In case you missed it – my review of the Fed statement and Powell presser can be found here.

Wiz:

Google is reportedly set to purchase a large, privately owned, cloud security vendor called Wiz for $32 billion. Wiz offers a cloud native application protection platform (CNAPP) ranging from cloud configuration, to hygiene and also runtime security with its Wiz Sensor product. It builds graphic visuals of all cloud assets and interactions to provide a birds-eye-view of operations, flag threats/vulnerabilities and rank severity of each risk to prioritize work. It’s very similar to how we’d hear Zscaler or CrowdStrike explain their cloud offering.

While configuration and hygiene are table-stakes offerings for pretty much every cloud security vendor, its runtime security is considered high-quality and more differentiated. Wiz is also natively multi-cloud, and is expected to be tucked right into Google Cloud Platform’s (GCP’s) infrastructure and bolster its already world-class security, utility and cross-selling engine. It’s widely expected that Wiz will continue to operate as a seamless, omni-cloud platform rather than GCP attempting to create vendor lock. That’s not surprising considering how well entrenched Wiz already is and how popular it is for cloud security operations to be multi-cloud – especially in the new age of GenAI. 

Price Tag:

Wiz generated about $500M in annual recurring revenue (ARR) in 2024. It has goals to reach $1B this year, although many view that as ambitious. I cannot stand sales multiples, but I’m going to use them here because Wiz is private and data is extremely limited. This is what we have access to: Assuming it reaches $800M, the mega cap is paying 40x forward ARR for this company. That’s similar to the multiple it offered to buy Wiz last year and is extremely lofty, but not egregious in my mind. I think this is the wildly rare case where 40x ARR actually does make some sense. Why? First and foremost, Wiz is a private market darling in the world of cloud security. And cloud security is the highest growth, most untapped area in a group of product categories that all boast long runways. It’s the shiniest house in a neighborhood of mansions. That’s why a disappointing 2025 would yield 60% ARR growth at scale, while every other high-growth peer in public markets struggles to deliver 25%-30% revenue growth this year. While CrowdStrike and Palo Alto enjoy more scale, at $800M in ARR, the scale difference is not nearly enough to explain this premium growth. And for more exciting context, Wiz was created in 2020 and reached $1M in ARR just 4 years ago. Insane progress in just a few years. This is a hyper-grower at scale that calls 50% of the Fortune 100 its customers.

Furthermore, The Search Giant is going to operate this in a hyper-efficient manner. Wiz will shed hefty cloud hosting costs, go-to-market and brand-building expenses, back-end maintenance needs and all other redundant costs as part of integration. While Wiz probably doesn’t have the margins of a CrowdStrike or Zscaler, the margin ceiling for Wiz under this mega-cap is higher than any pure-play cyber name on their own. There are a plethora of required input costs to nurture and grow a security business that Wiz will no longer incur. Similarly to Broadcom operating VMWare at sky-high margins, I think Google will do that with Wiz too. I do not think a 50% EBIT margin post integration is unreasonable to expect for this specific segment… I don’t even think that’s the ceiling. And considering this, Sundar Pichai probably sees 40x ARR as somewhere around 80x EBIT for one of the most enticing businesses in this equally enticing sector. It’s a lot more reasonable price tag with this in mind.

This deal faces considerable regulatory hurdles and skepticism. There’s a decent chance it’s not allowed. Still, the cloud-agnostic niche should help the company’s case, while its distant 3rd place in public cloud market share should too. If it closes, that will happen next year and could prop up overall growth rates by a low single-digit percentage. I’m also a bit encouraged to see Alphabet attempt this amid Department of Justice (DOJ) investigations. Maybe they’re seeing a more laxed regulatory backdrop and the M&A spigot is turning back on?

Public Peers:

I also think this is good news for all of Wiz’s competition – Cloudflare, Zscaler, SentinelOne, Palo Alto, CrowdStrike, Fortinet etc. Those companies trade at ARR multiples (again using my least favorite valuation metric because Wiz is private) ranging from 6x-18x. All of them, besides maybe SentinelOne, likely have much better margins than Wiz does today. Seeing Alphabet value a quality firm in the space at levels higher than what the public darlings enjoy should be supportive for their own valuations. 

Finally, Wiz is an innovation machine and an intimidating disruptor. It is taking its fair share of large logos and storming onto the security scene. When large established companies buy security disruptors, the cliché is for those disruptors to move more slowly and become easier to beat. Just like when Carbon Black was bought by VMWare. I think every single company mentioned would be glad to see this purchased. It’s not that I believe Alphabet will botch this integration (far from it). I just think Wiz being part of a massive organization will inherently slow it down just a tad. More red tape… more bureaucracy.

This was a highly technical presentation that got more into the weeds than we need to on individual hardware components. It also reviewed a lot of what Huang presented at his CES keynote. I did get into the technical weeds for that review, so if you want that level of detail, it can be found in section 5 of this article.

Why Inference Needs More Compute than Training:

The first part of Jensen’s keynote was spent talking about how much more compute is needed for agentic reasoning models vs. typical large language models (LLMs) for GenAI. According to him, these reasoning models need 100x more compute than the industry expected as of a year ago. And if you think about it, that makes sense. GenAI models that are used with ChatGPT are great at scraping information from pre-training experience to give a “one-shot” answer. They cannot dissect problems, strategize various approaches to solving them, test each method and then arrive at the very best answer. They cannot “think harder” or “contemplate” by taking more time to produce a series of output tokens (the response). They cannot arrive at correct answers in multiple different ways to practice consistency checking and lower hallucination rates. While LLMs were considered cutting-edge and groundbreaking not too long ago, they’re now called “traditional” or even “legacy” models, with agentic AI supplanting them as the exciting new technology on the block.

Agentic AI models can do all of this and field far more complex, abstract, multi-step questions by delivering a series of coherent tokens. Nvidia offers a new family of reasoning models (built on top of Llama) to help developers, as well as a few newer technologies to make model building and customizing easier.

Model distillation helps transfer knowledge from teacher models to smaller student models. Reinforcement learning, or offering positive/negative feedback based on quality of answers, helps assess and steer these student and agentic models as they learn. Synthetic data generation, using Nvidia Cosmos, creates an endless series of similar scenarios from one base scenario to more rapidly and scalably drive learning. Closed-loop training ensures we can perfect policy (or teacher) models and power synthetic simulations for learning. And finally, Nvidia’s Omniverse platform for building digital twins pulls from all of these agentic tools to power its Physical AI (AI-powered robotics) opportunity. 

Nvidia doesn't just provide the GPUs to make this all happen. It provides the “scale-up” capabilities with its NVLink technology to create larger GPU clusters; it provides the switches and ethernet equipment needed to connect disparate pieces of data centers; it provides a slew of models and guardrails for building industry-specific applications; it provides a software layer that seamlessly organizes and facilitates enterprise AI work. It does everything. And by doing everything, it ensures its world-class GPUs are indispensable. By open-sourcing a lot of this work, it inspires developers to work on optimizing Nvidia GPUs to extract even more performance. In turn, this means Nvidia enjoys more market share there while cross-selling more easily everywhere else.

Jensen offered an example of token output for Llama’s 3.3 70 billion parameter LLM vs. DeepSeek’s R1 reasoning model. In total, Llama answered a question with under 500 tokens while R1 used nearly 8,600. A 17x in token consumption is obviously good news for overall revenue and profit generation per next-gen data center (or AI factory). And that’s great news for compute needs and Nvidia demand. This is a big reason why Hopper shipped 1.3M GPUs during its peak year while Blackwell will ship 3.6M in year 1. And considering Blackwell delivers 40x reasoning performance gains vs. Hopper, that should merely grow from here (until it ramps up Rubin).

“I expect data center build outs to reach $1 trillion fairly soon.”

Founder/CEO Jensen Huang

More on Robotics & Physical AI:

As a reminder, Omniverse uses its demonstration data paired with Cosmos’s World Foundational Model (grounded in physical laws) to create massive amounts of hyper-realistic simulations. This greatly lowers the requirement for expensive physical data collecting and makes building robots for factories or AVs much easier and cheaper. Simply put, with these two products, Nvidia is greatly lowering the barrier to entry for building AI-powered robotics. And again, all of this is guided by agentic AI capabilities such as reinforcement learning, while the basis of Cosmos is tied at the hip to synthetic data generation.

Nvidia also showcased “Mega” which is its “Omniverse blueprint” to let developers test and assess the quality of rules for robots at scale. And its brand new Isaac Groot N1 foundational model for humanoid robots provides more instruction and guidance for building valuable physical AI applications. Considering how reliant quality physical AI is on models firmly understanding physics, biology and cause and effect in general, Nvidia is partnering with DeepMind and Disney to create a new “physics engine” to augment the quality of Groot.

Announcements:

Announced a new collaboration with T-Mobile, Cisco and Cerberus to build full stack, high-performance radio networks for 5g and 6g applications in the USA. Cisco will also integrate Nvidia’s SpectrumX connectivity hardware into its own offering.

Nvidia and GE Healthcare are partnering for autonomous diagnostics and imaging.

Nvidia and CrowdStrike deepened their tight partnership to include Charlotte AI. With Nvidia, CrowdStrike is fielding alerts 2x faster at 2x efficiency. With agentic AI reasoning models, token outputs are also becoming increasingly reliable. So important to minimize hallucination rates in cybersecurity especially.

Huang announced the latest version of its DGX SuperPOD (supercomputer).

GM selected Nvidia to build its next-generation self-driving cars. As part of this news, Nvidia unveiled Halos as its “chip-to-development” system for autonomous vehicle (AV) safety. Nvidia is the first company to have “all code safety assessed by third parties.” With this product and Nvidia Drive (driverless technology platform), it’s highly likely that Nvidia will continue to be the backbone that powers virtually every legacy and next-gen AV program. As an aside, that’s great news for a competitive future and market fragmentation… which is great news for demand aggregators such as Uber.

Nvidia shared its “Dynamo” operating system for AI factories. Jensen compared this to the virtual machine and software layer that VMWare provides within general compute data centers. This will orchestrate agentic AI activity that draws from the massive compute capacity kept in these AI factories. It will also be open source. Anything Nvidia can do to make usage of its hardware (and acceleration of valuable apps stemming from it) easier will be good for overall demand. Enterprise AI customers for Dynamo include Accenture, AT&T, Blackrock, Capital One, Deloitte, SAP and Servicenow. They’re all building their AI agent platforms and offerings on top of this.

From a roadmap point of view, Nvidia reiterated plans to transition to Grace-Blackwell Ultra (GB300) by the second half of this year. From there, by the 2nd half of 2026, Vera-Rubin will come next. Vera, the CPU, has 2x the performance of Grace, while Rubin comes with 2x GPU linking scalability of Blackwell and more memory bandwidth. The only thing that isn’t different about this GPU is the chassis (or frame), which should lower the risk of production delays endured when it first shifted from Hopper to Blackwell. Vera-Rubin ultra will come during the 2nd half of 2027 with 2x the GPU linking capabilities of the first Vera-Rubin iteration. Feynman is the name of the GPU platform after Rubin. It also has new switches coming this year, based on a newer photonics technology that should improve overall performance and efficiency. It’s always iterating. It’s always sprinting. That’s why it is so hard for everyone else to catch this company.

It’s no secret. The Trade Desk stubbed its toe on the launch of Kokai. That led to last quarter’s disappointing results and the equally disappointing guidance. And while the stock has certainly had a horrendous run, my confidence in the company remains high. I talked through that a lot in the review, but thought I’d revisit those ideas considering the company’s continued turbulence.

TTD missed the mark on buying processes, workflows and overall user interface issues that were not well received by customers during early release. While that’s not ideal, it’s also a lot easier to fix than targeting or identification issues. The team listened to these complaints and intentionally slowed down the roll-out of Kokai to address them. It could have generated much better results this quarter and delivered better guidance if it was fixated on rushing the platform out to meet short-term targets. That just was not the focus. Instead, it decided to prioritize its long-term reputation and financial engine, which led to the misses. Modest dissatisfaction with the imperfect launch is fair, but I think the stock price-fueled sentiment has gotten astronomically too negative. In my mind, ensuring consistently profitable compounding over the coming decades matters far more than meeting a revenue number over a three-month period. If this were a private company, it would’ve been a non-event. But? Public markets punish incremental uncertainty and execution risk. That’s especially true for fundamental darlings like TTD who have trained us to expect perfection and who have been rewarded with lofty multiples because of that perfection. 

With that said, I refuse to give up on Jeff Green because of prioritizing the next ten years over the next three months. I refuse to turn sour on a leader that has been flawless for the better part of a decade and who seemingly has an ad-tech crystal ball to forecast every market transition. I refuse to think Amazon is suddenly preventing its success instead of that sentiment stemming from price action. While it’s true Amazon is focusing more on its ad network, Google is focusing less on theirs and TTD was able to overcome that competitive threat for many years. I fully anticipate that TTD’s fixes will quickly bear fruit like they’ve told us to expect. And I think this will get right back to a wonderfully boring 20%+ revenue and profit compounder in the near future. This is an elite company that is now trading for about 32x forward FCF and 30x forward earnings. After a brief period of higher investments this year, it should get comfortably back to 20%+ profit compounding.

We dream of world-class firms stubbing their toes and opening the door for long-term entry. That’s what I think TTD did last quarter. I have finally been able to halt the trimming (as Max readers know) I did multiple times on the way up, and have been buying (as again Max readers know) consistently on the way down. The value proposition is firmly intact; their relationships with leading streamers, media platforms and retailers are firmly intact. Green is still a superstar. 

While I am never willing to unconditionally support a stock, I am willing to look past one mistake within a mountain of execution. I would need to see continued signs of financial disappointment to grow sour on this name. And I am always open to seeing that. For now, I remain bullish and in accumulation mode.

Nike guided to about -13% Y/Y revenue growth, which missed -12% Y/Y growth expectations. It also guided to 450 bps of Y/Y GPM contraction, which missed estimates by 110 bps. Notably, this includes the full impact of expected tariffs in Mexico and Canada. Good decision to bake in that weakness and leave any surprise here to the upside. It expects SG&A to rise at a low-to-mid single-digit clip, with demand creation expenses expected to rise and lower restructuring charges offsetting some of that growth. Guidance includes a 2 point foreign exchange (FX) headwind, which is about as expected. In the next section, we will work through everything Nike is doing to jumpstart a company turnaround. This includes expediting some product phase outs, eliminating some digital discounting and liquidating more stale inventory through outlet stores and wholesale. These are all financial headwinds that will be the most severe next quarter. Results should brighten from there, as sell-siders expect in about 12 months.. Again… There's so much to fix and it’s hard to overstate how poorly run this company was under the old team (and the team before that one). 

One more note on guidance. Due to less digital channel discounting, it expects -10% Y/Y digital traffic growth for fiscal year 2026 (which starts in June).

Nike trades for 38x forward earnings. Earnings are expected to fall by 46% this year, grow by 2% next year and grow by 33% the following year.

Table Setting:

Last quarter, new CEO Elliott Hill laid out his various priorities and plan of action for righting this troubled ship. 90 more days into his tenure, following meetings with teams and partners across the globe, he is incrementally more confident in this being the correct approach. They call it the “Win Now” strategy. Most of the call was spent talking about everything Nike is doing to turn the tide, with plenty of reminders that momentum will take more time to build and offset current weakness. Every piece of the plan is expected to form a compelling recipe for sustainably profitable growth as we move further into fiscal year 2026.

Turnaround Plan Progress – Faster and More Impactful Product Innovation:

Nike continues to urgently work on filling its product pipeline with more compelling offerings while it de-prioritizes older franchises that just aren’t working as well anymore. Starting with actual product assortment. Running performance enjoyed some successful early launches and was called a standout for its performance category. The Peg-41 enjoyed healthy volume growth while the Pegasus Premium sold out across North America and will grow throughout the rest of this year. 

In sportswear and lifestyle, things are still more challenged. Classic franchises like Air Force 1, Air Jordan 1 and Dunk are the largest examples of Nike accelerating established brand phase-outs in favor of more innovation and product newness. This is where that decision is having the sharpest financial impact and is exacerbating the current fundamental weakness. Nike is not eliminating these brands, it just has way too much of them on hand and is greatly pulling back on new orders. Its Vomero 5 (for lifestyle unlike the Vomero 18) doubled revenue Y/Y while its NIKE Shox lineup continued its rapid growth. Its new Air Superfly shoe is garnering significant social media attention early on. Finally, it teased a new Air Max platform coming next year, which leadership was clearly excited about.

For apparel, like everywhere else, it is determined to drive fresher assortment to cater to evolving tastes. Its 24/7 collection “brings performance materials to high-style training to tell a deep technical apparel story.” A little fluffy, but what isn’t fluffy is an opening month of sales that comfortably tacked ahead of expectations. It’s now scaling into this strong demand. And as previously announced, its new collaboration with Skims will feature “style-led product that sculpts and performs” and will launch on Nike Direct and Skims Direct next quarter. This relationship will start in North America and expand globally from there.

“The apparel space is right for fresh thinking, and I've asked our team to keep innovating across the spectrum of performance and style, and to seek out white space in the market to complement our brands and product portfolio.”

CEO Elliott Hill

This all sounds encouraging. So why are the results so bad? Nike is determined to have a stable of “seeds” and established brands, creating a consistent series of growth curves to power revenue. It has not created enough “seeds” over the last few years to re-grow a healthy forest of established brands. So? All of these positive anecdotes about individual products doing well early on are just not enough to offset declines within its core franchises – mainly legacy footwear. The Vomero 18, for the more casual runner, launched with wholesale partners in 1,800 stores, with plans to double distribution by the end of next month. Running and pockets of training are currently the bright spots within the healthier performance category.

“While we added innovation across our five key fields of play this quarter, it's not enough to offset the continued headwinds of our classic franchises.”

CEO Elliott Hill

We got our first hint of when new product launches will become material. According to the team, fall order books across North America, EMEA and Asia Pacific + Latin America are delivering enough new product growth to “almost offset” the rest of the portfolio. More patience will be needed, but this forward-looking demand metric does bode well for revenue growth turning positive once more as we exit this calendar year.

Turnaround Plan Progress – Product Messaging and Brand-Building:

Nike strayed too far away from its core of “placing athletes at the center” of everything it does. That sounds eerily similar to Bob Iger at Disney talking about “placing creativity back at the center” of the business. Nike (and Disney) deprioritized the core value proposition too much and lost needed focus. It is determined to recapture that focus through more relevant product newness, assortment breadth and messaging.

As a reminder, it’s further segmenting teams to more granularly address specific needs. Now, not only are teams split by sport, but also gender, with a hope that tighter focus will lead to more impactful work. Part of this entails “igniting the ground game” in key cities across the USA, China and the UK. This past quarter, that meant hosting youth basketball tournaments for the All-Star weekend and opening pop-up stores across New Orleans for the Super Bowl along with Nike and Jordan houses. It thinks this heightened ground game go-to-market focus led to its San Francisco store enjoying its best revenue day in a decade.

Next, it got too focused on marketing for its own channels, rather than using these dollars for overall brand building. That’s changing. The company ran its first Super Bowl ad in 27 years and had a dynamic campaign for Jalen Hurts (cleverly named Love Hurts) ready to go when the Eagles won. Finally, in an attempt to deepen relationships with younger customers, Converse introduced the Shai 1, which was designed by Shai Gilgeous-Alexander, who is the new face of Converse Basketball. It will launch in the fall.

Turnaround Plan Progress – Repair the Broken Marketplace:

As a reminder, Nike is repositioning its direct digital channel as more of a premium shopping destination. It got a bit careless with endless promotions and deep discounts, which diluted the quality of its brand and burned some relationships with wholesalers. This led to cutting promotion days and the level of discounts, which meant zero promotion days between January and February vs. 30 days last year. This obviously didn’t help growth. But? This is the kind of headwind that lasts four quarters, with normalizing comps inevitably coming thereafter. I think this is a correct pill to swallow, reinforcing the sustainable quality of a brand, rather than chasing quarterly sales through discounts that are structurally harmful to the company’s app. 

At the same time, it is temporarily doing more discounting in outlet stores and wholesalers to more quickly work through inventory… This is why GPM continued to plummet regardless of the digital shift. For now, the net impact of all of this is lower margins and slower revenue growth. In December, growth was actually positive, as it had not yet implemented all of these changes. For January and February, once changes had been made, growth slowed to double-digit Y/Y declines. At the same time, despite more outlet store deals, this also meant the percentage of revenue from full-price sales rose several points Y/Y. For Q4, it expects its classic footwear revenue contribution to fall by over 10 points as a percentage of total shoe sales. Progress is slowly being made.

The other part of the marketplace that needs help is wholesale. The company fixated too much on pushing maximum sales through its direct channels, and again, irrationally discounted to make sure that happened. This burned the trust of key wholesale partners, which Hill is determined to rebuild. They are pivoting to making sure the right product is in front of the right customer regardless of where that customer is. They’re not unnaturally steering traffic to their own channels via unsustainable pricing (which eliminated the margin advantage from direct to consumer anyway).

Nike is now taking a much more hands-on approach with these valued partners and making sure they know they’re a core piece of Nike’s future. They’re building co-go-to-market and revenue plans, optimizing displays, communicating on product roadmaps and unlocking more high-value inventory for these partners. For Foot Locker, securing access to a few shoe launches led to it having “lines down the block all weekend.” That’s how you motivate your partners to support your success.

Another ugly quarter. The core pieces of this business are struggling, and years of poor innovation and execution mean the product pipeline is not ready to fill the void. It’s nice that new launches are working, but it needs several quarters of this to happen before that can offset how poorly other parts of the business are performing. 

I view this company similarly to Starbucks. An iconic and beloved enterprise that has endured years of terrible leadership. There is no quick fix. While I do think Hill is the right person for this job, Nike is a complete mess and it will take several quarters to see progress. I think there’s a decent chance that I start a new position in the coming quarters and still see this as a globally iconic brand with boatloads of potential. I think it just needs a solid chunk of time with effective leadership, and I think it’s now getting that. The priorities are well-placed, the green shoots are encouraging and the road ahead will be bumpy.

This is not part of the core coverage network. I was able to find some time to read through the report and decided to expand this quarter’s earnings snapshot with a bit more detail. This is not as detailed as an earnings review (although I’m sure it’s more detailed than what others would call a “deep dive”).

Micron sells semiconductors for memory and storage. Its “Not And” (NAND) chips offer non-volatile data storage, which maintains stored information when a system’s power is turned off. Dynamic Random Access Memory (DRAM) offers volatile memory storage for personal computers, data centers, and mobile devices. This makes sure other processors have the context they need at any given time to minimize processing latency. These chips are considered to be at least partially commoditized at this point, with Micron’s cost advantages providing its edge.

These chips provide the foundation for its solid-state drives (SSDs), which are used in things like USB flash drives. Micron sells standalone DRAM and NAND processors and also SSDs with these chips in them. SSDs replace hard disk drives (HDDs), as they’re more power efficient, durable and resilient. It provides basic memory cards for things like gaming devices and cameras too.

Micron offers high-bandwidth memory (HBM) hardware that helps enable the vast data processing needs of Gen AI. This vastly expands data processing capabilities and the ability to pass context between CPUs and next-gen GPUs. Nvidia is a big customer and uses this in its Blackwell and future Rubin systems. It also offers higher-capacity SSDs to help with LLM storage.

Micron beat revenue estimates by 1.9% & beat its guidance by 1.9%.

GPM weakness was solely due to NAND industry pricing and capacity utilization. DRAM helped offset some of this decline, which means its most important products for growth are still coming with healthy margins. The pricing environment for NAND has also improved just a bit lately, but remains somewhat challenged.

Micron expects calendar 2025 DRAM bit (units of memory) growth in the mid-to-high teens range, with NAND bits rising in the low-double-digit range. Micron also expects to maintain market share in both categories and thinks supply growth will lag demand growth, which should support pricing. At the same time, it is still working through GPM headwinds from mix-shift and NAND underutilization (more later), which is why GPM is expected to fall Q/Q next quarter. It reiterated plans for Q/Q GPM expansion starting in Q4 of this fiscal year. Finally, it reiterated plans for $14B in CapEx for FY 2025.

For tariffs, they could have a small impact on costs, but Micron plans to pass all of that impact on to customers.

Micron trades for 11x forward earnings. EPS is expected to grow by 433% Y/Y this year, by 60% Y/Y next year and by 4% Y/Y the following year.

Micron thinks it’s in the “best competitive position in its history.” It is taking market share across all of its high-margin categories and rapidly innovating, ensuring it stays ahead of the pack.

HBM:

Within the highly promising HBM opportunity, revenue rose by 50% Q/Q to cross $1 billion for the period and surpass internal expectations. Its current HBM-3E product offers 30% better power efficiency vs. competing models, and its “12-high” iteration of this HBM-3E product offers even larger competitive leads in power, memory capacity and overall performance. It has begun scaled production of this product, and thinks this will “comprise the vast majority of HBM shipments during the 2nd half of calendar 2025.” HBM is expected to generate several billion in 2025 revenue, as Micron again raised its TAM estimate for this niche to $35B. The 12-high system will be an integral part of Nvidia’s Grace-Blackwell 300 system, while Micron has already started shipping units to a “third large customer.” HBM generation 4 will begin to ramp production during 2026, and offers 60% more bandwidth than its best-in-class HBM3E offering.

“High-performance processors are starved of memory bandwidth. HBM memory provides the bandwidth necessary to leverage these powerful processors in the most effective and efficient manner, and we are excited to see the growth opportunities ahead for this complex and high-value product category where our customers now recognize Micron as the HBM technology leader in our industry.”

Micron CEO Sanjay Mehrotra

It’s also working on a new DRAM node technology using extreme ultraviolet technology (called 1-gamma), which is expected to deliver 20% power advantages, 30% density boosts and 15% performance gains vs. its old 1-beta process.

For review, Micron is sold out of HBM capacity for 2025 and seeing “strong demand” for available supply in 2026. It remains on pace to reach HBM market share equal to its overall DRAM market share in Q4.

DRAM Capacity:

DeepSeek-Based Cost Deflation:

Micron spoke very positively about inference cost deflation and the impact on overall demand. Not surprising but good to hear.

“Hardware improvements, along with more efficient algorithms and software, drive down the cost of inference and make generative AI-based capabilities more accessible to new applications and use cases. This broadening deployment creates a powerful growth vector for aggregate AI demand, and recent innovations and those in the pipeline from key contributors to the AI ecosystem will continue to fuel this growth trend.”

NAND:

NAND demand continued to moderate in Q2 via channel inventory resets. It does think bit (unit of memory) shipments will start to grow again in the near future and considering its record-high market share here, it’s well-positioned to benefit. As briefly mentioned, NAND output is down around 15% from cycle peaks. It is underutilizing capacity, which is weighing on gross margin as it repurposes some of this capacity for HBM and its AI-related bucks. It sees a 10% decline in NAND capacity exiting FY 2025 and will keep tightly managing supply availability.

Devices:

It thinks personal computers (PCs) as a market will rise by roughly 5% this year. That’s supported by refresh cycles via Windows 10 end of life this October and more upgrade motivation via new AI-powered tools with newer models. These AI PCs require 33% more DRAM capacity than older models, which should be yet another demand tailwind for that revenue segment.

It continues to expect low-single-digit smartphone growth this year as well. 

In automotive, robotaxi platforms need about 20x-30x more DRAM gigabytes than traditional cars. It has the industry’s first SSD product for driverless cars, which is now testing with customers.

As someone who doesn’t follow this name very closely, it is clear to me that they will continue to lead the HBM portion of the GenAI hardware revolution. Nvidia works closely with them and Micron continues to sell everything it can make within this category. That should support growth with solid margins for as long as this hardware demand boom lasts. When will it end? That is the trillion-dollar question investors need to ask themselves to get comfortable with investing in a highly cyclical business like this. And that is why Micron and this sector overall are in my ‘too hard” pile. 

There was a ton of repeating from PayPal’s investor day. My review of that event can be found here.

A Reminder on Restructuring its Product Suite:

PayPal Open is PayPal’s unified suite of checkout options ranging from PayPal to Venmo to Pay Later and Fastlane. It had been running fragmented go-to-market for each checkout offering, and has since unified this effort under this new label. It is the company’s cohesive umbrella of checkout offerings that enables merchants to onboard all of these products with fewer steps and work. PayPal is in the process of sunsetting all legacy checkout flows and integrations to bring merchants onto its latest and greatest PayPal experience, and all other experiences it offers via this product. That will take a few years, considering how poorly managed and maintained all of these integrations have been for more than a decade. 

Merchants can easily utilize PayPal Open through its new commerce API, which also features data on all PayPal users for merchants to leverage for checkout experience personalization. Whether it’s more rewards, more prominent display of preferred checkout options or just upgraded flows (especially on mobile), PayPal has already delivered a 1-4 point uplift in shopper conversion rates. Eventually, through its Payment Ready API, it will augment this personalization by IDing returning customers in real-time to guide merchants on how to perfect displays on a by-shopper basis and to seamlessly pre-approve BNPL customers. This is now in early testing and, between this initiative and its personalization work overall, it sees a lot more progress to make. And? That progress should come more consistently and quickly following this checkout platform overhaul and move to PayPal Open/Commerce API. It will ensure all merchants are on its latest integration, vs. less than 50% total today, so that onboarding updates will be rapid and seamless, rather than one painstaking manual task after another.

Keller reiterated expectations to have 80% of its merchants on the latest integration by 2027. It started with the cohort of U.S. merchants on its latest integration (30% of total) and is now moving to the rest of its U.S. customers. From there, it will move to Germany and the U.K. where about 50% of merchants are on its latest checkout. That should mean a faster ramp. When most of its merchants are on this new flow, it expects to more aggressively sunset older processes and push the stragglers to this new product. Importantly, it is not offering direct incentive to drive this volume. Instead, it is finding communication of the value prop (higher conversion) and demoing the new flows as more than enough to convince merchants.

Sources of Branded Acceleration were Reiterated:

Apple Pay Competition:

Concern is rising over Apple Pay extending to non-Safari browsers. Based on the data PayPal has available, they have seen zero impact on market share from this. Generally speaking, PayPal now thinks its web and mobile checkout options are both world-class. Mobile had the most catching up to do, and it seems like that has happened. I know the checkout I now use for PayPal has quickly gotten much better on my phone. This, paired with its offers and ads platform, and its suite of value-added services continues to shift conversations with merchants to more holistic value creation partnerships rather than transactional button arrangements.

Fastlane:

PayPal is still not focused on directly monetizing Fastlane. It is using the Meta playbook here and wants to first build critical mass and knows it will indirectly benefit from the uplift to overall market share, as well as engagement uplifts seen from early users. Fastlane’s popularity and powerful channel partnerships are also turning this product into a great top-of-funnel tool for cross-selling its other services like fraud management, chargeback protection and everything else as another indirect perk. It’s making prospective merchants more interested in revisiting what PayPal can do for them, which is uniformly positive for the company. And while I do love indirect value creation, I fully expect this to be directly monetized as another service down the road. When PayPal can deliver such dramatic boosts to checkout conversion like it does here, that value should be seamlessly monetizable.

Braintree:

As part of the PayPal Open rebranding, Braintree will now be called PayPal Enterprise Payments and Hyperwallet will be called Enterprise Payouts. I’m a fan of simplicity and naming products after your already ubiquitous brand. Good decision in my mind. No reason to support awareness for brands beyond PayPal and Venmo.

Keller reiterated that PayPal Enterprise payments delivers a 5-point boost to authorization rates vs. leaders such as Stripe and Adyen. That has everything to do with its vast customer base and the detailed data profiles (hello superior checkout personalization) it has on each of these users. And? Pairing this with PayPal Open’s connection to over 100 payment service providers globally means PayPal can orchestrate these private label transactions in a more efficient manner than others.

The company is “still competitive on price” but no longer attempting to always be the cheapest to chase that “vanity metric.” As a reminder, the pricing-to-value change is currently a roughly 5-point headwind to revenue growth, but a 1-point tailwind to transaction dollar growth. A lot of the “tough conversations” with merchants on pricing updates have now been had. Zero merchants churned, and it has also already recovered some of the lost volume from these clients thanks to its service suite and superior performance. This is what we want to see… private label processing competing based on value proposition rather than predatory pricing.

In terms of growth levers, international growth, omni-channel processing through partnerships with companies like Verifone and localizing product offerings by industry (very similar to what Block is doing with Square) are core priorities. 

Other Value-Add Services (OVAS):

PayPal is now packaging its risk-management product for volume processed by other players such as Adyen. This should be a nice unlock for this service’s traction.

It’s finding much more demand for its payment orchestration product than expected within merchants doing $4 million to $50 million a year in revenue.

“I think for OVAS, we are just getting started. We have not had highly focused investment into this. Some of the services are best-in-class but we haven’t exposed them to the outside.” – Frank Keller

The Mexican peso and Brazilian real have strengthened from cycle lows by about 5% and 9%, respectively. Most of this strengthening happened after companies issued Q1 2025 guidance. Meaning? Currency headwinds baked into forecasts are probably lighter than feared. Even for a company like Uber, where only part of its business is in Latin America, this has been a multiple point headwind to overall growth and the source of some estimate misses in recent reports. The opposite should be true this quarter if these trends hold. This is great news for headline growth rates for Uber, Nu, Mercado Libre and every company collecting revenue in those currencies while reporting in U.S. dollars. The list of beneficiaries is quite long. Think Latin American-domiciled or globally-oriented businesses like Airbnb, Duolingo, Apple, Spotify, PayPal, Lululemon etc.

Tesla secured one of many needed approvals to launch robotaxis in California.

Amazon is offering its new Trainium chips at steep discounts to Nvidia. Important to note Amazon is buying all of the Nvidia GPUs they can find and the two remain very close partners. Still, it has been working on these chips for use cases that can get away with having something other than the very best option. Sometimes customers care more about lower cost than higher performance. That’s what this is for.

JP Morgan upgraded Cava to overweight due to growth optimism. They hosted Cava for private investor event and clearly liked what they heard.

Morgan Stanley sees Lululemon revenue outperformance driving profit outperformance when it reports Q4 results next week.

Meta began the rollout of Meta AI in Europe.

I sent out my coverage of the Fed statement and Powell presser during the week.

Output Data:

Consumer & Employment Data: