Author: Techub News Compilation

In this conversation at a16z, two heavyweight guests—former Tesla VP of R&D Drew Baglino and former SpaceX engineer Doug Bernauer—shared their profound insights on the current energy infrastructure crisis and the solutions their respective companies are proposing. This conversation is crucial because it addresses the most pressing reality under the wave of AI: Can our power grid support the future? These two engineers, who were deeply involved in Elon Musk's cutting-edge projects, are now applying "Tesla/SpaceX-style" engineering thinking—focused on modularity, manufacturability, and software-defined systems—to the most traditional yet critical field: power systems.

The Grid Has Reached Its Limit: Efficiency Bonuses End, Transmission Becomes the Bottleneck

At the start of the dialogue, Baglino pointed out a long-ignored fact: over the past forty years, total energy service in the United States has grown, but the actual net electric power delivered has remained essentially flat. This is thanks to ubiquitous efficiency improvements—from industrial variable frequency drives to lighting and cooling equipment. Even in data centers, significant advancements in computational efficiency (floating point operations per watt) mean that the growth in computing power has not significantly increased the share of power consumption.

However, the times are changing. Baglino pointed out that the rise of AI and its unprecedented widespread application is intersecting with trends such as the electrification of transportation and industry, creating a broad foundational demand for electricity. The growth in electricity is surpassing historical levels at an unprecedented rate. This is not only due to the diminishing marginal effects of efficiency improvements, but also because "we have found more reasons to consume electricity."

Bernauer approached the topic from a different angle, bluntly stating that "the grid is collapsing." The real bottleneck is not generation capacity—last year, the U.S. grid added record capacity, and this year is expected to be even higher—but rather transmission and distribution systems. Those transmission lines crisscrossing the nation constitute an extremely complex and massive "organic machine," but there have been few fundamental changes for decades. Due to stagnant growth, top talent has flowed into other fields (such as South Korea, Japan, and China), leading to a hollowing out of the knowledge base. Now, the explosive growth in demand forces us to reinvest and rebuild this system.

In Baglino's view, the explosion of AI is a "good forcing function." It compels the U.S. to relearn how to build large-scale infrastructure and reminds us that energy, power, and delivering it to where it is needed are crucial.

New Paradigm: Software-Defined, Decentralized, and Edge Growth

The traditional grid model is unidirectional: electrons flow from large centralized power plants to your power outlet. But in the face of increasingly complex and more volatile loads (whether they are data centers, edge defense use cases, or others), what we need is resilience, decentralization, and software-driven workflows.

Baglino emphasized that it is impossible to expand the grid for all potential peak power demands (even at the most remote edge use cases). Therefore, the core of energy investment thesis is: how to transform the grid into a more software-defined, resilient, decentralized, and streamlined system. This requires microgrids (behind the grid, within the grid, off-grid, hybrid models) and immense flexibility.

Bernauer offered a more fundamental perspective: "The grid itself is civilization." Electricity is civilization. He envisions a "rebirth" of civilization based on a brand new electricity transmission architecture. The key lies in utilizing all "free energy": sunlight is free (captured through solar panels), and uranium is also free (it's right underground). If we do not harness it, it will spontaneously decay and "disappear."

More critically, the growth of the grid and the possibility of expanding from the edge is something the U.S. could hardly imagine in the past 50 years. The grid has always been a top-down project. Today, data center operators wanting to connect to the grid are profoundly experiencing this: the process is like a nightmare. Part of the reason is that the foundation of the grid is a mechanical system, slow to respond, lacking telemetry data, and having (or only very slow-responding) software. In such a world, it is hard to imagine a bidirectional, dynamically controlled grid.

Baglino’s company is providing solutions for this. Their first product, “Heron Link,” is a 5-megawatt bidirectional solid-state transformer that converts 800-1500V DC to 34,000V AC (this is the secondary transmission voltage for data centers, large battery stations, and solar facilities). At its core, it uses power semiconductors and software, replacing traditional transformers that rely on large coils, magnetic steel, and oil drums for low-frequency voltage conversions. Just as smartphone chargers evolved from bulky "power bricks" to compact and efficient multi-output devices, they are applying this high-frequency, high-power density conversion technology on an industrial scale.

This modular architecture (composed of 30 x 165-kilowatt modules) possesses "fault-tolerant" characteristics, meaning even if a single module fails, the entire system can still operate. This provides a foundation of software control and dynamic response for the grid.

Micro Nuclear: Turning Nuclear Reactors into "Products"

Bernauer's company, Radiant, focuses on a more radical concept: manufacturing nuclear reactors as "products." His inspiration stems from the dilemmas encountered while designing energy solutions for Mars colonies at SpaceX: solar solutions require "a series of miracles," while Musk's suggestion was "you should look into nuclear energy."

He bluntly stated that "the nuclear industry does not exist" at present. Many nuclear energy startups are approaching a critical milestone: before July 4 of this year, several companies will complete and achieve a critical state with newly designed reactors. It feels a bit like "talking about flight before Kitty Hawk."

Radiant's goal is to manufacture mass-producible, portable micro-reactors. They are currently focusing on truck-sized specifications (about 1 megawatt). Their product is designed for off-grid use: built at factories, transported to customer sites by truck or plane, and activated to power customer locations within 48 hours from the time the wheels stop turning. It can run sustainably for 5 years, equivalent to the energy of 2 million gallons of diesel.

Bernauer emphasized that their reactors use "melt-down proof fuel" and are completely safe at customer sites. At the end of their operational life, the company will remove them, and customers need not deal with waste or other complex issues. This fundamentally changes people's perceptions of nuclear energy: it is now a viable option and appears to be better than almost any other form of energy.

He does not believe micro nuclear is the only choice, but advocates a combination block of "solar + batteries + nuclear." Micro nuclear fills a specific niche: it does not compete with large centralized reactors (which economically prevail) but competes with diesel generators. Their economics are competitive when diesel prices are about $6.50 per gallon. And in many edge regions (like Hong Kong, Northern Europe, and islands), diesel prices are much higher, creating a vast market.

Radiant does not aim to solve the "miracles" faced by large nuclear power plants that dig big holes in the ground. Startups cannot handle too many miracles at once. Their planned production scale is about 50 reactors per year, with a maximum of a few thousand. They do not want to pursue tens of thousands, as that would require different products and economics.

Bernauer also mentioned a key advantage: extremely low on-site workload. Unlike traditional nuclear power plants that require massive cooling towers that can be visible for miles, their solution is "skyline-free" and rapidly deployable infrastructure. This can significantly alleviate community "NIMBY" (Not In My Backyard) sentiments. He even proposed redefining NIMBY to mean “Nuclear In My Backyard.”

Manufacturing Revolution: From "On-Site Construction" to "Factory Production"

Both men place extreme importance on manufacturability and modular design, clearly stemming from their experiences at Tesla and SpaceX.

Baglino pointed out that there is always an optimal balance between capital investment and manufacturing costs. Integrating the same functionality on highly automated production lines beats the total cost of "stick-built" power plants on-site. Their first factory goal is to produce 40 gigawatts per year. This accounts for about 10-15% of their product category's global market excluding China, equivalent to half the peak power of Texas (about 4% of total U.S. electricity). The choice of 40 gigawatts is based on market scale considerations and because approximately 60-second cycle times can maximize the factory's capital efficiency. Automation brings tremendous advantages in quality and cost.

Bernauer completely agrees with this concept. Building nuclear reactors in a factory is a “quite radical” idea. They are constructing reactors within existing facilities and have already expanded a second building due to the need for more vertical integration and machining. They plan to build a factory on an 80-acre site in Tennessee, with initial goals of finding buildings with sufficient power for component manufacturing, then transporting them to Tennessee for fuel loading, and finally delivering them to customer sites. The factory itself will undergo multiple evolutions (like factories before the "Gigafactory"), gradually automating what can be automated.

Baglino added that they are also following a similar path: building about 10 prototypes manually this year, then constructing 30-50 prototype systems at the factory location next year, and only beginning automation after learning from these two rounds of construction. “You must repeat practice.”

Supply Chain and Scaling Challenges

For the nuclear industry to truly thrive, Bernauer believes several conditions need to be met: nuclear fuel and enrichment technologies must enter a fully competitive free market, solving challenges through innovative startups; and a centralized waste storage facility must be established (which is safer for the existing nuclear fleet operational since the 60s). Radiant can handle waste using dry storage tanks on their own site due to its small scale, but large power plants need national-level solutions.

Baglino delved deeply into the supply chain of power electronics. Key components include: ferrite (essentially iron oxide, lacking rare earth elements but produced by the world’s largest manufacturers in Asia, and he is working to restore production in the U.S.), film capacitors (supply chains also primarily in Asia), as well as copper and aluminum foils. Other materials (plastics, sheet metals, aluminum castings, busbars, etc.) are abundant in the U.S. He is excited to see power electronics transitioning from device scale (like charging laptops) to the grid itself. Although approximately 3 terawatts of power semiconductors entered electric vehicles last year and peak power in the U.S. grid is less than 1 terawatt, the grid needs multiple power conversions along the way, and the total amount of power electronics that support it could be as high as 800 gigawatts. The slowdown in electric vehicle growth means we can redirect that momentum to new problems: the power needed for data centers, industrialization, economic growth, and sustainable energy.

Data Centers: A Problem for the Grid and a Solution for the Grid

Regarding the impact of data centers on the grid, Baglino provided a dialectical view. Overall, data centers will be "very beneficial" for the grid.

The root of current negative reports lies in the way data centers are designed to connect to the grid. To maintain computational availability (pursuing "six nines" reliability), data centers isolate and switch to backup power or UPS as soon as any anomaly occurs in the grid. When data centers are small (10 megawatts), this has little impact on the hundreds of megawatts or gigawatts of the grid. However, when constructing gigawatt-level data centers, this instantaneous disconnection poses a serious stability risk to the grid (recent events in Washington State and Virginia exemplify this).

However, this problem can be resolved through software, modern power electronics, and dynamic grid formation controls. Data centers can be equipped with a small amount of energy storage and remain online, even helping to stabilize the grid rather than disrupt it.

Regarding the commentary that data centers will drive up electricity prices, Baglino rebutted from a physics perspective. Electricity price is the total cost of maintaining/supplying power divided by the total kilowatt-hours delivered. Data center customers are “ideal customers”: they almost always operate close to maximum power, while residential customers may only reach 10% of maximum power for about an hour a day. Therefore, data centers increase total kilowatt-hours (the denominator), while utilities typically spread costs across all customers. More stable loads like data centers and factories will actually reduce overall electricity costs for everyone due to improved utilization.

He concluded that new generation capacity is not the problem; transmission is. Data centers will make transmission more economical by increasing the utilization of the transmission system. Therefore, data centers will overall drive down electricity prices.

This conversation clearly outlines a path: in the face of the tsunami of energy demand driven by AI, we cannot merely patch the old systems. We need new architectures that are software-defined, modular, and distributed, especially new options like "micro nuclear," starting from the edge to reshape our energy civilization. This is not just a technical challenge but also a fundamental shift in mindset.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。