it avoids a lot of the labor, operating and geopolitical headaches that come with miners. I’m not trying to go all-in on a rate-cut story here. I’d rather keep $TFPM as the steadier core exposure and use $CDE only when I want more beta. Gold can stay messy for a while. That doesn’t mean the whole trade is broken.

Many people have the misconception that once OCS arrives, it will take market share from CPO. Let me clear that up. I actually think they solve two different problems. CPO is about widening the road. Its core is moving the optical engine closer to the chip, cutting the power and loss of high-speed electrical links. As we go from 800G to 1.6T and 3.2T, bandwidth per watt becomes more and more important. OCS is about making the road able to reroute itself. It reconfigures paths directly in the optical domain, so the AI fabric can dynamically schedule traffic and jobs. The two can benefit together: CPO addresses bandwidth and power. OCS addresses how the whole network is scheduled. Google’s Jupiter network has already shown the value of OCS plus SDN: roughly 30% higher throughput and about 40% lower power. The potential of this ecosystem is still growing. I think this is a lane where it makes sense to add some positions. $LITE $COHR

Latest news OCS is the next optical communications segment awaiting a re-rating after CPO. Citi sees the market potentially reaching $11 billion by 2030. It skips the optical-electrical-electrical conversion and directly dynamically schedules optical paths in the AI fabric, cutting power consumption and improving cluster efficiency. Microsoft, Meta, and NVIDIA are entering the OCP OCS ecosystem, and $LITE and $COHR are expected to lead the commercial market. The current perception gap is huge. Given how much they have already risen, I’m surprised they outperformed the cloud providers today.

Guys, the market is still trying to call the top of the $MU cycle while customers are already putting down deposits to reserve future capacity. Micron’s SCAs increased from 16 to 26, customer prepayments rose to $32B, and RPO is now around $150B. More than 75% of FY27 shipments are already committed or allocated. The visibility on AI memory demand is starting to stretch from quarters into years. Out of all these numbers, the part I care about most is the pricing framework inside those long-term agreements. Customers get supply certainty by locking in Micron’s capacity early, and both sides are doing the math. I still like the long-term memory trend. Now I’m watching contract execution, the HBM ramp, and whether all this capacity expansion ultimately shows up in cash flow. Collecting deposits looks great. At the end of the day, they still have to keep the profits.

Everyone is going to talk about the $150B buyback. I’m more interested in the Agent security platform. $NVDA releasing an open Agent security framework tells me the next AI battle is already moving beyond chips. As Agents get more autonomy, somebody has to control what they can access, what they can execute and when they need to stop. That starts overlapping with companies like $AKAM. Akamai still has a real edge-network footprint, so I wouldn’t treat this like NVIDIA suddenly replaces the whole business. But the software-security premium gets harder to defend when a company like NVIDIA starts giving part of the control layer away. AI security is becoming a much bigger battleground than I expected a few months ago. The chip war was only the first round.

OpenAI shifting away from GPT-6.1 Astra and pushing the cheaper GPT-6.1 Sol + always-on agents is more interesting than the model names themselves. Inference is getting cheaper. That means enterprise Agents are getting cheaper to deploy too. And that starts putting pressure on companies like $NOW. ServiceNow has spent years selling workflow, ticketing and automation modules into large enterprises. If cheaper Agents can handle more of those workflows directly, the long-term pressure on software pricing gets harder to ignore. I still wouldn’t treat $NOW like it disappears overnight. Enterprise customers are sticky, and replacing core workflows takes time. $MSFT is in a different position. It has exposure to OpenAI, but it also owns Azure. Cheaper models can hurt model economics while simultaneously driving a lot more inference demand into the cloud. That setup is pretty interesting. I’d rather buy $MSFT on weakness than chase it here. AI disrupting enterprise software looks more like a slow squeeze than a one-day collapse.

Everyone is focused on enterprise Agents. But the consumer side may be starting to show something interesting. Meta Muse quickly climbed to the top of the U.S. App Store free rankings, with downloads far ahead of ChatGPT at the same stage. That tells me consumers are willing to use AI that actually does things for them, not just another chatbot. For $META, that could eventually open up more than advertising. Subscriptions. Transaction fees. AI services layered on top of a massive existing user base. The advantage is obvious: distribution is already there. $SAP is almost the opposite setup. It’s focused on enterprise workflows like manufacturing, supply chains and finance. Much stickier customers, but adoption moves slower. So I’m watching two different Agent models here: $META for consumer scale. $SAP for enterprise depth. Consumer AI can move fast. The part I want to see next is whether that attention actually turns into paid usage.