Shownotes

Making AI More Sustainable: Insights from NeuralWatt's Chad Gibson

In this episode of Asynchronous and Unreliable, host Anne Currie speaks with Chad Gibson, co-founder and CEO of NeuralWatt, about innovative approaches to reducing the energy footprint of AI infrastructure. Discover how layered technologies, transparency, and customer-driven model choices are shaping a greener, more efficient AI future.

Key topics

• NeuralWatt’s multi-layered approach to AI energy efficiency, from GPUs to inference cloud services

• The role of software optimization in reducing power consumption and improving model efficiency

• How the demand for open source and Chinese models influences energy efficiency strategies

• Challenges in serving large, long-context models like GLM 5.2 and their operational impacts

• The importance of transparency and customer demand in achieving sustainable AI practices

• The potential of autonomous agents for further optimizing energy consumption

• Industry trends in hardware, model size, and software innovation pushing AI toward scalability

• Real-world examples of data center practices that either hinder or help sustainability goals

Timestamps

00:00 - Introduction to the podcast and guest Chad Gibson from NeuralWatt

02:09 - NeuralWatt’s comprehensive tech stack for power-efficient AI operation

04:07 - The origin story: from Microsoft experience to sustainable AI solutions

06:24 - Expanding energy-efficient AI with stranded power and inference cloud

08:48 - The role of transparency and customer demand in energy-conscious AI deployment

11:10 - Challenges with varying model sizes and use cases in AI inference

12:25 - The shift towards open source Chinese models and their efficiency implications

16:02 - The rapid pace of software innovation improving AI efficiency

20:00 - Overcoming technical hurdles when evolving inference technologies

22:57 - The complexity of supporting diverse customer calling patterns and models

25:02 - Future trends in model size, hardware advancements, and innovation

29:13 - The influence of larger models and sustainability considerations over the coming years

33:54 - Practical advice for organizations aiming to align AI with energy sustainability

36:42 - The importance of transparency from suppliers and the role of customer awareness

40:16 - The exhilarating speed of AI tech evolution and the ongoing nature of problem-solving

Transcript

Anne (00:01) So hello and welcome to Asynchronous and Unreliable, a weekly podcast where we discuss the latest ideas and concepts in tech. I'm your host, Anne Currie co author of Building Green Software, the Cloud Native Attitude, and author of the science fiction Panopticon series.

Today I'm going to be talking to Chad Gibson, the co-founder and CEO of NeuralWatt, a technology company that builds the software to make AI infrastructure radically more power efficient, and who eat their own dog food by offering AI hosting via the Neuralwatt Cloud. So, hello and welcome to the podcast, Chad. Thank you so much for being on.

Chad Gibson (00:38) Thank you for having me. It's really exciting to be here.

Anne (00:42) I was just saying to Chad that I'm really keen on you being on this because people have been asking me a lot recently about hosting AI in a way that is more sustainable. And I commonly refer them to NeuralWatt. So why don't we start by you talking me through what NeuralWatt does, and how, 'cause you actually take lots of different approaches to cutting the energy use of AI, don't you? And it'd be quite good to talk through all of them.

Chad Gibson (01:16) Basically NeuralWatt provides different levels of technology to help AI operate effectively with less power. One of the general themes we believe is that a lot of AI's growth and optimism can be handled by current resources and current energy. So there's several layers of our software, if you will. The lowest layer sits right above the data center GPUs, which are probably the thing that's consuming the most energy and growing the most radically in terms of power consumption. And we have some algorithms and efficiency models that basically allow us to keep these GPUs most efficient. We optimize for like tokens per joule or tokens per watt, so we can basically modulate the GPUs. That's like the lowest level and probably the deep tech of what we do. And then on top of that, there's different ways that we can expand upon that for different cases.

We have this product that allows us to deploy GPUs in stranded and underutilized or flexible power locations, that again is part of the theme of allowing more AI to be done within current resources. And then on the top of it, we do offer, as you mentioned, an inference cloud for folks who want energy efficient inference. And the really cool thing about that is we provide full transparency, like you know exactly how much energy per request, we provide carbon observability, and some unique things—we actually offer inference by the kilowatt hour. So all those things compound in a way that allows us to offer more and more inference or token output with less energy.

Anne (02:53) It is interesting, you don't just have one story. That's a lot of different stories. Where did you start, presuming you didn't start with all of those ways all at once?

Chad Gibson (03:01) Our story, we offer all those things based upon our experience. The short story—I don't want to get super long-winded about the origin story—but Scott and I have known each other for a long time. We both were at Microsoft for a long time, and the last team at Microsoft we worked on was a product called Zune, ancient Microsoft's iPod competitor. We were both on the Zune team. We had actually met before that, but our paths at Microsoft took different routes basically. But Scott at Microsoft had invented a lot of the techniques for carbon observability, so answering the questions of how much carbon is being emitted by this network of Windows machines. That led to optimization, that led him to Intel, where he did a lot of data center work to basically minimize emissions based on data center build outs and footprints and hardware. And so through that work, Scott had built some pretty deep domain experience on how to basically modulate chips for an optimization target. So that's where we started. NeuralWatt really began at the end of 2024, and that lowest piece of technology is where we started. We built this energy efficiency model that through observation allows us to understand the most efficient way to run a GPU based on the work that's being asked of it.

We built that system in the model and we trained that model based on a lot of simulated work, a lot of simulated inference workloads, and started talking to customers and doing more trials. That core tech, 2025 was all about that. NeuralWatt in 2025 was just that piece of software, and we did lots of trials. The trials grew in size and energy impact, and we did a trial over the winter with a utility on the East Coast, where the utility wanted to show and prove that a data center could flex its power to help unlock more flexible power to different facilities. There's a big theme about time to power for data center build outs. We did that work and when we completed that trial, we actually had built an inference stack in service of that. When we did the inference stack, we realized, we discovered so much that the energy efficiency opportunities actually expand as you go up the stack. That also unlocked all these opportunities with stranded, flexible power. So we launched that product in March. That's NeuralWatt Cloud. Through NeuralWatt Cloud, we entered into some partnerships with some other customers who want to utilize the power in their facility that is stranded today. All these things build upon each other. NeuralWatt Cloud uses all the technology in our portfolio, if you will, but we'd love to talk to customers at all aspects of this to really help AI do more with less, basically.

Anne (06:03) That's very interesting. It's unusual that you've retained all of the stack there. Quite often people will take their own products and then run with it, like you have with NeuralWatt Cloud.

Chad Gibson (06:19) Yeah.

Anne (06:20) But you also offer all of the stuff all the way up for people who essentially will be your competitors.

Chad Gibson (06:26) We thought a lot about that because when we did NeuralWatt Cloud, one alternative path for us to take was we're just going to be energy efficient inference. It simplifies—we have one class of customer, which is people and inference, and all the other things become part of that. But our mission really is to provide energy efficiency to AI, and so that would kind of preclude us from offering the deeper levels of technology to really help us achieve this goal. It may be a more complicated path because we have different customers at different levels, but it helps us achieve more and accomplish more in service of the mission. We're AI optimists, and we absolutely believe that we can do a lot to address the fear about energy and power with AI, and it involves employing all these different tactics.

Anne (07:21) I am also an AI optimist, and I find it unbelievably frustrating that the data centers industry and the AI data center industry is not a really key part of the energy transition because it could be helping. The interesting thing is why it so often isn't. So I'm quite interested in asking you the question: Why is it so often not, when obviously it can be, because your data centers are?

Chad Gibson (07:53) Well, it absolutely can be. In most of the last few data center conferences I've been to, there's been a huge desire to tell that story because there's a lot of pretty awesome things going on with data centers and helping the energy transition, having data centers be actual assets for the grid, not things that are going to hurt the grid. It's a complicated story to tell, and there's a lot of fear that needs to be overcome. When the fear is prevalent, it's kind of hard to tell the complicated reason for why that fear isn't always justified. But there are pretty awesome things happening.

A good example is here in Seattle. I live in Seattle, and there's a moratorium done on data centers in the city because there's just this belief that data centers are going to make everyone's power bills go up. Here in Washington, our grid is mostly renewable energy with hydro. Seattle actually has a pretty good grid. Where there's a good grid, accessible renewable energy seems like a great place to have data centers, but the fear needs to be overcome. The fear requires telling the broader story. There's a lot of investments on data centers to actually subsidize power infrastructure, transmission lines. There's a lot of those stories happening, but I think some work needs to be done to tell the stories.

To your point, we also believe that a lot of the data centers are not fully utilized. The way that energy is provisioned, the way capacity is allocated—it's of no fault to the data centers. It's just people traditionally think that if I'm going to deploy a bunch of servers, I need to have this fixed amount of energy. We provide solutions that allow more servers per that capacity, and more of that current capacity to be utilized. I don't think it's going to take away the need to build more data centers, but it's going to allow us to do more with what we have, like a lot of that stranded capacity, if you will.

Anne (10:04) Your comment about there being a lot of hydro in Seattle made me cast my mind back. I lived in Seattle for two months once when I was working on the Exchange project. It rained every day, apart from one day when it snowed. But yes, plenty of rain in Seattle.

The NeuralWatt story is giving me quite a lot of difficulty in knowing what to talk about because you've got so many layers. So I will focus on just one layer, which is very much at the top of your stack—which is not what we've been talking about up till now—which is the question that comes up over and over again because I'm mostly talking to enterprise users. Most of the models that you run are Chinese models. I think they're all open source models, aren't they, in your hosted stuff in NeuralWatt Cloud? Tell us a little bit about why you made that decision to go all open source, mostly Chinese.

Chad Gibson (11:09) When we launched NeuralWatt Cloud, we had a diversity of models. We had some Mistral models, we had the GPT OSS models from OpenAI, and we actually just recently started hosting the Gemma model that Google creates. The shift towards the Chinese models has really been customer demand. The open weight models out of China, in terms of closing the gap with closed weight frontier models, that gap is closing fast. When we launched NeuralWatt Cloud, I think we launched with eight models. For us, model architecture is actually a really fascinating story about AI energy efficiency, both with the emergence of MOE models, which is a way of having the weights be only partially activated at a moment—which is great for efficiency—versus dense models where the whole network is being used, as well as small, medium, and large models. Small models for many tasks are amazing and hyper-efficient.

So we launched with an array of models across small, medium, and large. We launched with some models from the US, like Mistral's European model, and the demand kind of shifted all towards the frontier large weight models from China. Part of our goal and the reason we have these different layers of stack—we would love Anthropic to license our technology to help make Claude and Opus more efficient, we would love to open it up. So we would love to do more with closed weight models and frontier models, and we would love to license different parts of our stack. The fact that NeuralWatt Cloud uses open weight models is largely because that's what we have access to. But it allows us to basically improve our energy efficiency model at the core of our tech and provide energy efficient inference. The more inference work we observe, the better it makes our energy efficient model at the base of our system.

That serves the purpose of helping train our model and make it more energy efficient. Open weight models—part of NeuralWatt Cloud has grown very quickly since March, largely because that gap between open weight and closed weight models is absolutely closing, largely because of a lot of the work happening in China right now to expand the frontier. I use ChatGPT and Anthropic models a ton, and I use the models we host a ton, and seeing that gap close has been pretty fascinating to observe.

Anne (13:45) That's interesting. As an observer of that—probably a fairly close observer compared to most people, I would imagine, since you've got loads of data there—as the gap closes in terms of functionality between the frontier models and the open weight models, what's happening to the efficiency? Because in the past, the open weight models, especially the Chinese models, had become extraordinarily efficient compared to the frontier models.

Chad Gibson (14:19) The thing about the closed weight models is we don't entirely know. We know that folks like Anthropic and OpenAI, and even their partners like Amazon and Microsoft who host their models, have huge compute and energy footprints, but in terms of joules per token, that is data that only they have access to. I can theorize, but I don't really know. It very well could be that there's some state-of-the-art energy efficiency happening with some of those models that we don't know.

With the open weight models, a lot of it—I think part of it is necessity of invention. There are limited accelerators and GPUs, and so it pushes a lot of innovation in the software stack in terms of serving efficiency. If you take a look at some of the recent innovations, specifically like DeepSeek-V4 and the recent GLM-5.2 launch going to a million context, there is so much innovation on model architectures, on attention mechanisms, on KV cache formats. Each one of these innovations—the thing that we've really learned and accelerated in the last six months—is the real opportunity of efficiency in the software layer. A lot of folks view that AI efficiency is really a hardware thing, like you need new accelerators which makes it more efficient, but there is profound opportunity in the full stack of software—all the way from the layer we exist at (right above the GPUs), to the serving layer, to the model layer, to the GPU kernel layer.

There are profound efficiency opportunities, and that whole ecosystem of software efficiency is moving so rapidly. Even with its rapid speed, there's still so much efficiency available. That's the most fascinating thing about what we observe and see. These model releases are happening every four to six weeks, so with each of them we can really assess how much efficiency this model architecture or serving pattern has assisted. The one difference between open weight and some of the closed models is on open weight, you can observe every step of the transition and build more optimizations around that as well.

Anne (16:49) It really does help to be able to look inside. It's really interesting that you're doing all of this stuff. You seem to be strangely far ahead of everybody else.

Chad Gibson (17:03) That's awesome. We've been focusing on this problem for a while, and our thesis is that energy and power limits will be a constraining factor. So we have been relentlessly focused on inference, which in early 2025 was a little bit questionable because there was a lot of talk about training and training clusters. Training generally requires a lot of co-location of GPU compute—needs to be in bigger facilities with all of the compute together—and so that took a lot of the early attention in terms of data center build-outs. That's why you needed a lot of these gigacampuses to have super large training clusters. Inference still benefits from that, but can largely be a more fragmented approach. Since our beginning, we've been laser-focused on inference and laser-focused on making the accelerators efficient. That focus was early on our journey questionable, but now it's paid off because we can go really deep on the efficiency of inference.

Anne (18:18) What real, sticky, difficult things have you hit while doing this? What were you not expecting to be as difficult as it's turned out to be, if you can tell me?

Chad Gibson (18:31) I don't know if there's one. There's so many hurdles.

Anne (18:35) Yeah.

Chad Gibson (18:38) Model evolution speed. The challenges haven't been big blockers; they are temporal blockers moving so quickly. First, inference demand is kind of insatiable right now, specifically as open weight models are getting stronger, better, and cost-effective. The demand is growing, so we had this wild experience from mid-April to the end of May where we had just started with NeuralWatt Cloud. We had our first wave of growth. A common pattern when you deploy inference is you just throw more capacity at it. An easy way to handle more growth is throw more capacity. In that era, we weren't entirely sure of NeuralWatt Cloud, so we weren't confident in deploying more capacity. We got into this pattern of just manifesting headroom via software efficiency.

We would do some work and evaluate: okay, we could generate 30% efficiency in this layer, and we'd do that and get 30% headroom. That headroom would be swallowed. It was like Jevons' paradox every three days: you free up some headroom, you make things faster, it's consumed immediately. For four weeks, it was like Groundhog Day. We're like, "Okay, we thought that headroom would last us 10 days; it lasted 24 hours. What do we do now? Okay, let's do it again." All while in the midst of a new model change.

The other big challenge was when GLM-5.2 came out. A lot of the open weight models at that moment were all 200,000 to 256k context, so they had a max context limit which puts some bounds on serving efficiency and makes it easier to serve in a way. Then when GLM-5.2 came out along with DeepSeek-V4, they expanded to million context length, like a lot of the Anthropic and OpenAI models. That was a really difficult serving challenge because part of the nature of our cloud is we have capacity fragmented all over the place, and managing longer contexts makes the fragmentation challenge harder to overcome. That was a really interesting technical challenge for us to overcome that took us a good 10 days. Those are two noteworthy things that come to mind.

Anne (21:05) That's interesting because something we say a lot in building green software is that it's really hard to run data centers. Especially if you use cloud services rather than lift and shift, you get the benefit of specialist folk looking after really quite complicated systems that need to be tuned and managed. It sounds like models are just like that, only on steroids—on a completely different level.

Chad Gibson (21:37) It's as much the calling patterns of your customers as the model. It's that combination. In our cloud, our customers are mostly software engineers. Roughly three-quarters of them are software engineers, and a quarter of them are building autonomous agents to do things. The autonomous agents themselves have their own calling patterns, and so it's the combination of those classes of customers with the models that makes things very challenging. Across our customer base, some calling patterns are way more complicated than others with certain models. Being able to identify it, route it, address it properly, and understand how to handle these classes of calling patterns combined with the different models becomes very challenging.

Anne (22:42) You support a lot of models, and on your website it looks like you're going to support even more in the future.

Chad Gibson (22:50) Our view on models is to go where there's demand and interest. The one bias we have is we would love usage across a broad array of model sizes. We've been doing better on the small end, like the Gemma model at 31 billion parameters. We have a small MOE and a small dense model, and those get a lot of great use. The middle size models we have not found as much—it's really stratified between the small models and the extra-large models. And the extra-large models are generally the hardest to serve with the most demand as well.

Anne (23:40) Is that because people are still basically over-provisioning on model, using an all-purpose approach?

Chad Gibson (23:52) Part of it is simplicity. If I take my workflow, there are still workflows where I'm using a big model to do something that is overkill, but it's on me to delegate to the small model the task that the small model can do. It's simple just to use one model for a given workflow.

There's things we're doing to help with that, and ways you can make it easier for different tasks to be delegated, but simplicity wins. The cutting edge is moving so quickly that people are still realizing how to fully exploit or maximize these frontier models. When the new frontier comes out, you need to relearn again: how can I re-establish my workflow now knowing the model is 20% or 30% more capable? The fact that the frontier is pushing while people are still trying to figure out how to maximize these models in their workflow pushes to this point of simplicity: "I'm just going to push everything on the high model for a bit and see what I can learn and evaluate from that." Then right when I get comfortable with that, yet another new model comes out and sets the bar even higher.

Anne (25:13) And as you say, there's so much potential ability to effectively get more out of the software that we haven't really even scratched the surface of yet.

Chad Gibson (25:26) Right. Totally.

Anne (25:30) So there's really no end in sight, is there?

Chad Gibson (25:34) I don't see the end. Conceptually, it's an interesting thing to contemplate. Now the new open weight frontier models are coming out with 2.8 trillion parameters. Generally, if you take a look at the landscape of hardware—NVIDIA, AMD, and others—they look at hardware capacity of a single chip and a single server, and then they have their scale-out pattern. An example with NVIDIA: they have Blackwell, where there's a Blackwell chip, a B300 server that has eight of these interconnected, and then a Ruben rack scaling this out to the biggest compute in one rack. These frontier models are now pushing the point where you need just about two 8-GPU servers. Theoretically, parameter counts could increase to a point where you may see a model where you need a full rack to serve it. Conceptually speaking, I don't think the ceiling on parameter counts is necessarily in sight.

While parameter counts are being pushed, you're also seeing models get better at smaller parameters. While the frontier is pushing model size, there's amazing innovation happening with making a 300 billion parameter model better and better, making the value of each weight higher. NVIDIA is pushing the increased capability of each chip, which increases the capability of each server and each scale-up unit. The fact that all these things are progressing, I don't have a good sense for which of those three things in the next 12 months is going to be a limiting factor. I can see 12 months of growth across all three. In 24 to 36 months, it's unclear because it's moving so quickly.

Anne (27:51) Right, and it's a bit terrifying because you've got a good data center, doing all the things that we said in Building Green Software—adding flexibility to the grid by working with grids to dial up and down, using hardware or excess electricity that was going to waste, and providing data to people so they can act on it. But there are so many data centers out there which do sound pretty terrifying. They're hooking up brand new gas-fired power stations. You can kind of see the #NotAllDataCenters, but there are some pretty evil data centers out there.

Chad Gibson (29:04) On-site power—if you're impatient, the easiest thing to do (it's not easy, but) minimizing reliance on the grid is one way to do it. There are ways to do that where you could put a modular data center where there's a big solar array with abundant solar energy and batteries, and there's some pretty cool stuff happening there. But there's also ways of just hooking up more diesel generators. If we can shine a light on some of the more constructive ways of doing it, we can address that fear we talked about and that anti-data center sentiment. Oftentimes, the cases you describe are the ones that become the examples used.

Anne (30:00) The difficulty with that is that people hear only about the bad stuff and think there is no choice—it's like AI or being sustainable and green, an either/or where people are always going to choose AI. We need to shine a light on the fact that there are alternatives; you do not have to use these giant data centers.

Chad Gibson (30:31) When we look at the transition to cloud computing, when people first started moving on-premise workloads to AWS, Azure, and Google Cloud Platform, all those platforms out of necessity started providing carbon observability for cloud workloads. That's now a feature of all those cloud platforms, evolving to provide more regional-based tracking versus doing tricks with carbon offset credits. That level of transparency we absolutely believe is going to come to AI. We're seeing it in Europe already, where customers want a direct correlation between where workloads are run and the state of energy generation correlated with it. I do believe that will come worldwide at some point.

The challenge in the immediate short term goes back to speed. Right now everyone is racing to understand how to leverage AI, driving more pressure to build out capacity to service these workloads. Even the suppliers themselves are trying to figure out how to scale effectively and help others use AI. Speed is putting pressure on all aspects of this.

Anne (32:08) If you had advice for listeners and watchers of this podcast about what they can do, what should they be thinking about and opening their minds to in AI?

Chad Gibson (32:31) First of all, to plug NeuralWatt Cloud, there are ways to build and use AI where you can correlate it to energy and carbon usage. If you want to answer the question of how much energy your AI use is consuming, we can answer that for you.

Second, customer demand is going to change this. If you're a huge Anthropic fan (which I am as well), letting Anthropic know that you'd love to know how much energy is being consumed by Opus and that you want that transparency—that's a meaningful thing. We've proven it's possible and can be done. We would love to help Anthropic satisfy that need.

When it comes to the question of data centers and power bills, I hope this conversation has surfaced that there are constructive ways of doing it. Not all data centers are going to put gas generators in your backyard. Allowing capacity to come online that makes the grid more flexible and consumes renewable energy is a good thing, and in some cases will mitigate how much new turbines and generators we need to create elsewhere. Hopefully that provides a view on the nuance. There are options to consume AI and be an AI champion and optimist that allow you to understand energy consumption.

Anne (34:26) That was very useful for us all to be reminded that you do have options here—it isn't AI or the planet. If you can't do anything else, just ask your suppliers. Say, "I care about this; can you start telling me what your plans are, because I will be making purchasing decisions based on this?" Is there anything else you want to add? I'm perfectly happy for you to keep plugging NeuralWatt because you offer a service that should be offered more widely.

Chad Gibson (35:16) When we offer inference, we price it by the kilowatt-hour, which is both a really cool thing and a complicated thing. For users used to paying by the token, consuming AI by energy is a challenging shift. We've had to make that easier by providing a lot of transparency. You know how much energy you use, and now you can optimize for that.

How do you optimize for that? There are certain times of the day where energy is more abundant for AI, and you can do this with carbon as well. There are times of the day where your grid is emitting more carbon, and times when it's not. This is a whole new frontier of optimization opportunities. A lot of our current customers are pushing us on this. Largely, inference today for our cloud is a human piloting something calling our service, but more and more it's transitioning to a human who created an autonomous agent that is calling our service. I was at an AI conference a month ago where they're planning for a future of autonomous swarms of software agents.

The cool thing about agents is if you give them data to optimize, they're really great at optimizing. We can provide transparency: "Hey, there's way less carbon being emitted on the grid right now," or "Energy is really cheap and abundant right now." For tasks that don't require an immediate response, let's do those tasks when there's less carbon being emitted. The optimization potential becomes pretty profound. If you give a human software engineer fifty variables to optimize against, it's hard for them to figure out if they should wait an hour to run code. But for workloads that are increasingly autonomous, there's phenomenal optimization potential.

That's an exciting proposition that goes down the path of doing more with the resources we have. It not only minimizes consumption, but it also makes it cheaper, and everyone loves that.

Anne (38:31) They love things to be cheaper, but they don't like to go to too much effort to get it. But the barrier to what is difficult is changing because of AI.

Chad Gibson (38:50) I agree. All these features and capabilities—we're trying to move as quickly as we can while everything around us is changing. It's exciting. I think about my career at Microsoft—it's an awesome place where I got to work on massive, high-impact products—but the speed at which I'm moving now is one of the reasons I wanted a post-Microsoft career. Being around the technology this quickly, new interesting problems emerge every week.

Anne (39:44) We all said we wanted to be in the tech industry to solve difficult problems; we had no idea, really, did we?

Chad Gibson (39:50) I know. Are they ever going to be solved? There's going to be new ones all the time.

Anne (39:58) Well, on that happy note—and it is a happy note because we did go into the tech industry to solve difficult problems, and now it's more difficult and faster-moving than ever. Nothing has changed at the rate we're seeing at the moment.

Thank you very much indeed for being on the podcast. It was a delight, really interesting, and very inspiring. There are things that we can do and tricky problems to solve. Thank you very much for being on the podcast, and thank you to all our watchers and listeners across the globe. Hopefully I will catch you again on a future episode of the Asynchronous and Unreliable podcast. Thank you very much.