80-hour work week in the books. In case you missed it, here's what was sent since Monday:

Other earnings reviews sent so far this season:

6 more coming next week. Can't wait.

Demand:

Margins:

Balance Sheet:

Q4 Guidance & Valuation:

SPOT trades for 44x forward EPS. EPS is expected to grow by 37% this year, 64% next year and 27% the year after that.

Demand:

AppLovin beat revenue estimate by 4.8% & guidance by 5.6%.

Profits:

Balance Sheet:

Guidance & Valuation:

App trades for 47x forward EPS. EPS is expected to grow by 50% next year and by 32% the following year.

Demand:

Profits:

DDOG beat EBIT estimates by 15.6% and beat guidance by 16.3%. It also beat $0.46 EPS estimates by $0.09 and beat guidance by $0.010.

Balance Sheet:

Guidance & Valuation:

These Q4 raises are especially notable considering DataDog generally bakes in an extra degree of prudence into its forward guidance.

DDOG trades for 86x forward EPS. EPS is expected to grow by 10% this year, 15% next year and 24% the following year.

Demand:

Axon beat revenue estimates by 0.8%. Its x% 2-year revenue CAGR compares to 33.6% Q/Q and 32.7% 2 quarters ago.

Profits:

OpEx rose by 40% Y/Y due to accelerated R&D investments and some cost pressures from tariffs. That was materially larger cost growth than expected. They don’t see tariff headwinds getting worse from here.

Balance Sheet:

Guidance & Valuation:

AXON trades for 87x forward EPS. EPS is expected to grow by 6% this year, 21% next year and 28% the year after that.

For a full overview of all the product news announced at the March 2025 Robinhood Gold event, click here.

More growth metrics:

Robinhood increased its annual OpEx guide by 2%. Meeting performance targets and higher payroll taxes from the soaring share price led to this raise. Prediction market and Robinhood Ventures investments added a bit to cost growth as well.

 So far in October, annualized net deposit growth rate is 20% vs. 29% this past quarter. Its margin book growth only slowed modestly from 153% to a still great 150% in October. Equity volumes accelerated from 126% Y/Y this quarter to 150% in October. Options contract growth acceleration from 38% to 60% Y/Y. Crypto growth moved from 300% Y/Y to 176% Y/Y as Bitstamp M&A keeps helping and underlying trends remain solid. All of this leads to consensus revenue estimates calling for 32% Y/Y growth.

HOOD trades for 50x forward EPS. EPS is set to grow by 25% this year and by 8% in 2026 and 2027. The reason for that sharp assumed slowdown is an expectation that comps get a lot harder and tailwinds supporting its large retail crypto and options businesses don’t remain quite so strong. This is the kind of business where modeling profit estimates two years from now is especially difficult as nobody today can predict how cyclical factors will play out in 2027. They do buffer their inherent cyclicality with strong market share gains and some level of product diversification, but still, 45% of their total revenue is from options and crypto. That is the risk here, with the hope being that other products proliferate and make them less reliant on these buckets. Here’s the big HOOD risk summed up in one chart:

Some of these new products below are still entirely tied to crypto and options, but things like futures and gold subscriptions aren’t. This is the bull case – product diversification: 

Priority – Win Active Traders:

As it does every quarter, Robinhood launched several new products aimed at making it the best place for active traders. Leadership says it all the time. They want these active traders to feel “disadvantaged” by not being on Robinhood. Those are the customers that deliver the most transaction volume – hence the strong focus. 

This quarter, they announced Robinhood Social as a Stocktwits-like destination for exchanging ideas. While that may not sound like a great revenue tool, it is. This should be a positive for engagement, which will feed retention and lifetime value.

On the Desktop Platform (Robinhood Legend), they launched new indicators and scanners, with plans to consistently launch more tools going forward. They added new market data tools and potential option trade profit visualization. More updates include multiple individual brokerage accounts and highly anticipated shorting, which is coming soon.

The prediction market ramp is also going very well. This product is the fastest ever to reach  $100M in annualized revenue. That makes sense, as it effectively offers online legal sports betting to Americans that didn’t already have it, but it’s still impressive. October looked very good as well, with total volume eclipsing all of Q3. They’re hard at work on offering new contract categories. The only caveat here is that Polymarket, DraftKings and eventually FanDuel will all soon be launching their own sports prediction markets. DraftKings only has a sportsbook offering in 25 states and is about to launch sports predictions in the other 25 where it wasn’t allowed to operate until now. It will be interesting to see how this growth curve looks once there’s real competition.

The clearest sign of effectively winning active traders is asset class market share. Those trends look great for Robinhood:

“We have more. We have more for you guys. We cannot wait to share more next month at our first-ever AI event on December 16th.” – Founder/CEO Vlad Tenev

Priority – Go Global:

Global expansion is another high priority. Partially thanks to Bitstamp M&A closing, they now have 700,000 customers in Europe, with a rising institutional base responding well to its beefed-up crypto offering. It has 400 U.S. stocks offered in the EU through tokenization, which is up 2x this quarter. They’re still in the process of building this out, with plans to eventually unlock secondary trading through Bitstamp’s infrastructure, as well as self-custody, collateralized lending, margin etc. They charge just a 10 bps FX fee, yet this volume comes with a slightly higher take rate than their U.S. equities business. Good news. They see this as a massive opportunity, with hopes that non-U.S. grows to 50% of total revenue (with a 50%+ institutional business skew) over time. Not only would that mean more revenue, but less cyclical revenue compared to retail.

“Ten years from now, the aim is to have over half of our revenue be outside the U.S. and also cut another way. Right now, we're majority retail. We think we can get to over half being non-retail, institutional.” – Founder/CEO Vlad Tenev

Grow Wallet Share:

Its third core priority is wallet share expansion. Robinhood Banking will be a main ingredient, with the ultimate goal of establishing more direct deposit and primary banking relationships with their customers. The rollout here will be very quick and complete in the coming weeks. This will not be like their Gold Card business, as there’s not brand new credit risk that makes prudence the necessary decision.

Robinhood Gold:

Gold continues to deliver a 3.5x retirement adoption boost alongside a 20% lift to deposits and 400% spike to AUC. They are accelerating credit card issuance based on encouraging repayment trends and following a rightfully conservative rollout. While 14.4% adoption is yet another quarterly record, new customer cohorts are opting in at 40%, pointing to plenty of runway remaining.

Private Markets:

Robinhood is going deeper into private markets. They already offer IPO access to non-accredited investors. They now plan to offer venture-level access. It’s likely coming soon with an application already sent to the SEC. They’re hard at work on structuring this so that investors get needed levels of company information and the companies listed on the platform can also benefit.

New CFO:

Jason Warnick is retiring. Shiv Verma, who has been with the firm for 7 years, will replace him. He’s really not going to change much, with Robinhood in a strong financial position and continuity the best path forward.

“In terms of what to expect, big picture, more of the same. Our top goal is still to grow and to keep delivering for customers, to ship amazing products with high velocity. We also believe in a lean and disciplined culture, and this is personally where I spend a lot of time. We obsess about capital allocation and ROI.” – New CFO Shiv Verma

Very pretty numbers for sure. I just worry about the durability of these trends heading into 2026 and beyond. Robinhood is still reliant on retail options and crypto markets remaining red hot to keep delivering fast enough growth to appease shareholders. I do not think they can diversify their business fast enough to create meaningful insulation from cyclical ebbs and flows. It’s great that 11 products are contributing $100M+ annually, but still, 45% of its revenue is crypto and options trading. That’s quite hard to ignore. Currently turbocharged growth is supporting explosive fixed cost leverage. And that leverage is what’s pushing margins high enough to make a 22x forward sales look reasonable. That reasonableness requires net income margin to remain exceedingly high, which, to me, is dependent on this cycle continuing to rock and roll.

Nothing about my perceived Robinhood margin fragility means I think they’re shady or unimpressive. Far from it. I think this team has executed admirably and shipped good products at a world-class cadence. I just don’t think that will free them from the ups and downs of their industry. I feel as though the forward multiple treats this firm like a purely structural and hypergrowth SaaS name at this point. And they’re certainly not. As comps get harder and other things like prediction market competition kick in, I question whether or not this will be growing above a consensus 17% Y/Y clip for 2026. If macro tailwinds remain intact? They should do much better than that. If not? They could struggle to get there. We’ll see. Either way, not my favorite investment. 

AMD designs CPUs and GPUs, offers software and provides connectivity equipment for a wide array of cloud and enterprise clients. If there’s one thing the semiconductor industry loves, it’s constantly changing the names of products with a swarm of acronyms for us to juggle. Fun, fun. Those acronyms all fall into neat categories: chips, networking and connectivity, and software. What matters? AMD’s positioning within these areas and how it’s integrating them with its rack scale data center offering called Helios. Not that they’ve memorized what an MI350 HBM3E chip stands for.

GPU: Graphics Processing Unit. This is an electronic circuit used to process information and data. The accelerated compute needed for GenAI apps and models pulls from next-gen GPUs. It thinks its “MI” series of GPUs (part of the “Instinct” product family) boasts best-in-class memory and bandwidth, which Nvidia would certainly disagree with. AMD also thinks its 2025 Instinct release will compete with Nvidia’s world-class Blackwell platform.

CPU: Central Processing Unit. This is a different type of electronic circuit that carries out assignments and data processing. CPUs fall in the general compute bucket. General compute CPUs are still optimal for static, step-series and instruction-based tasks. They’re also much cheaper than deploying next-gen GPUs when they can work for the specific use case. AMD’s new AI data center CPUs “extend leadership in performance per watt and dollar.” AMD's family data center CPUs is call EPYC. The latest iteration of EPYC is called Turin.

NPU: Neural Processing Unit: Used for AI-enabled personal computers (PCs).

TOPs: Tera Operations Per Second. This measures NPU performance, with more TOPs being better. TOPs superiority is imperative for running Copilots and GenAI apps on PCs with optimal latency, hallucination rates and performance.

“Given what we see today, we see a very good demand environment into 2026.” – CEO Lisa Su

Data Center – Accelerated Compute GPUs:

Everywhere you look for this segment, momentum is notable. The MI300 Instinct GPU product line continues to perform very well, as deployments are “broadening.” Customers like IBM and Cohere are using it to train LLMs, while Luma and others lean on it for inference cost and performance advantages. MI350’s ramp is accelerating and yielding market share gains with neocloud players like DigitalOcean and Vultr in this important segment. Oracle just launched the first MI355X hyperscaler cloud instance and Cisco will use that product for their AI infrastructure as well. In terms of what’s next, the MI 400 family will launch next year and is already seeing “rapidly building customer momentum.

“The MI400 series combines a new compute engine with industry-leading memory capacity and advanced networking capabilities to deliver a major leap in performance for the most demanding AI training and inference workloads.” – CEO Lisa Su

AMD is hard at work on advancing its open-source software offering (ROCm) for GPUs. This area is vital and somewhere that Nvidia excels. AMD is starting to advance as well, as they “resonate with developers” across platforms such as Hugging Face (open-source AI tooling company) and drive traffic. That traffic entails optimizing performance of deployed AMD GPUs, meaning successful scaling on the ROCm side will make this company more competitive with Nvidia’s GPUs as well. They launched ROCm7 as their latest version. It’s delivering up to 4.6x inference performance and 3.0x training performance gains, and there’s plenty more where that came from in terms of upcoming iterations to accelerate GPU productivity. They expect this product to be extremely competitive with Nvidia’s leading offering.

MI 400 will be a vital piece of the Helios launch and 2026 ramp.  As a reminder, this is a rack-scale AI platform from AMD designed to support up to 72 MI400 family GPUs in one server rack. It relies  on “ultra accelerator link” (UALink) connectivity equipment, which is an open standard version of Nvidia’s NVLink that is co-developed by several tech giants. It combines the best of AMD’s GPU, EPYC CPUs, and its latest Pensando Network Interface Cards (NICs). AMD calls these data processing units (DPUs). DPUs are high-performance compute accelerators that assume some of the work from generalist CPUs to allow those CPUs to focus on main tasks. Simply put, Helios is a large advancement in AMD’s ability to provide end-to-end systems for AI with better interoperability and performance. With this launch AMD will own a lot more of that product delivery than in the past. They expect most MI400-level demand to come from this rack-scale product.

“Development of both our MI400 series GPUs and Helios rack is progressing rapidly, supported by deep technical engagements across a growing set of hyperscalers, AI companies, and other partners.” – CEO Lisa Se

The talent acquired from ZT systems has been instrumental in building Helios. Speaking of ZT Systems, AMD sold the manufacturing arm to Sanmina and will use Sanmina as the leading Helios manufacturer. This will allow AMD to move faster in scaling this offering for large customers like Oracle, which will be the first to offer this product as part of a 50,000 GPU compute cluster next year. 

Overall, the AI segment is on pace for “tens of billions” in annual revenue by 2027, with signs of that being attainable looking good thus far. Other GPU wins this quarter:

“From everything that we see, we think the AI compute TAM is going up… 500 billion sounded like a lot when we first talked about it, but we think there is a larger opportunity for us over the next few years, and that's pretty exciting.” – CEO Lisa Su

Open AI:

Leadership is understandably excited about that new deal. They did have to give up warrants for up to 10% of their company based on deal execution, but the revenue contribution is enormous. This could eventually generate $100B in total business over several years (current annual revenue is $33B for context) as long as the contract is honored by OpenAI.

“With OpenAI, we are planning multiple quarters out, ensuring that the power is available, that the supply chain is available.” – CEO Lisa Su

Data Center – General Compute CPUs:

While GPUs rightfully get a lot of buzz, CPUs are not going anywhere. GenAI and agentic AI need general compute. Agents are highly complex pieces of software that run on accelerated compute and require GPUs to enable automated multi-step work across various parts of the digital ether. At the same time, the basic tasks within these complex workflows are most efficiently completed by CPUs. They can do this basic, formulaic work for less energy and money than a GPU can. Meaning? If more agents are being deployed and their productivity sequences are getting longer and more advanced, there will be more basic tasks within all of that traffic requiring more and more CPUs. This is why CPU demand signals have been outperforming for a few quarters now and why those signals should remain robust well into 2026, per leadership. 

“Supporting increased demands from AI is serving as a powerful new catalyst for our server business… I think we expect the CPU demand environment in 2026 to be, let’s call it, positive. I do feel like the trends we are seeing are durable.” – CEO Lisa Su

The 5th generation EPYC processor (called Turin) set a quarterly record and generated half of total data center CPU sales. Sales of the 4th generation EPYC model also remain strong, as demand across cloud and enterprise shows zero signs of dissipating. EPYC cloud instances rose 50% Y/Y and “many customers” are looking to greatly expand their CPU ordering plans with AMD. Overall cloud adoption of EPYC 3Xed Y/Y, with its dominant on-premise positioning giving it a large edge with customers embracing a hybrid footprint with EPYC. In enterprise, as of two quarters ago, they were hoping to cross 150 Turn platforms in the market around now, and they crossed that goal comfortably with 170 in circulation across Dell, Lenovo etc. Overall sell-through rates meaningfully improved and AMD again took CPU market share. Their total cost of ownership and improving go-to-market is proving difficult for Intel and others to compete with… quarter after quarter… year after year.

In 2026, the 6th generation Venice EPYC processor will debut. It’s expected to come with “substantial performance, efficiency and compute density gains.” Preliminary interest in the product is “stronger than ever seen.”

Non-Data Center Segment Highlights:

Ryzen PC CPUs within the client segment are enjoying strong momentum. Quarterly sales set a new record. Its desktop CPU specifically enjoyed “record channel sell-in and sell-out” powered by superior performance, as overall customer demand picked up substantially. Overall sell-through rates for Ryzen rose by 30% Y/Y thanks to some “large wins” and a strong demand environment.

Its Embedded revenue segment is pacing for a consecutive year of record-setting design wins. They’re currently at $14B year-to-date. Modest demand improvements in some industries helped enable Q/Q growth, and they expect the segment to return to Y/Y growth next quarter. Finally, in gaming, Sony and Microsoft holiday-related orders helped demand.

Useful Life Debate:

Su was asked about the useful life of a GPU debate. Encouragingly, she’s seeing signs that customers may be willing to stretch these estimates beyond 5-6 years. That’s great news for AI infrastructure demand, and would be fantastic news for hyperscaler depreciation growth if they do end up making the move. There is simply too much demand for compute right now to think about retiring previous GPU models. Customers are finding ways to make these aging assets productive and that’s awesome to hear.

Fine quarter. I think some were disappointed by how much of the data center and overall outperformance came on the CPU side. Still, there are reasons to be excited. The massive OpenAI deal, strong relationships with Meta, Oracle and Microsoft as well as Helios developments are all notable. Those signs point to accelerated GPU growth coming next year. While AMD may not be able to operate at a 60% EBIT margin like Nvidia can, they still clearly have good enough GPUs to attract large deals with some of the most important customers out there. And for as long as this AI supercycle lasts, there is plenty of GPU demand for AMD to do just fine. Beyond that, the CPU demand acceleration stemming from AI is also encouraging, and AMD is the best-positioned company to take advantage of that. If I’m going with a GPU vendor, I am going with Nvidia, but there is nothing wrong with this company in the slightest. Great team. Fantastic job passing Intel on the CPU side. And now? Signs of being a highly formidable player on the GPU side.

Starbucks announced a new joint venture with a company called Boyu Capital. Starbucks will sell 60% of its China business and retain 40%. The deal is valued at $13B. That's higher than previous rumors of $10B, but not surprising after Niccol told us the business is worth "more than $10B" recently. This should be good for sentiment and should mean an American brand gets a large potential target taken off of their backs. If trade drama escalates again and China chooses to target U.S. companies, they'd now be directly hurting their own institutions if they single out Starbucks. I love this. Rely on your chosen partner to run and grow this. Let your brand get you paid. Shed a lot of the headache.

AWS signed a large 7-year $38B cloud deal with OpenAI. That must have been what Jassy meant by “unannounced October deals” on the recent call earnings. The arrangement will start with OpenAI tapping into existing Nvidia GPU-based AWS capacity and will begin immediately. OpenAI usage will ramp over the coming years & AWS will eventually build new data center capacity for this new customer. This helps OpenAI diversify away from Azure, while giving AWS gets a big chunk of business. The deal should start contributing to AWS revenue this quarter, with all of the capacity set to be in place by the end of next year. There are also options to expand the deal beyond 2027. The deal should be worth more than $5B in AWS revenue per year or about 3.7% of annualized revenue.

Zscaler is buying a company called SPLX. They're looking to upgrade AI asset security with this move. Here's what Zscaler gets from the purchase:

Generally speaking, this makes ZS a much better partner for AI use cases beyond its core network and cloud environments. It pushes them firmly into AI model and app security, making them an end-to-end vendor in a world obsessed with point solution displacement. In addition to accelerate AI adoption through better security and safety, this launch bolsters the platform-level cross-selling, retention and lifetime value edges it already enjoys over the field. Terms were not disclosed and the company has raised just $9M since inception. The price tag was probably very small and this likely won't be material to near-term results. I'm a fan of this.

“AI is creating enormous value, but its full potential can only be realized when it can be secured. By combining SPLX’s technology with the intelligence of the Zscaler Zero Trust Exchange and its native data protection... Zscaler will secure the entire AI lifecycle on one platform." -- Co-Founder/CEO Jay Chaudhry

Output Data:

Inflation:

Consumer & Employment Data: