Insane Demands of AI & Datacenters: Exowatt’s Mission for Affordable, Sustainable Power| E2039

AI search energy consumption and corporate willingness to pay for AI (0s)

• Typical AI search uses 10 to 25 times more energy than the search used just a year ago, making it super energy hungry, with the energy demand and growth being extremely high (4s).
• To support a large number of users, such as 100 million or a billion, the energy requirement would need to increase by 10 to 25 times, which is an insane energy demand (21s).
• Despite the high energy costs, corporations are willing to pay 5 times the rate to win in AI, showing that AI knows no cost boundaries (37s).
• This willingness to pay high costs for AI is surprising, especially considering that modern corporations have been cutting back on costs, trimming headcount, and reducing software spend (34s).
• The high energy consumption and costs associated with AI search are not yet fully understood, and the extent of the issue is not yet fully appreciated (26s).

Introduction to power, electricity, and Hannan Parvizian of ExoWatt (1m30s)

• The world is facing an enormous power crunch, with Europe potentially seeing a 40-50% gain in power demand by 2033, according to Goldman, and data centers in the US possibly boosting power demand by 20% (1m52s).
• The increasing power demand is driven by the construction of bigger and more powerful data centers, which requires a better power generation system and grid (2m6s).
• Hanan Parvizian, co-founder and CEO of ExoWatt, is working on a solution to help plug the gaps in domestic and global power systems (2m23s).
• The US net on-grid demand forecast through 2035 shows a plateau in power demands from 2005 to 2020, followed by a significant increase in forecasted demand (2m55s).
• Parvizian believes that the forecasts might be underestimating the actual power demand due to the recent AI boom and the energy-intensive nature of data centers (3m25s).
• The shift in focus from swapping out coal-fired power plants for solar or hydro to adding new power generation capacity changes the calculus of power investments (4m16s).
• The increasing power demand is driven by the growing use of AI products, which are energy-intensive and require significant amounts of power to operate (3m50s).
• The industry, federal, and state levels are concerned about the amount of energy needed to power energy-hungry data centers (3m39s).

Renewable energy, fossil fuels, and data center energy consumption (4m21s)

• Renewables are no longer seen as a substitute for fossil fuel power generation systems, but rather as part of a mix that includes both, and even that may not be enough, prompting the search for more sources of generation (4m29s).
• Companies like Kitecraft in Germany are exploring innovative solutions, such as flying turbines, to address the growing energy demands (4m42s).
• Data centers are a significant contributor to the increasing energy consumption, with global data center electricity consumption expected to rise sharply in the future, as shown in a chart from Goldman (4m54s).
• The average 100-megawatt data center consumes the equivalent of 80 to 100,000 households worth of energy, highlighting the energy-intensive nature of data centers (5m36s).
• The energy density of data centers is expected to increase, with data center racks becoming even more energy-dense, leading to higher energy consumption (6m5s).
• The goal is to make data centers more energy-dense, allowing a 100-megawatt data center to potentially output or consume 500 megawatts from an energy perspective (6m23s).
• There is a need to prepare for a significant increase in energy generation to power the growing number of data centers, which will require a substantial amount of energy (6m36s).
• A gigawatt data center, equivalent to 10 100-megawatt data centers, would consume energy equivalent to more households than there are in some small states, such as Rhode Island (6m55s).
• The growth of data centers is a global phenomenon, with a series of gigawatt data centers expected to be built around the world, leading to a massive increase in energy consumption (7m9s).

AI race, energy demands, and founding of ExoWatt (7m15s)

• Big tech companies are announcing ambitions for large data center campuses, with sizes reaching 100 GW, and are willing to spend heavily to win the AI race, with no signs of conservatism in their spending due to the "winner take all" nature of the competition (7m27s).
• The foundation story of ExoWatt began with its founding in 2023, as a recent startup that originated from a venture studio called Atomic (7m49s).
• The idea for ExoWatt was born out of a thesis developed by Jack, co-founder of ExoWatt and CEO of Atomic, and his partners, which focused on the second-order effects of the AI boom (8m16s).
• The thesis was developed at Atomic, a venture studio where ExoWatt was started, and was driven by the observation of the AI boom and its related effects (8m13s).

Key challenges in the renewable energy sector (8m25s)

• The focus is on the infrastructure layer of AI, including data centers, power, and chips, which presents a huge opportunity for growth and is expected to expand rapidly (8m25s).
• The goal is to power AI data centers and data centers as a whole with renewable energy, which is not the current standard for bringing power (8m50s).
• A modular system that can be scaled with different projects and data center sizes and deployed quickly is necessary to bring the product to market and start powering AI data centers with renewable energy (8m58s).
• The modular system would enable the powering of both existing and new AI data centers being built using renewable energy (9m12s).
• The core idea behind the project is to develop a scalable and quickly deployable system to power AI data centers with renewable energy (9m21s).

NetSuite - Download the CFO's Guide to AI and Machine Learning for free (9m24s)

• The future of business is uncertain, with varying opinions from experts, and to navigate this uncertainty, businesses need to future-proof themselves with the help of NetSuite by Oracle, the number one ERP tool trusted by over 40,000 businesses (9m42s).
• NetSuite blends accounting, inventory, financial management, and HR into one single platform, providing a single source of truth, real-time insights, forecasting, and visibility and control to make quick decisions (9m57s).
• To learn more about the opportunities and challenges of AI and machine learning, the CFO's guide to AI and machine learning is available for free download at netsuite.com (10m14s).
• The goal of Exowatt is to produce electricity as cheaply as possible, making AI applications more feasible for a larger audience and reducing the cost of energy, which accounts for 60% of data center operating expenses (10m42s).
• Exowatt aims to make electricity so cheap that customers can make an economic decision to choose the cheapest energy possible, without having to pay a renewable energy premium (11m21s).
• The company started with an ambitious goal to produce electricity at 1 cent per kilowatt hour and worked backwards to develop a solution, which led to the creation of the XP3 product, a combination of generation and storage (11m49s).

Cost of electricity, energy bills, and Texas energy market (12m23s)

• The cost of electricity for a typical household can range from 12 to 15 cents per kilowatt hour to the high 30 cents per kilowatt hour range, depending on the location (12m47s).
• A typical household consuming 800 kilowatt hours of energy per month would pay 800 times the cost per kilowatt hour, which is $160 at 20 cents per kilowatt hour (13m3s).
• The cost of 1 cent per kilowatt hour is two orders of magnitude cheaper than the typical household energy bill (13m13s).
• In the wholesale markets or commercial energy, people pay on average between 8 to 12 to 15 cents per kilowatt hour, depending on the state (13m33s).
• The price of energy in the wholesale markets is cheaper than in the retail markets, but not by a large margin (13m27s).

Exowatt’s unique technology and competitive edge (13m40s)

• The energy market in Texas is a free market where people pay for energy at the cheapest price available, resulting in fluctuating prices that can range from a few cents to hundreds of cents or even dollars per kilowatt-hour (13m41s).
• The Texas model has pros and cons, with the pros including driving competition and promoting renewable energy development, but the cons include price volatility and potential high energy bills during winter (14m32s).
• Texas is the leader in renewable energy development, not California, due to its free market approach (15m8s).
• However, the Texas model can be problematic during winter storms, causing consumers to suffer (15m21s).
• The idea of an "all of the above" approach to bringing more power generation online is appealing, but nuclear energy may not be a viable solution to the AI power crunch in the near future (15m31s).
• Nuclear energy faces significant regulatory, community acceptance, technical, and manufacturing challenges, making it unlikely to solve the AI power challenges within the next 10-15 years (16m10s).
• The promise of nuclear energy is amazing, but the reality is grim, and it may not be a solution to the current energy crisis (16m51s).
• Burning fossil fuels for another 10 years to power AI data centers while waiting for nuclear energy to come online is not a viable solution (17m4s).
• Nuclear energy has the advantage of consistency and high base load, but it may take a long time to come online, and storage is being better coupled to renewables to resolve the base load question (17m22s).

Introduction to ExoWatt's P3 product (17m46s)

• ExoWatt's P3 product is a modular system that fits into a 40-foot shipping container, consisting of a series of batteries that store heat and solar lenses that convert the sun's energy into stored heat for electricity and industrial needs (17m50s).
• Each module is roughly the size of a 40-foot shipping container, approximately 40 feet long, 8 feet wide, and 8 feet tall, made of free elements (18m21s).
• The P3 system has three main elements: the renewable energy collection system, the heat battery, and the power generation system (18m29s).
• The renewable energy collection system is powered by the sun, using custom-developed lenses that capture energy throughout the day and convert it into high-temperature heat (18m31s).
• The heat is stored in a custom heat battery cell that can store energy with minimal losses, and the battery works as a sensible heat battery, meaning it is a solid block that doesn't undergo chemical or electrical chemical reactions, mechanical parts, or phase changes (18m49s).
• The heat battery can store energy up to 1,000 degrees Celsius or about 2,400 degrees Fahrenheit and retains this energy for as long as needed (19m20s).
• The main advantage of the heat battery is that the cost of storage is significantly lower than electricity, allowing for longer storage with no degradation and at a much cheaper cost (19m32s).
• This makes the P3 system useful for applications that require dispatch for 12-24 hours a day, such as data centers, which can store energy very cheaply with renewable energy (19m55s).
• Traditional solar panels and lithium-ion batteries would become cost-prohibitive for this application due to high efficiency losses and the cost of storing electricity (20m20s).
• Electrochemical batteries, like those used in phones and other electronic devices, degrade over time, becoming less efficient and requiring replacement, whereas heat batteries do not have this issue (21m0s).

Gusto - Get three months free when you run your first payroll (21m34s)

• Starting a company does not necessarily mean having to run payroll, as Gusto is available to help with this task (21m36s).
• Gusto offers a range of services, including running payroll, handling benefits, onboarding, and HR, all in one place (21m40s).
• Over 300,000 businesses trust Gusto today, and the platform can grow with a startup as it scales (21m48s).
• Gusto handles state and federal taxes for staff across the country (22m3s).
• The platform also offers 401K plans and helps with compliance, with three out of four employees saying Gusto helps them be government compliant (22m8s).
• Gusto is simple, easy-to-use software that allows users to focus on building their company (22m17s).
• Gusto offers a discount with no hidden fees, and new users can try the platform and get three months free at gusto.com/twist (22m29s).

Energy storage technologies and comparison (22m39s)

• The drawbacks of lithium-ion batteries are well-known, and various startups are working on alternatives such as sodium-ion batteries, pressurized gas energy storage, and gravity batteries, but a new approach involves using a substance that can store energy in the form of heat for a long period of time without degrading or undergoing phase changes (22m39s).
• The concept of sensible heat batteries isn't new, as people have been using them in other applications, such as storing electricity from the grid and dispatching heat to support industrial heat loads (23m13s).
• What's unique about the Exowatt approach is the design of the battery material, geometry, and heat exchangers to capture energy from the sun directly, store it effectively, and dispatch it to an engine that can convert it to electricity (23m30s).
• The technology and physics of heat batteries aren't new, but the innovation lies in the material, geometry, and heat exchangers used to capture and store energy from the sun (24m1s).
• A simple example of how heat batteries work is by using a rock to store heat from the sun, which can then be used for cooking, and Exowatt's technology is similar, but uses a magnifying glass-like approach to heat up a rock material throughout the day, allowing it to stay hot throughout the night (24m10s).
• The Exowatt technology is not a new invention, but rather a revisiting of an old concept, as people used to use rocks to store heat for cooking before the advent of indoor ovens (25m8s).
• Batteries have tradeoffs, and the Exowatt heat battery has its own set of advantages and disadvantages, including energy density, which is measured in watt-hours per kilogram (25m22s).
• The energy density of the Exowatt heat battery depends on the material used, but it is generally lower than that of lithium-ion batteries, which can store between 300 to 400 watt-hours per kilogram (25m36s).
• Heat batteries have a lower energy density compared to lithium-ion batteries, requiring more material for the same amount of energy, with an energy density of 50 to 100 watt-hours per kilogram (26m3s).
• However, heat batteries are significantly cheaper, costing between $1 to $10 per watt-hour, compared to lithium-ion batteries which cost around $245 to $300 per watt-hour (26m16s).
• The tradeoff between energy density and cost makes heat batteries a viable option for certain applications, particularly those where weight is not a concern, such as data centers (26m43s).
• Solar lenses, also known as Fresnel lenses, are used to collect solar energy and are essentially large, flat magnifying glasses with grooves on one side and a smooth surface on the other (27m47s).
• These lenses are used in modules that collect enough energy throughout the day to charge a battery pack, and multiple modules can be combined to deliver a desired amount of energy (28m8s).
• Each module is capable of powering a heat engine for up to 24 hours a day, making them a suitable option for data centers and other applications requiring a reliable source of energy (28m39s).
• The use of lenses allows for more efficient collection of solar energy, with typical solar PV panels converting incoming solar energy at an efficiency of around 20% (29m11s).
• Concentrated photovoltaic (CPV) systems can capture or collect up to 80% of the incoming energy, which is a lot from an efficiency perspective, with each individual lens being very powerful. (29m29s)
• The power of each individual lens can reach temperatures of up to 1400° Celsius or about 2,800° Fahrenheit, even in non-summer months in locations like Miami. (30m1s)
• The lenses used in the system are relatively cheap, which is why they were chosen to help achieve the goal of getting to 1 cent per kilowatt hour. (30m31s)
• The system is designed to put all the heavy components behind the lens, on the ground, eliminating the need to build structures to suspend anything or put it in the air. (30m19s)
• Locations with a lot of direct sunlight, such as California, East California, Mojave Desert, Nevada, Arizona, and Texas, are ideal for this system. (29m49s)

Washington Post - TWiST listeners can subscribe for just 50 cents per week for your first year (30m47s)

• Subscribing to The Washington Post is recommended for those who care about current events and want reliable news coverage (30m48s).
• The Washington Post offers various features, including apps, podcasts, and custom newsletters, making it easy to stay informed even for busy individuals (31m5s).
• One of the features of The Washington Post is the ability to have articles read aloud, allowing users to stay up-to-date while on-the-go (31m12s).
• A special offer is available for subscribers, providing 80% off the typical price, making it just 50 cents per week for the first year (31m32s).
• This offer can be accessed by visiting washingtonpost.com/twist and subscribing to The Washington Post at the discounted rate (31m43s).
• The cost of the subscription is equivalent to the price of a cup of coffee every three or four months, making it a great deal (31m51s).

P3 modules' energy capacity and storage (31m58s)

• Each P3 module can have multiple battery packs, which determine its energy capacity, with a standard off-grid module having two battery packs that produce 25 kilowatt hours of electricity (32m19s).
• The P3 module also has a thermal capacity, which varies depending on the temperature at which it is charged or discharged (32m36s).
• In an industrial setting, heat is a valuable commodity, almost as valuable as electricity, as it is used as an input in various industries such as manufacturing, steel, chemicals, food and beverage, and cement (33m1s).
• Heat is used at different qualities and grades, from low-temperature heat for boiling water and purification to high-temperature heat for cement and steel industries (33m28s).
• The carbonization of industrial heat is a $60 billion industry, with many startups and companies focused on this application alone (34m8s).
• The P3 module's primary focus is on providing electricity for powering data centers, rather than heat, as data centers do not take heat as an input (34m20s).
• The mechanism by which heat is converted to electricity is not specified, but possibilities include heat exchangers, Sterling engines, or turbines (34m48s).

Heat-to-electricity conversion mechanism (34m55s)

• The primary factors in selecting a heat engine were a modular system with a scalable supply chain, leading to the study of various heat engines such as turbines, Brighton cycles, and semiconductor-based heat exchangers (34m55s).
• A Sterling engine was chosen as the primary heat engine due to its simplicity, reliability, and available supply chain, as well as limited maintenance requirements compared to turbines (35m30s).
• Sterling engines depend on a heat differential to oscillate a cylinder, using the differential between a hot end and a cool end to expand gas and spin a wheel (36m3s).
• The Sterling engine can be a simple, low-tech device, with examples available for purchase on Amazon, and can be powered by a heat source such as a coffee mug (36m20s).
• The heat differential between the hot and cool ends of the Sterling engine creates a spinning motion, which can be used to generate power (36m59s).
• The use of a Sterling engine to power cutting-edge digital technology is seen as poetic, as it combines old, elemental technology with new, innovative applications (37m17s).
• The company believes in standing on the shoulders of giants, using centuries of technological development to innovate and make existing technology more useful for powering AI and other cutting-edge technologies (37m37s).

ExoWatt's company background, funding, and scaling challenges (37m56s)

• Exowatt has raised a $20 million round of funding from investors including Jason Horowitz, Sam Altman, and Atomic, which has helped the company launch its XP3 product to the public (37m58s).
• The public launch of XP3 in Anaheim generated a significant amount of interest, with orders received for the next 3 million units, equivalent to 5 gigawatts of demand and 85 gigawatt-hours of energy (38m46s).
• The demand for XP3 systems has exceeded expectations, with 90% of interest coming from data centers, and the company expects this demand to double (39m50s).
• Exowatt is currently executing against purchase orders from early customers and plans to deploy units in the field in places like Florida and Texas in the coming months, with operational deployment expected by Q1 2025 (40m14s).
• The cost of a single XP3 unit is $77,500, with a decent amount of gross margin built in, and the company expects to bring down the cost of building these units further, increasing gross margin and benefiting customers with a cheaper source of electricity (40m48s).
• There are recurring revenue components to the XP3 business, although the specifics of these components are not detailed (41m31s).
• A company has sold 3 million units at $72,000 each, which is a significant business achievement (41m35s).
• The company also offers software on a subscription basis, which includes control software for running modules and a digital twin for predictive and proactive maintenance (41m47s).
• The digital twin allows the company to predict the performance of the system and perform maintenance tasks (41m58s).
• The company's team of PhDs has developed a high-fidelity model of the system, which is a testament to their work (42m9s).
• To scale up manufacturing, the company may need to raise additional funds, potentially $100 million, to meet the demand from their order book (42m15s).
• However, the company is not planning to do vertical integration of manufacturing; instead, they are partnering with leading contract manufacturers to build and ramp up production (42m37s).
• The company is optimizing for bringing the product to market quickly, as their data center customers prioritize time to power over cost (42m57s).
• The company is working with a contract manufacturing network to achieve this goal (43m10s).

Variable pricing, customer demand, and AI's energy consumption outlook (43m14s)

• Tech companies often use variable pricing, and given the scale of demand, some customers might be willing to pay a premium to skip ahead, with some customers willing to pay up to 500% premium for access to behind-the-meter power or power for data centers (43m14s).
• Modern corporations are willing to pay 5x for power despite being parsimonious and cutting back on software spend, which is a surprising but understandable trend given the energy-hungry nature of AI (44m1s).
• AI searches, such as those using ChatPT or Google, use 10 to 25 times more energy than searches from just a year ago, leading to insane energy demand and growth (44m21s).
• The energy demand for AI is expected to continue growing, with some predictions suggesting 1.9% annual growth, but this number may be conservative, and actual growth could be 5-6% annually (44m57s).
• Historically, predictions about energy markets have been wrong, and it's likely that the growth of AI and data centers will be underestimated (45m24s).
• Data centers currently represent 2% of global emissions, but if they grow at predicted rates, they will account for 10% of global emissions by 2030, which is comparable to the emissions of the cement industry (45m44s).
• The cement industry is responsible for 5-8% of global CO2 emissions, and data centers could potentially reach up to 2x the emissions of the cement industry (46m7s).
• The demand for power is high, and people are willing to pay a premium, but the actual number of power modules that can be created and built next year is uncertain, with estimates ranging from 20 to 20,000 (46m24s).
• Exowatt aims to produce a couple of thousand units next year and ramp up production to a million units in the next couple of years, with a goal of making affordable and sustainable power more accessible (46m36s).
• The company's modular system is designed to be easy to install and set up, with the ability to build millions of units in a factory to bring down the cost curve (47m5s).
• The system is an "install and forget" solution, with no on-site construction required, which helps to reduce costs and make it more efficient (47m23s).
• While the system is prioritizing commercial applications, it is technically possible for individuals to use it for their homes, but financing may be a challenge (47m43s).
• The system is designed to be scalable, and it could be possible to cut it into smaller units to power a single home or a weekend house (48m17s).
• The system can work with varying amounts of solar energy, but it is more effective in high-sun environments, and it may not be suitable as a round-the-clock solution in lower-sun areas (49m24s).
• However, it can still work in lower-sun areas, but it may only be able to provide power for fewer days per year (49m34s).
• The system's effectiveness is dependent on the amount of solar energy available, with areas like East Oregon being suitable for its use (49m2s).
• The company's goal is to make sustainable power more accessible and affordable, with the potential to power entire operations with a single unit (48m35s).
• A location with approximately 300 days of sunshine per year can effectively provide the equivalent of base load power, allowing for the complete operation of an application offline if desired (49m47s).
• However, it is not recommended to completely go off-grid for applications requiring a certain level of reliability, backup, or redundancy (49m59s).
• Theoretically, it is possible to run an application completely off-grid in a location with abundant sunshine, but this may not be suitable for all applications due to reliability concerns (50m7s).

American manufacturing dynamism and AI's impact on the job market (50m10s)

• American manufacturing dynamism is crucial, and companies like Exowatt are making a statement by choosing to base their manufacturing in the US, despite the challenges and potential cost savings of manufacturing in countries like Mexico or China (50m12s).
• Exowatt has made trade-offs in material selection to ensure that their products can be sourced from the US, and while it wasn't the easiest task, they have worked hard to establish a US-based supply chain and manufacturing process (50m41s).
• The trend of onshoring manufacturing is becoming more prominent, with initiatives like the American Dynamism fund pioneering the way, and opportunities for local manufacturing and recycling of materials are increasing (51m15s).
• Despite the progress, the US is still not super cost-competitive, but it's getting there, and the effects of having manufacturing back onshore are being felt across the broader American industrial footprint (51m51s).
• The growth of AI and data centers is creating new job opportunities in areas like green technology and climate-related jobs, which will benefit the American population and job market (52m35s).
• Companies like Exowatt are working to address concerns about the environmental impact of AI and data centers, and their efforts are contributing to a more sustainable future (52m53s).
• The development of sustainable power solutions is crucial to support the growth of AI and data centers while minimizing their environmental impact (53m0s).