Greetings, everyone, and welcome to the Calix third quarter 2026 earnings conference call. At this time, all participants are in a listen-only mode. A question-and-answer session will follow the brief prepared remarks. If anyone should require operator assistance during the conference, please press star 0 on your telephone keypad. As a reminder, this conference is being recorded. It is now my pleasure to introduce Nancy Fazioli, Vice President of Investor Relations. Nancy, please go ahead.
Thank you, Sam, and good morning, everyone. Thank you for joining our third quarter 2026 earnings call. Today on the call, we have our President and CEO, Michael Weening, and Chief Financial Officer, Cory Sindelar. As a reminder, yesterday after the market closed, Calix issued a news release, which was furnished on a Form 8-K along with our stockholder letter. It was also posted in the investor relations section of the Calix website. Today's conference call will be available for webcast replay in the investor relations section of our website. Before I turn the call over to Michael for his opening remarks, I want to remind everyone that on this call, we will refer to forward-looking statements, including all statements the company will make about its future financial and operating performance, growth strategy, and market outlook, and that actual results may differ materially from those contemplated by these forward-looking statements.
Factors that could cause actual results and trends to differ materially are set forth in the third quarter 2026 letter to stockholders and in the annual and quarterly reports filed with the SEC. Calix assumes no obligation to update any forward-looking statements, which speak only as of their respective dates. Also in this conference call, we will discuss both GAAP and non-GAAP financial measures. A reconciliation of GAAP to non-GAAP measures is included in the third quarter 2026 letter to stockholders. Unless otherwise stated, all financial information referenced in this call will be non-GAAP. With that, Michael, please go ahead.
Thanks, Nancy. We have just returned from Calix ConneXions, where thousands of customers and partners joined us to discuss the path forward and the reality that there is no version of the future where AI is optional, a future we intend to lead for our customers and the markets we serve. With Calix One deployed since March, we deliver the trust and security that our customers require to be successful within the new AI reality, allowing us all to move from experimentation to execution safely. In these uncertain times, we believe that predictability is highly valued by our customers and their subscribers, our team members, partners, and investors.
For our customers, we provide predictability of cost through our leadership with open-weight models, the platform, and the evolution to specialized models that will land on all devices over time, and return on investment as we operate billions of workflows annually, giving us unmatched insight into which agents and workflows will yield the greatest returns for our customers. Predictability for customers extends into predictability for investors, as evidenced by the records that we broke in third quarter and the end-of-year forecast that will see revenue at the midpoint hitting 19% year-on-year growth while OpEx leverage begins. At ConneXions two weeks ago, the new CEO of ALLO clearly stated the value of our platform pre-AI. He had come from a multi-vendor broadband provider to ALLO, who has partnered exclusively with Calix.
The Calix platform at ALLO yields 40% lower truck rolls, 60-point higher net promoter score, and is supported by 25 IT people versus 400. That is the power of our platform before AI. With Calix Agent Workforce, which is rolling out at a rapid pace, those results will expand as we automate functions and workflows, yielding new subscribers, revenue growth, reduced churn, and even higher profits for broadband providers, while our platform expands to manage service providers for apartment buildings and multi-dwelling units, and then on to hospitality. Last week, we released the next step in our AI strategy, the Calix Communications Language Model, or CLM, and Calix Gems, which are designed to bring communication-specific intelligence to the right place across the platform, from the cloud to the network to the subscriber premise.
This architecture will enable us to deliver increasingly specialized capabilities for platform customers while enabling large service providers to partner with Calix by consuming our AI component capabilities into their unique environments. That combination of a differentiated platform, expanding segments with the ongoing expansion of MDU, agentic workflows, AI agents, and the addition of specialized language models, coupled with a customer success organization that is leveraging AI to enable customers to adopt at scale, gives us great confidence in the opportunity ahead and our ability to lead. Our third quarter results demonstrate that customers are responding. This is the payoff from the investments we have made and the partnerships we have built with our customers. We went slow to build the platform correctly. Now we move fast with our customers to compete, win, and expand. Cory, over to you to walk through the details of a strong quarter.
Thank you, Michael. We delivered record revenue of $302 million, up 3% sequentially and up 14% year- over- year. A key highlight was record software and service revenue of $52 million, up 3% sequentially and 19% year-over-year, as our customers continue to win new subscribers and take market share. Calix One contracts signed in the third quarter more than doubled from the prior quarter, with RPOs reaching a record $396 million, up 2% sequentially and 12% year-over-year. One important note, in connection with our Customer Supply Assurance Program introduced early in the third quarter, we began receiving non-cancelable appliance orders, which are added to RPOs. To avoid distortion of this key metric, we elected a practical expedient to eliminate the value of RPOs for contracts with a duration of one year or less, thereby eliminating hardware from RPOs.
As part of that elimination, the value of software and service contracts with a duration of one year or less, totaling $6 million, was also excluded. These software and service contracts were included in prior periods. Without giving effect to this change, RPOs attributed to software and services would have grown sequentially by 4% and year-over-year by 13%. Non-GAAP gross margin was 53.5%. Non-GAAP software and service gross margin reached a record 67.2%, increasing 340 basis points sequentially, driven by subscriber growth and continued optimization of our 3rd-gen platform. As our customers continue to focus on subscriber experience and take share, our software and services revenue will continue to increase, expanding our software and service gross margins. Non-GAAP appliance gross margin was 50.7%, which represents a sequential decline of 220 basis points, primarily due to higher memory component costs.
We also continue to demonstrate operating discipline and early productivity gains from our human-centric AI investments. Third quarter non-GAAP operating expenses were $124 million, representing 41% of revenue and down from 42% of revenue last quarter. Our balance sheet remains strong. We ended the quarter with $155 million in cash and investments after deploying $50 million to repurchase 1.4 million shares. DSO was 38 days, and inventory turns were 2.8, reflecting improved collections and deliberate inventory investments to support customer demand and product availability. After quarter end, we increased our stock repurchase authorization by $100 million, adding to the $44 million that remained at the end of the quarter. Turning to guidance. For the fourth quarter of 2026, we expect revenue to be between $311 million and $317 million. This outlook reflects the ongoing broad-based deployment data and order visibility across our customer base.
Our fourth quarter non-GAAP gross margin guidance is 52.25%-54.75%. The range reflects component cost expectations as well as customer and product mix. As noted last quarter, we chose to include surcharges on new orders. As such, we expect to recover a greater percentage of higher memory costs in the fourth quarter as we work our way to gross profit neutrality. Our fourth quarter non-GAAP operating expense guidance is $129.5 million- $131.5 million. The range reflects fourth quarter investments in our annual ConneXions Customer and Innovation Conference and five additional days in the quarter, partially offset by continued productivity gains. Setting aside the fourth quarter bump we see every year due to ConneXions, we expect operating leverage will continue as OpEx as a percentage of revenue continues to drift downward to 40% and below, as we continue to deploy human-centric AI internally.
Our guidance for non-GAAP EPS is $0.41- $0.49. At the midpoint of guidance for the fourth quarter, our 2026 non-GAAP EPS is $1.76, which would represent an annual EPS growth of 30%, compared to revenue growth of 19%. Michael, back to you.
Thanks, Cory. We have built the industry's leading AI-native platform. Now we go fast with our customers, amplifying their teams and ours to compete, win, and expand at a rapid pace. Nancy, let's open the call for questions.
Thank you, operator. We're ready for questions.
Thank you. We will now be conducting a question-and-answer session. If you would like to ask a question, please press star one on your telephone keypad. A confirmation tone will indicate your line is in the question queue. You may press star two if you would like to remove your question from the queue. For participants using speaker equipment, it may be necessary to pick up your handset before pressing the star keys. One moment, please, while we poll for questions. Our first question is from Ryan Koontz with Needham & Co. Please proceed with your question.
Great. Thanks. Good morning. I wanted to ask about RPOs and specifically software bookings there. Are you pleased with where you're at there in your recovery from the cloud transition? Any trends within the mix of RPOs you'd point out? What sort of year-over-year booking rate you think you can get to next year on software RPOs?
Yeah, I'll talk to the trends, and then Cory can talk about one specific nuance in RPOs. From whether we're pleased, yeah, I'm wildly pleased, because what that showed was customers in Q3 actually started to commit to our AI, where we're going from an AI point of view, even though our first workflows just started to roll out in August, which is the beginning. As they start to see what is possible, a huge percentage of them just jumped in, as you saw with the doubling of the contracts, because they agree with where we're going. Frankly, what it validated is that we're going to be the easy button for AI.
Everybody in every market needs to figure out what their AI strategy is, and if you're the one who can provide that in a trusted and secure way, but also give them predictability of cost, which as we know with tokens and all those things is an issue, we've resolved that. More importantly, give them a roadmap that shows where they can get predictability of return on investment, which is very hard to acquire, and we can actually give that to them really clearly, saying, "Here's the stack ranking of the workflows.
We're implementing them in this order, and based upon everything and how you run your broadband business, this is the ROI on it." Those were the early adopters, but I believe in the previous call, I had a customer who, and I'll reiterate this, who basically stated, they've never bought into our longer-term vision with regards to what we're doing on the smart side of stuff. No matter how much I tried to sell them on SmartHome and SmartBiz and where we're going, he never bought in. They signed up in third quarter, and the reason why is for five years, and the reason why is because they know they need artificial intelligence. It was time for them to get on the experience bandwagon, and doing AI and experience at the same time was highly valuable.
Having just come out of ConneXions, where we had almost 3,000 people there, the momentum was incredible. If you haven't had the opportunity to watch that event, that's the best event we've ever done. I could stand up there with confidence in front of all those customers and say, "We have nailed it. We absolutely 100%, if you sign up with us for the future with AI, we will give you everything you dreamed." I can commit to that because, as I said on stage, our Chief Product Officer, who, congrats to him, he just won CPO of the year here in the Valley, which is a pretty huge accomplishment.
And the reason why he won that is because he could stand up in front of our board in the last board meeting and say, "We will deliver more innovation on our platform in 2027 than we have in all of 2026 and in all of 2025 combined." More importantly, he will do that at flat to down from a budget point of view, showing a massive jump in productivity. Do I feel good about this? Hell yeah. For sure. Cory, why don't you talk about the numeric nuance?
Yeah, Ryan. As I said in my prepared remarks, we wanted to avoid a distortion created by including appliance gross margins or appliance RPOs in that metric and keep it a pure software and services number. With that, we elected a practical expedient to eliminate contracts of a year or less.
The 6-
You need to add back 6 million to have a comparable number to prior year.
What would RPOs and the percentage have been then?
4% quarter-on-quarter and 13% year-on-year.
Which is pretty good.
Great. That's great, guys. Michael, I totally agree with you on the certainty around costs and the ease of adoption. It's very, very compelling. It's exciting.
Yeah. Ryan, we're the easy button, right? All you have to do is sign up, and then what happens is these workflows and agents just start popping up. At ConneXions, I demonstrated when I was on stage just how trusted and secure this can be, because I gave the example of there was one of the employees of our customers did something funny. He tried to break one of our agents, and this was like you would use Microsoft Copilot or something like that. He basically pounded on our agent inside Service Cloud and tried to get it to answer how many calories are in a McDouble, which I didn't know what a McDouble was, because I'm an Egg McMuffin guy, but it's basically a two-patty cheeseburger, and it has 400 calories in it. Lord knows a lot of you have.
It took him 51 tries before he could break it. So, we also leave them with the certainty that from a data privacy and security point of view, that we're going to be able to protect them, because you think about the data that they have, that's really important, especially in the Look what happened from a cyber hacking this week, where $6 billion got stole off of a crypto exchange, right? So, the attacks are increasing from, and pretty much guaranteed that came from Korea. So, these crypto attacks from these nasty nations continue. So, we give them that security, we make them really secure, and then we can demonstrate, "Hey, by the way, if you implement this, we're going to help you get 30% or 40% productivity.
That's great, guys. Maybe just a quick follow-up if I could. Could you maybe characterize where you are in terms of some of your tier 1 engagements, what kind of conversations you're having there, what kind of the nature of the business opportunities looks like, and maybe give a little color on say the larger?
Yeah, so we have two large tier 1s in North America that we deal with on a regular basis, as you're well aware of, and those continue to progress at a very rapid rate. We're doing really well. On the international stage, we have started those engagements in 2026. You know these are 18- 24 or 36-month life cycles, and we have many engagements going on internationally. But those are 2027 through to 2028, where 2028, what happens is you go and win some business in 2027, demonstrate it, build footholds, and then you start to really drive revenue through the latter part of the next year and into 2028. So those are going well. I would say that if you look at some of the press releases that we put out at ConneXions, there were two specific ones that are relevant to large customers.
The first would be our announcement that the SmartLife products are now on prplOS. If you're not familiar with what prplOS is, prplOS is something that a bunch of tier 1s have come together and built a standard around operating systems on-prem. Our SmartLife products are now available so that they can actually run those in a prplOS container. That's just the first step for us. We will actually take our entire operating system, everything that we've done in EXOS on the prem, and we will make it prplOS certified in the early part of 2027 so that anyone who wants to adopt our operating system can, again, prplOS certified. The second part is with what we did with language models, and I cannot overstate how important this is.
The release of the Calix Communications Language Model was where we demonstrated that we're not just a data and experience company, we're an AI company. Let's be really clear. We released a language model for the communications sector that is ahead of everybody else. No one else has done what we did. What it is we basically implemented specialized training so that language model would understand what communication service provider needs. Again, if you go to calix.com, you can see that in my keynote, I actually talk about the release of the language model. What differentiates that is that you can take a large language model and I can have it paint Cory on a pink unicorn flying over a mountain, which was exactly my reference, although I used John Durocher on stage. All of that is useless when I'm running a broadband company.
What I need is, I need to understand how to network and provision. I need to understand how to have great Wi-Fi, what is the experience in an apartment building, all those kind of things. So we stripped away all that superfluous components, got the model down to a significantly smaller state, and then we trained at a very low cost. In fact, the way that we trained it was really the innovation that we did with AI. Out of that will come, so that's one of our CLMs. We will have multiples of these open-weight models that we have trained specifically for the communications market.
The next big step for us is, that's where I announced the release of Calix Gems, which is what'll happen is that we will then take those and slim them down further into very bespoke and specialized capabilities. So for example, we will have a CLM that is really great at just doing security, and it will be small enough because we're going to get it down to a couple hundred million parameters where we can run it for cybersecurity on a premises system, or we can run it for Wi-Fi management on a premises system, or we can get a bigger version of that and run it in the network on an E9-2.
The evolution of artificial intelligence is frankly what we are at the forefront on for the communication sector, because we are taking these open-weight models and transforming them into specialist models that our customers will leverage. Our existing customer base will not need to understand any of that other than to know that we are doing it, because it is just going to show up. It is going to show up on an E9-2. It is going to show up on a GigaSpire. It is going to show up all across the new switches that we have just released.
That will just show up on the appliances running as part of the operating system in the cloud, and they are just going to, back to why we are the easy button for AI, because we are so far ahead of what is going on in the AI world that if you did not see the CNBC interview that we did, look at how that interviewer is leaning forward because we are way ahead and we are going to keep our customers way ahead. Then to your question on the tier 1s, that will be consumable by a tier 1 where I have this massive bespoke custom Gordian knot of an infrastructure, and I will be able to say to Calix, "You know what? I am running prplOS on a premises box, and because I am so slow, I can only have one, so I am running prplOS there.
But what I am going to do is I can run SmartLife on it. Oh, and by the way, I can also grab a Cyber Gem from us that will constantly be updated and will be the best way to protect their customers from a cyber point of view in small or extra small language model that is specialized for Wi-Fi capabilities or cyber, and they will be able to consume that from us." Our differentiation in these large tier 1s is that we are going to make, for them, we will make artificial intelligence an easy button. That is elements if you did not parse that out from ConneXions, because ConneXions was at our regional customers who range up to about a half million subscribers.
Right
Embedded in there were some very clear messages for tier 1s that will be the messages we bring to Mobile World Congress and all those, and Network X, and those places as we continue to roll those out at an aggressive pace.
That's really impressive, Michael. Thanks for that rundown. Appreciate it. Thanks for the question.
Yeah. Thanks, Ryan.
Thanks, operator. Next question.
Our next question is from George Notter with Wolfe Research. Please proceed with your question.
Hi, guys. Thanks very much. I was just looking at the appliance growth in year-on-year terms. I think 12.5% is the number I am looking at. I am curious what that number looks like in unit terms. Obviously, there are price increases, I think, embedded in that improvement in the appliance side. I am just curious what that looks like in units. Thanks.
Yeah, George, we do not break that out. For the GigaSpire, our deployment rates of GigaSpire continue to go up. So, we see an increase in appliance sales, but to your point is that customers have been burning down inventory, frankly, because of the fact that they have been trying to work through that. But GigaSpire deployments continue to increase in our customer base, which means that the core demand affiliated with it, George, is still there. So that is why we look forward into 2027, and we still see very strong demand because GigaSpire growth keeps going up.
Got it. Okay, super. That is great. And then
Good question.
just trying to understand some of the changes you made here in the current RPO metric. Could you explain what the Customer Supply Assurance Program was all about and why we are including appliance revenue in that number, and then I guess subsequently backing it out? I am just curious on the puts and takes of the current RPOs. Thanks.
I will explain the program, you explain the financials, Cory.
Go ahead.
Okay. The program from an assurance point of view is that they give us a non-cancelable PO, and then we guarantee the price for a period of time. The benefit to the customer is we are basically giving them some price assurance, as it has been a pretty volatile time from an appliance with memories, right? What we get is we get assurance on their order flow. That is what we put in place, and then every single month, George, we, unlike in COVID, where we only did one price increase in COVID, this is a very different time. So every month, we change the surcharge on memory, and we keep it up to date. Therefore, if they do not make that commitment, then they risk as if memory goes up again-
The next order
the next order will be at a higher price. That was kind of the incentive for them on why we did that, and it was well received. Obviously, no one likes price increases, but simple reality is everything in the world has gone up in price, whether it's gas or memory. Cory, why don't you walk through, George, on the RPO metric and what we did there to make sure it was not polluted?
Yeah. George, to ensure that we don't have an RPO number that includes both hardware and software, and we're distorting that growth rate on software, we decided to back out contracts of a year or less. All the appliances will be shipped within less than a year. The certainty is for a six-month window. The $6 million applies both to the RPO number that you would add back. You would also add it back to the current RPO number. It goes into both because it's all short term, one year or less.
Super. Thanks very much. I appreciate it.
Thanks, operator. Next question.
Our next question is from Chris Schwab with Craig-Hallum Capital Group. Please proceed with your question.
Hey, guys. It's Ben Taxdahl for Christian Schwab here. Just a quick question. We took the top line. We're at 19% or above the previous 15% to 20%. I guess just any commentary on how we should think about 2027. I know you guys probably can't guide 2027, but just any commentary on, as it compares to 2026 would be very helpful.
Yeah. So I would say the biggest thing you should look at is the fact that we've deployed the platform that was in March. We have all of our R&D organization on full throttle to build out sellable things to drive our sales organization so our customers can be more successful and win more. As we go through 2027, I stand here telling you that we will deploy more capabilities into the market than we have in the last two years. So, the number I gave on stage at ConneXions was 1,400 capabilities versus what we did in 2026 and 2025, which comes out to about 1,300. So the value there is twofold.
First, as evidenced by the momentum that we saw in contract closings in Q3, which was really the early jumpers, even before we got to ConneXions, the release next week is a big release with a whole bunch of workflows and new agents and all these capabilities. It is kind of like pre-sold, it is there. Then we went to ConneXions, showed our customers what is going on, so I would expect that there is going to be a lot of enthusiasm out of our base for these capabilities. Because again, there is no reality where AI is not the center of everything, and anyone who does not have a kickass AI strategy is cooked. We are that easy button for 1,000 service providers.
As I evidenced in the previous commentary on where we are going with SLMs, we are going to be the easy button for tier ones too. Then, with regards to the incremental capabilities, as I continue to mention, we have launched our SmartMDU product. There is another big release in November, and then there is one for SmartMDU, what the R&D organization says it is never done, but it gets to a significant level of completeness in February where the SmartMDU consumers are like, "That is the product that I needed." Then you are going to see a massive expansion into new markets as we go after MSPs who sell to MDUs, so those are non-fiber-based providers. They are just the ones who build out custom integrations into complex buildings, and the expansion into hospitality.
You are going to start seeing us do what we have never done before, which is actually expanding beyond broadband providers into MSPs and hospitality and those areas through the year. How do I feel about the year? Cory is not going to guide to that. Pretty damn good.
Ben, let me give you. We will fall back to our long-term financial target of 15% annual growth. We feel good about that, and that is our baseline as we roll into next year and look to work to improve upon that number.
Perfect. Thanks. Just thinking about the long-term targets shared earlier in the year. Gross margins are going to be flat, consolidated gross margins guided to be flat. I guess it kind of implies to me, we're kind of expecting to maybe stabilize here, then as we go into next year, we could see that increase from these levels. Is that kind of the right way to think about it?
Well, I would personally like to see the blowup of the artificial intelligence market to happen, because those jackasses are the ones who are driving up all the memory costs across the world, while losing hundreds of billions of dollars, while those of us who actually make money suffer at it. As that actually happens, then what will happen is you'll see the memory market collapse because it's artificially high. If you look at Micron, it is making 90 points of margin or 85 points of gross profit. At some point, that's going to What? Micron, making 85% of gross profit. At some point, those guys are going to be punished aggressively for their greed, and for what they've done to the broader market. Then at that point, we'll get to a more normalized state, and there'll be other countries who also start manufacturing memory.
As that capacity comes on, you're going to see a massive glut in the memory market, and it's going to do what it does every freaking time, which is the prices are going to drop like crazy, and we're going to throw that straight to the bottom line. But when that happens, that's what we all got to wait for. It's great to see Google lose $0.5 billion a mark, or $0.5 trillion worth of market cap, so maybe they'll actually get punished with regards to the stupidity on their capital investments, which will be great. Then that all falls to the bottom line for us so that we can actually have proper margins again. No, I'm not being hyperbolic.
Hi, Ben. Let me give you a little bit of perspective.
If you are wondering, Cory is waving me down, so it is true.
From Michael's commentary, you can take that it is likely to be a very noisy environment.
Correct.
The AI data center build continues to put pressures on the supply chain, as well as we realign global supply chains to take into consideration geopolitics. Both those two things create a very noisy environment for supply chain and cost as we roll into 2027. We are looking to do our best to manage through this with our program as we partner with our customers to get to a gross profit neutral state by passing along the cost without putting profit on top of it. I cannot tell you as I sit here, the rate of change as it relates to memory prices. We are expecting it to flatten out, relative to what we have seen in 2026, which should make it more manageable. I would also say that the pressures put on the AI supply chain buildouts are pressuring other parts of the supply chain.
Memory just happens to be the single largest one of those. I think you have it right. I think we look to stabilize appliance margins at these levels and try to build on where we are now, knowing that it could go up or down as we move into 2027, and we are uncertain about that.
That's extremely helpful. That's all I got. Thanks, guys.
Thanks, Ben. Next question, operator, please.
Our next question is from Joseph Cardoso with JP Morgan. Please proceed with your question.
Hey, good morning, guys, and thanks for the question. Maybe just following up on the margin question, but focusing on the other side of the portfolio here on software and services. Obviously, saw a nice bump this quarter to the record that you guys kind of articulated previously. Maybe just curious, I think you put out in the past, talked about kind of this maybe a 70% type bogey going forward as maybe a little bit of a near-term or medium-term North Star. How are you thinking about trending there? Is that something we can see in the next couple of quarters? And maybe what are the moving pieces there that we should be considering? And then I do have a follow-up.
Yeah. I think the volatility and the noise associated with the platform transition is now well behind us.
Yeah.
We articulated that we would get back to a new record. We did. From here, it's all about software revenue growth. We are optimistic about the trajectory of RPOs and where that's headed with all the AI work that we're doing. Now it's just about adding subs and helping our customers go faster to accelerate the software revenues. Yes, 70% is on the horizon. Is it in one quarter, two quarter, three quarters? I can't say for sure, but it's inside that 2027 envelope, and it's just going to continue to build there. There's not a single day that goes by that software revenue does not go up, because remember, all of our software stack is all tied to subscriber adds. As our customers do better, we do better.
With an NPS score of north of 70, 80, 90, our customers are the share gain winners. They are taking market share because their whole focus is on subscriber experience, and when they win, we win. So, it's going to continue up. There's just no stopping it.
Got it. Then maybe switching gears just more on working capital dynamics and free cash flow. Obviously, you guys kind of talked about it already in terms of the higher price inventory kind of taking a toll here, but just curious as we're thinking about how that trend's going forward and maybe what needs to happen, anything, any other levers at your guys' disposal that can kind of help with the trends as we think about our models in the subsequent quarters?
Yeah. We made a lot of progress this year on building that supply assurance with our customers to make sure that as they are taking share, we have appliances that can back them up. I think we've added almost $75 million or more onto the balance sheet over the last 12 months, ensuring that we have the inventory necessary to meet those customer demand environments. As I look out at 2027, I don't think we're going to need to be making the same level of commitment as we did in 2026. From a working capital perspective, you would expect to see free cash flow improve year- on- year inside of that trend.
Understand, at some point, I don't know when that some point is, the investments we had in inventory will reverse themselves, and then obviously the free cash flow generation will accelerate much higher than where we've been. But in this entire environment, we are looking to make sure that we are supply certain as our priority, and it allows us then to focus on footprint expansion. That's what we'll continue to do, and knowing that free cash flow generation will accelerate at some point here in the not-too-distant future.
No, got it. Thank you, Cory. Appreciate the color.
Thanks, Joe. Operator, your next question.
Our next question is from Scott Searle with ROTH Capital Partners. Please proceed with your question.
Hey, good morning. Thanks for taking the questions. I apologize, I got on a minute or two late. I am not sure if you addressed this, but Mike, from a high-level macro perspective, certainly some momentum coming out of ConneXions, but are you seeing any slowdown in terms of decision-making or otherwise, given the current macro environment and interest rate environment? It sounds like you are still comfortable with that long-term 15% growth rate, but wondering what you are seeing on the margin. Looking to 2027 and 2028, if you could comment just on the MDU opportunity in terms of what you think that is going to represent in the mix and how that ramps up in 2027 into 2028. Lastly, operating leverage, it seems like from an internal standpoint, you are seeing an incredible amount of productivity from your own AI implementation.
You laid out some goals at the analyst day in terms of 2027, 2028, I think from an operating leverage standpoint. I am wondering how you are feeling about that and if you are feeling better in terms of some of the internal productivity gains that you are seeing. Thanks.
Well, that is three questions. All right, so first, what was the first one?
Revenue and-
Yeah, I did address that, so I feel very good about that as we went through the Because there's multiple, there's three really elements to that. The first one is that we saw the momentum going up to ConneXions because the early adopters are actually seeing that we are the AI easy button. As we presented at ConneXions, again, we had all those folks there. It was an incredible event, and we demonstrated the roadmap for AI based upon the insights that we have with regards to workflows, but more importantly, the fact that we can do this at a really rapid rate. We have a big release next week where we're going to be releasing more and more workflows in AI.
We also talked about the fact that we were releasing, or we announced the release of Workflow Studio, which is the ability for a service provider to actually take, think of our workflows, the reference workflow, then they go into Workflow Studio, and they have the ability to see the MCP connections they have to other systems. They're able to see agents, the agent-to-agent that are connected to other systems, and expand out their workflows and do customizations to improve their business. That's a critical component. At the same time, tomorrow or next week, we release our agentic-based CommandIQ application, which is a huge step forward from experience, and then also AI Control Center, which is the ability to see everything that's going on from an AI point of view.
I feel very comfortable that as we go through 2027, that our customers who have to face the reality that they need to do something on AI, and building it themselves is a fool's errand, and frankly, is going to lead to death, and if not, the hacking of their companies. Because there's so much complexity around AI, that we have the ability to give them something that's secure, that they can trust, and that will provide predictable outcomes, because we're looking at billions of workflows and building out what are the right workflows to improve a business. On the existing business, I feel very comfortable with our ability to continue to grow, and that growth comes from one thing, them winning more subscribers and growing revenue per sub while reducing churn.
The metrics are very clear that we are adding GigaSpires at an increasing rate, so that is going to continue. The other two segments, I also announced, I am not going to reiterate that because you did join late, I talked about the release of Calix's Communications Language Model, where we are going with Calix Gems, which are specialized language models that are large, medium, small, and extra small. Those become yet another thing on top of the work that we are doing, where SmartLife just got released to prplOS, which is a standard that large customers use. So, our ability to actually go and partner with tier 1s and help them press easy buttons on experience and AI continues to expand at a rapid rate.
The third part of that, which also addressed your second question on MDU, is that the SmartMDU product, there is a big release in November and then one more in February, which SmartMDU gets to a level of completeness that frankly puts us as the dominant provider or dominant capabilities to serve the MDU market for broadband providers. Then we are going to see a significant focus through 2027 as we take this massive growth in capacity that we have, as the people have moved away from building the platform into the R&D organization, who has completely evolved their processes to drive massive productivity gains in research and development, and now be focused on managed service providers who support apartment buildings and MDU facilities, and then the expansion into hospitality, which is going to come at a very rapid rate through 2027.
That is all rooted in what I said on stage at ConneXions, which is, we will deliver more innovation in 2027 than we have in all of 2026 and all of 2025 combined. That number is roughly 1,300, and we expect to do 1,400. So, it is a massive jump in productivity. My confidence with regards to our productivity on R&D side is evidenced by the fact that their CPO told a very interesting story that I shared on main stage too, which is, we had a function for a RADIUS server scoped at 15 man-months, and they did it in 3. So, that answers your last question, which is, how am I feeling from a productivity point of view on AI work internally? That shows you what the R&D organization is doing, which is about a 30%-40% jump in productivity.
Also evidenced by the fact that we expect the R&D budget in 2027 to be flat. So, having a flat R&D budget delivering 40% more product functionality is a pretty incredible testament to the power of AI. Across the broader organization, look, implementing artificial intelligence inside an organization requires a massive managerial shift. Managers to be successful in the past, there is a number of parameters on what you need to be. To be a good leader, there is a number of things that you need to do. Now wait, I am going to say this clearly. I am going to explain what a good leader is. A good leader is able to have a strategy. They understand how to implement that strategy and execute against it.
They are able to attract talent and also manage talent, which means that they remove low performers, they coach people to a higher level of performance, and that attracts people to the team. Then the next thing that they have to add to that skill is the ability to analyze workflows. This is one of the biggest skills that new leaders are going to require, which means you need to take a step back and understand, and this is what is old is new again. This goes back to, look, I am old, so we go back to the Six Sigma work, right?
It is not unlike Six Sigma work where you need to become, as a leader, if you want to be successful, you got to do the hard grunt work of actually applying a Six Sigma mindset to it, go through what are all your processes, understand what your humans on your team are doing, and then going through piece by piece, pulling all that apart and saying, "What will I agentify so I free my team from doing repetitive work that adds no value?" Then, "How is my team going to evolve to achieve outcomes like the R&D organization is going to do?" What I didn't tell you was that we are going to drop R&D down by half. I told you that R&D is going to be flat and deliver 40%, like a massive jump in productivity.
That in the end is what we are doing from a leverage point of view, which is every single quarter, and in fact, I have my ops review just before the board on Monday and Tuesday next week, and every team is getting, in essence- say that is crude. An exam of their business from me and my chief of staff, and what they are expected to do is grind through their organization top to bottom, understand how everything works, pull it apart, and agentify it. But this takes time. I think what you see in the results in Q3 is you saw some sparkles of that coming through. You hear the R&D stories, which is pretty bloody clear what the massive value is going to be there.
But then, in the other areas of the business, you saw us go from 42 to 41, and you are starting to see the leverage come through. The way I would describe this is, look, there was an article written about me in Selling Power about a month ago, and one of the things they called out is that I am a grinder. Well, you know what? The good thing is my leadership team is filled with grinders. This is what we do. I think that the Calix organization is going to be dramatically more successful than other organizations, primarily because of the fact that I am surrounded by a leadership team who all thinks the same.
We know that we have to grind our way through our organizations and understand how all the work gets done, and root out inefficiencies, replacing those with AI to allow our teams to do higher value work. For example, if I have X amount of people in customer success today, if I double the size of the company, the goal should be I keep that flat, and I do that work with agents and humans who are more effective. That is the mindset that we are achieving, and I think you are going to see that from us, from myself and Cory, every single quarter we are going to be sharing with you how something else that we grinded through and added agents to allow our teams to be more productive. Am I feeling encouraged?
The good thing is that the distraction of the platform lift that we went through for 2.5 years is now behind us. The road forward is really freaking clear, and my Chief Product Officer, Shane, and I joke about this all the time. The step that we are currently in is we are no longer trying to figure it out. This is just work. This is just grind this out. That is what we are doing. We are going to grind it out through the end of the quarter and through 2027. We are in a unique position because we do not have where other companies are still trying to figure out what the heck they are going to do in this massively changing world, it is really clear. Now we just execute and grind.
Okay, perfect. Thanks so much. Nice quarter. Keep grinding. Thanks, Mike.
Have a good day.
Thanks, Scott. The next question, please.
Our next question is from Tim Savageaux with Northland Capital Markets. Please proceed with your question.
Hey, good morning. This is kind of good timing. Question was going to be to follow up on the operating leverage comments you guys made during the call. I guess, Cory, my question to you was going, well, first going to be, Michael, did you get R&D flat to down? It looks like you did. Then the piece was-
Wait, I want to comment on that. Let me just comment on that first part is that, you know what's great is we didn't need to do that. Actually, the leadership team inside the R&D organization, that was always their goal. 2.5 years ago, I said this in the If you haven't taken 10 minutes to watch the CNBC thing that Shane and I did last week, you should. We'll repost this out, the entire interview. What was interesting back in 2023, this is where Shane started to educate us all, and his leadership team started to educate us on the value of open weights and building the platform and all those things. Then he also started to identify that we could really start to agentify how we build products.
I would say what's great is that that comment about R&D being flat in 2027 with massive productivity gains is frankly just him doing what they promised to do. They executed against that vision. We made the significant investments over what is now three years of hard freaking work to get here, and now it's going to pay. That's kind of how I would contextualize it. Sorry, I just wanted to comment on that.
Oh, great. I guess carrying that forward, I mean, Cory, the original question was going to be, over what timeframe do you expect to see OpEx drift below 40% of revenue? It looks like you'll be there for the year in calendar 2027 in this context. I just want to kind of bounce that off you in terms of expectations for operating leverage. Let's just start there, and I'll follow up.
Yeah. So, we outlined in our investor day back in April that we were looking to grow OpEx at half the rate of revenue growth in 2028. I think we've taken the goal internally to see where we can do to accelerate that into 2027. So, while we're not committing that now, that is still a goal that we have, is to accelerate that into 2027.
What we have to balance out is going into the new markets. I don't care what anyone says about AI, it's not replacing sales reps. It's just not. It's going to make sales reps more productive when they're doing less. They're going to do less of what they hate anyways. They all whine about it. I was a sales rep for 15 years. I whined about it, which is internal stuff and analysis and all those things. Selling deals or selling. I want to be real clear. Who thinks that AI is going to replace sales is probably a marketing or product person who's never sold anything in their life. Those of us who understand the sales organization know that AI will augment their capabilities, but you need sellers.
The investments that we make in 2027 are actually going to be in sales and marketing, because we're entering into new markets. For example, take the MSP builder market with what we're doing on MDUs. Our brand is well-known with broadband providers. It's not well-known in that market. We got to go make those investments to build brand. We'll balance that out with massive productivity gains and OpEx leverage, but we got to go balance out how we invest to grow brand.
Got it. I do. Can you hear me?
Yes, we can.
Okay, great. I kind of hesitate to bring this up, but we talked about memory a little bit. My own observation in the early part of the call was that you seem to have gotten a pretty good handle on this, even though it's still a fairly volatile environment. But if you look at the compression and appliance margins that we've seen throughout the year, Cory, can you estimate what this memory price increase has cost you this year? Maybe from an earnings standpoint, is it 500 bps on appliance margins, and that's how we should look at it, and is it potential to should we see some more favorable movement, what we could see on the flip side of that?
Yes, but it's a time statement. I don't know when. Right? I don't know what 2027's going to bring. I don't see it relenting in 2027. I still think there's going to be a fair amount of noise. You're also then looking into a cost environment that is spreading beyond memory. I think 2027 will be as challenging, if not more challenging, than 2026. But again, we're focused on supply, making sure that we have that, and price is secondary to that, Tim. But I can't tell you which way I think 2027 goes. You tell me kind of where memory goes, and I could probably give you a better answer.
Hopefully, NVIDIA is a trillion dollars in market cap. That'd be great. A couple of those folks, so we can actually see some of this. By the way, it'll be interesting to even What'll be interesting is the elections today, because there's a lot of questions with regards to data centers are an element of the election conversation, right? Do we just see this unmitigated, unstoppable building of data centers, which is consuming memory, or does something shift? We'll see. Do people actually need to make profit, right? Lord knows, OpenAI and Anthropic should be bankrupt. We can hope.
Thanks a lot. Appreciate it.
Bye.
Thanks, Tim.
We have reached the end of the question- and- answer session. I would like to turn the floor back over to Nancy Fazioli for closing comments.
Thank you, Sam. Calix will participate in a number of investor events during the fourth quarter. Information about these events, including dates and times and publicly available webcasts, will be posted on the Events and Presentations page of the Investor Relations section of calix.com. Once again, thank you to everyone on this call and webcast for your interest in Calix and for joining us. This concludes our conference call. Have a good day.
This concludes today's teleconference. You may disconnect your lines at this time. Thank you for your participation.