$AAPL IS BUILDING A LOCAL AI STACK OUTSIDE THE CLOUD Apple is turning the Mac into a local AI workstation since new Mac mini with M6 delivers up to 4x faster AI performance, while the new Mac Studio with M5 Max or M5 Ultra scales to 512GB of unified memory, 1.2TB/s of bandwidth and an 80-core GPU. Apple is also adding Thunderbolt 5 with RDMA so multiple Mac Studios can be linked together for roughly 3x faster distributed inference. The more interesting part is that Apple is competing on memory capacity rather than raw compute. Inference is often memory-bound, so 512GB of unified memory lets a single desktop hold models that would otherwise require multiple data center GPUs, while a four-system M5 Ultra cluster comes to roughly $22,000 and creates a compelling setup for startups or research teams that want private inference without recurring cloud costs or data leaving the building. That is already showing up in the business. Mac revenue grew roughly 29% to $10.35B last quarter as developers increasingly used Mac minis and Studios as local inference machines, and these refreshes push directly into that demand. High-margin memory upgrades also help mix, while the same architecture increasingly looks like a desktop version of the AI server strategy Apple is reportedly building toward longer term. The limitations are still real. Thunderbolt RDMA remains better suited for inference than training, software support across MLX, llama.cpp and other frameworks still needs to mature, and the 512GB configuration does not arrive until late October. Add in sharply higher DRAM prices and worsening supply constraints and the next question becomes whether Apple can build enough of these memory-heavy systems to meet demand without pressuring margins.

CPU stocks rally as Meta's Muse fuels the Agentic AI trade. $INTC +13.9% $ARM +14.5% $AMD +9.1% (ATH) Meta launched Muse earlier this month. It’s now the most downloaded free app on the U.S. iPhone App Store, surpassing ChatGPT. Why does that matter for CPUs? Because AI agents do a lot more than generate text. They browse, call tools, execute code, and coordinate multi-step tasks. That can make the workload far more CPU-heavy than chatbot inference. Industry estimates suggest CPU-to-GPU ratios could move from 1:8 in training to 1:4 in inference and potentially 1:1+ with agentic AI. Citi now sees the CPU market reach $237B by 2030. But the supply is already the constraint. $INTC: Intel CEO Lip-Bu Tan says Intel can currently meet only ~50% of customer CPU demand. $INTC has also raised prices several times this year. $AMD: AMD reportedly plans ~10% price hikes in Q4. Its server CPU TAM is modeled at >35% CAGR through 2030. $ARM: Arm-based CPUs already account for ~50% of the AI host CPU segment. $NVDA: Shipping Vera, a CPU built for agentic AI and inference workloads. Agents could make CPUs the next major bottleneck.

$AMD PRICE HIKES SIGNAL A BIGGER SHIFT ACROSS THE CHIP SUPPLY CHAIN The underlying trend looks real even if the specific AMD pricing details are still only a supply chain leak. ChannelGate says AMD has told partners to expect roughly 10% higher pricing in Q4 across AI accelerators, consumer GPUs and motherboard chipsets after $TSM reportedly raised advanced node wafer pricing by roughly 5% to 10%. AMD has not confirmed the increase, but the broader backdrop is consistent with what we are seeing across the industry as foundry, memory and packaging costs all move higher at the same time. What matters most is that AMD may be willing to pass those costs through rather than absorb them. A 10% increase in wafer costs does not automatically require a 10% increase in finished chip pricing because wafers are only one part of the total cost, so pricing roughly in line with the wafer increase would suggest AMD believes demand is strong enough to protect margins. That also reinforces how much pricing power TSMC continues to hold because advanced nodes now represent the majority of its wafer revenue and higher foundry pricing gets pushed through the rest of the semiconductor supply chain. The real question for AMD is whether those increases actually stick. If customers absorb them, gross margins remain protected and the higher cost simply moves downstream. If competition forces AMD to absorb part of the increase, margins come under pressure just as the company is trying to scale its AI accelerator business. Consumer GPUs are probably the easier place to push pricing through, while AI accelerator contracts are negotiated directly with hyperscalers that have more bargaining power and more alternatives, which makes that part of the leak much less straightforward. So the cleanest read through is that TSMC likely benefits the most while AMD becomes the test of how much semiconductor cost inflation customers are willing to absorb. GPU pricing has already moved higher because of memory costs, and now wafer and packaging inflation are being layered on top. For AMD, Q4 gross margin will tell us whether the company can successfully pass those costs through, while for TSMC the bigger story is that the foundry layer continues capturing more value as the entire AI supply chain gets more expensive.

THE NEXT BIG AI TRADE MAY BE CYBERSECURITY The next major leg of the AI economy may be security because the same agents that make enterprises more productive also create a completely new attack surface. $CRWD CEO George Kurtz says the unit of threat is no longer the hacker but the autonomous campaign, where coordinated agents can execute attacks at machine speed. That changes more than attack volume. Kurtz argues that “sophistication is dead as an attribution signal” because AI can give a lone actor execution that once required nation state resources, which means defenders have to rely more heavily on identity, infrastructure and intent to understand who is behind an attack. The commercial implication is even bigger with every AI agent effectively becomes a privileged identity with access to data, applications and eventually payment rails, so enterprises will need least privilege access, short lived credentials, traceable actions and a kill switch built into the architecture from day one. That pushes security closer to runtime where endpoints, cloud workloads, SaaS and identity become the actual control points. Governance documents cannot stop an agent moving at machine speed, which is why I think security increasingly becomes embedded infrastructure rather than something companies layer on after deployment. This is where $CRWD, $PANW, $ZS and the broader identity stack start becoming more important to the AI economy. CrowdStrike has SafeMind with NVIDIA, Cloudflare is working around OpenAI models, Zscaler is working with OpenAI and Anthropic and Palo Alto is building its own AI security stack, so the market is already moving from theory into an infrastructure land grab. CrowdStrike’s advantage is that partnerships can be copied but fifteen years of deployed telemetry cannot. If every blocked attack feeds back into detection and makes the next defense better, the Threat Graph becomes more valuable as agent activity scales and security starts to look like a network effect. There is also a regulatory angle that could become meaningful. Kurtz wants AI weights, training clusters and APIs treated as critical infrastructure, which could eventually bring tighter standards, incident reporting and procurement requirements around the entire AI stack and create another durable spending layer around compliance and protection. The unresolved problem is speed. If attacks happen in milliseconds but humans still own the highest consequence decisions, the defender remains the bottleneck at exactly the moments that matter most. That is why autonomous defense will likely become one of the most important software categories of the next several years. So while the market is debating whether frontier AI should slow down, enterprises still have to secure the AI already being deployed today. That is why I think AI security becomes one of the clearest second order beneficiaries of the AI economy regardless of how quickly the next model arrives.

$AAPL JUST OPENED A NEW $2,000 IPHONE ERA Apple just gave us a much better iPhone revenue setup than we were expecting where Pro pricing moved up ~9% and Pro Max ~8% which should help absorb higher memory costs while the absence of the iPhone 18, 18e and Air points to Apple splitting the upgrade cycle and pushing more near term demand toward higher priced Pro models while Apple priced the 18 Pro at $1,199 and Pro Max at $1,299. The biggest upside surprise for me was iPhone Duo. Apple entered foldables with a $1,999 device that opens to a 7.6 inch display roughly 50% larger than Pro Max while still fitting in your pocket. I was previously thinking Duo could be around 5% of iPhone revenue in FY27, but after seeing the product I think 10% is much more realistic. Even modest migration from Pro Max into Duo can have a measurable impact on both iPhone and total Apple revenue. The other big shift is Apple clearly building its hardware around personalized AI. The company framed iPhone as the “intelligent personal hub,” with Siri using personal context, AirPods extending that intelligence hands free and Apple Watch turning health data into personalized guidance. Apple is gradually turning the devices people already carry into the sensing layer for a much broader AI experience. Put it together and this was less about another annual hardware refresh and more about raising the value of the installed base. Higher pricing supports margins, the split cycle can improve mix, Duo creates an entirely new premium tier and AI gives users another reason to stay inside the ecosystem. That is why I think the Street may still be too low on Apple’s growth into 2027.

$META IS THE MOST UNDERVALUED MAG 7 NAME I still think Meta is the most undervalued Mag 7 name because at roughly 18x 2027 earnings, the market is pricing it like an advertising business while Muse is beginning to show what a second consumer platform could look like. The real signal is not the assistant itself but the infrastructure decision behind it. Meta is giving users a persistent isolated compute environment with a browser, which turns Muse from a chatbot into an agent that can actually act on your behalf. Very few companies can afford that fleet economics at scale, and Meta can pair it with WhatsApp, billions of users and an identity graph built across its entire ecosystem. The monetization opportunity is where this gets interesting. If Muse becomes the layer that compares products, fills forms and eventually completes purchases, Meta can move from monetizing attention to monetizing intent. An agent that knows your calendar, purchase history and goals could be a higher value signal than any feed impression, with merchant placement or transaction economics potentially becoming a much larger ARPU opportunity. Unlike Google, Meta also has no massive search business to protect from that shift. There are still real reasons the market is not paying for it yet. Meta has not disclosed Muse users, retention, pricing, take rates or the cost of running dedicated compute per user, while the model still needs to prove it can handle complex tasks reliably. There is also a real risk that successful agents reduce time spent in the feed before new commerce revenue is ready to replace those impressions. That is why Q3 disclosure matters so much. Show me retention, unit economics and a business model, and the market can start valuing Muse as a product instead of another capex line. Until then, I think you are still paying for the core ads machine and getting one of the largest consumer AI distribution bets as optionality.

AI's next big test could start with these earnings. $AVGO, $SNOW, $HPE, $NTAP on Sep 2. $CIEN, $ZS, $IOT on Sep 3. $ORCL, $ADBE on Sep 10. $MU on Sep 30. Save this earnings calendar. Join 8,000+ traders in our Discord to discuss and trade around these names. ⬇️

$NVDA PATH TO $10 TRILLION The path to $10T for Nvidia is becoming less about whether AI demand exists and more about how much of that demand Nvidia can physically serve. Revenue just crossed a $385B annualized run rate while management still sees FY28 growing roughly 70% even though it says demand is closer to 100%. The customer base is also broadening at exactly the point bears expected concentration to become the problem. Non-hyperscaler ACIE revenue is already 45% of Data Center and grew 138% YoY as neoclouds sovereign AI enterprises and AI-native companies become a second demand engine alongside the major clouds. At the same time Nvidia keeps capturing more economics from every AI factory it sells into. Revenue opportunity per gigawatt has climbed from roughly $18B with Hopper to $25B with Blackwell and potentially $40B with Vera Rubin as CPUs networking software and rack-scale systems become part of the platform while agentic AI drives 15x to 100x more compute per task. That combination is why I think Nvidia has the clearest path to becoming the world’s first $10T company. The business is still supply constrained while demand diversifies and Nvidia captures more revenue from every unit of compute deployed which gives earnings multiple ways to keep compounding even after hyperscaler growth eventually normalizes.

$NVDA has beaten earnings for four straight quarters. The stock sold off every time. Nvidia reports Q2 tomorrow, with ~$91B in revenue guidance. Consensus expects $91.9B in revenue and $2.08 EPS. But the market might no longer be just satisfied with it: - Q2 is largely de-risked after four straight ~$2B beats. Investors are looking to Q3 guidance. - $NVDA has $119B in supply commitments + $26B inventory + customers' equity stakes. The capital model only holds if compute pricing stays firm. - Blackwell demand must stay strong even as Rubin ramps, without purchase delays. - The bigger test is whether demand visibility extends into 2028. Read more on why $NVDA's print could reset the AI trade:

JPMorgan’s chart looks like a web of circular AI financing but the more important takeaway is that only about $46B of the capital shown is equity while roughly $879B represents multi-year purchase commitments for compute, GPUs and infrastructure. The true circular relationships are much more concentrated with examples like $MSFT investment in OpenAI, $AMZN and $GOOGL investments in Anthropic and $NVDA investment in OpenAI sitting alongside much larger purchase commitments but there are also massive one-way contracts with no equity attached including $ORCL ~$300B OpenAI commitment, $AMD $90B supply agreement and $CRWV compute contracts with OpenAI and Anthropic which makes the broader picture look more like a capital-intensive buildout than a closed financing loop. That distinction matters because the underlying demand is increasingly showing up in actual revenue, backlog and infrastructure commitments rather than just venture funding since the labs are growing faster than their balance sheets can comfortably support so hyperscalers and chip suppliers are helping finance the buildout in exchange for equity while locking in years of future demand across compute and data center capacity. The bigger risk is therefore not circularity by itself but how correlated the entire ecosystem has become around the same assumption that AI demand keeps scaling.

$RKLB delivered record $234M revenue, a $2.36B backlog and more than $1B of new launch and Space Systems contracts, so the selloff is not about demand but about Neutron timing, weaker near-term margins and rising cash burn. Peter Beck says “the window for an end year launch is narrowing” because Rocket Lab is prioritizing qualification for reuse and high cadence over rushing Flight 1, which matters because Neutron only becomes economically transformative if it can progress from one launch to three, five and eventually many more. That added spending is painful near term but underneath Neutron keeps getting stronger as Space Systems approaches $190M quarterly revenue, HASTE expands Rocket Lab deeper into missile defense and hypersonics, and its component stack gives it more control over spacecraft economics. If Neutron works and Iridium closes, Rocket Lab increasingly becomes a full-stack space platform that can build the components, manufacture the satellite, launch it, operate the constellation and eventually monetize recurring communications services on top.

$AMC JUMPS +20% ON ITS BEST QUARTER IN 106 YEARS. AMC reported record quarterly revenue of $1.6B and an all-time high Adjusted EBITDA. CEO Adam Aron said, "In AMC's entire 106-year history, there has never been a quarter like this one." Blockbusters like The Super Mario Galaxy Movie and The Odyssey helped drive moviegoers back to theaters: - Domestic ticket revenue +11.4%. - European attendance +18% YoY. The stock is up ~45% YTD, but still down 99% from its 2021 high. Is $AMC finally back?

New York is imposing the nation’s first statewide moratorium on new hyperscale data centers using 50MW or more of power. The one-year pause takes effect immediately while the state studies energy, water and environmental impacts, though already permitted projects are exempt.

Other prop firms force options traders to adapt. Vanquish doesn't. - Trade advanced options strategies - No restrictions on spreads or hedging - One-step evaluation - Keep 100% of profit split. Get 35% off your eval with code "VANQ".