Elon's Former Battery Chief on Making Transformers 100x Smaller | Drew Baglino, Heron Power

Summary

• Baglino’s claim: properly designed data centers can push electricity rates down, not up. His mechanism is utilization — a home has roughly 40kW of service and averages 1-2kW, while a gigawatt data center uses ~800MW on average, making them “basically like aluminum smelters or steel mills… the customers that make the grid affordable for everybody else.” He notes states with the highest data-center penetration have had the lowest rates and actual rate reductions, and with on-site storage absorbing “training ripple,” data centers can shift from grid liability — three gigawatts “just this week” turned off and disconnected together — to grid-stabilizing asset.
• Heron Power’s first product, the Heron Link, is a 5MW solid-state transformer that halves grid-to-chip loss — worth ~35 megawatts of extra useful compute per gigawatt data center. Wide-bandgap silicon carbide lets you build “a 10,000-volt transistor… actually smaller than the GPU,” so galvanic isolation happens at hundreds of kilohertz instead of 60Hz, shrinking the transformer 100x volumetrically per unit of power and letting the power electronics eliminate the additional motorized blade-switch stage.
• The market opening: US load growth ran at 1% or less from the ’80s through the 2010s, which “calcified” the utility supply chain. Utilities earn a guaranteed return on deployed CapEx rather than on kWh sold, so a flat-volume supply base got “fat and happy” charging more per unit — Baglino calls it “towing a dinosaur… a boat anchor on the path toward electrification.” With electrification growing at 3-5% per year and requiring roughly 3x today’s electricity, that supply base is the bottleneck Heron attacks.
• California’s high retail rates come from wildfire liability, a rural distribution footprint, aging assets, and net-energy-metering math — and Baglino says load growth may now reverse them. NEM forced utilities to buy rooftop solar back at retail rather than spot price while removing those kWh from the denominator over which fixed costs amortize; with EVs, heat pumps, and data centers restoring load growth, PG&E CEO Patty Poppe is saying rates could actually go down.
• His LCOE stack for baseload: 100% solar-plus-storage lands around 10 cents (7-8 in ideal desert), Fervo-style geothermal is targeting 5-6, fully depreciated nukes run 2-3 cents, and the new Georgia nuclear build came in above 10. “I’m down for nukes,” but he thinks geothermal “is gonna give nuclear a run for its money,” and Asia’s falling nuclear build costs suggest the US could relearn cheap construction — with the caveat that China has “subsidized everything that went into the nuke in some way or the other.”
• The definitive Elon management vignette: presented a rigorously modeled 8,000-8,800 cell requirement for Model S, Elon said “It can’t be more than 7,200” and ended the meeting. Baglino’s retrospective read is that impossible targets are the method — “loading everybody’s back equally” forced a more efficient powertrain and pushed Panasonic beyond its comfort zone, and the margin proved necessary when EPA range measurement changed unfavorably. Same pattern on Autopilot: the weekly question was “Can it be next week?”
• Baglino explicitly separates his own management philosophy from Elon’s. His creed — “decouple your work product from your ego,” fail fast, hire better than you — describes what he demands of his Heron team. When Lukas pushed that Twitter-era Elon hardly models zero ego, Baglino clarified: “I’m not talking about Elon.”

Deep dive

1. A Stanford field-trips class, not software, routed him to Tesla

• Baglino’s origin story runs through Gil Masters’ energy field trips class — Hoover Dam, wind farms, the SEGS parabolic-trough solar plants in the Mojave — where he met early Tesla founder JB Straubel. He kept in touch while coding economic models in Stata or R at the DC think tank Resources for the Future, “bored kind of out of my mind” because the work “was translating into nothing… in the physical world.”
• His first Tesla project: replace the analog motor, charge, and battery-management controls licensed from AC Propulsion with an embedded DSP. The hook was immediacy — driving Roadsters around the Mission with a laptop showing live signals, and converting a gas smart car bought in Mexico to electric, parked on the Bryant Street sidewalk: “it didn’t even block the sidewalk.”

2. First Elon encounter: “It can’t be more than 7,200”

• Modeling Model S battery sizing, his team walked Elon through assumptions — weight, drag, tire efficiency, EPA range — landing on 8,000-8,800 cells. Elon’s entire response: “It can’t be more than 7,200… meeting’s over.” The team’s reaction: “what just happened?”
• Baglino’s retrospective: Elon had context he lacked — cell costs and pressure on Panasonic’s energy density — “but I don’t actually think he did either. He just was doing the math in his head.” The genius is the mechanism: seemingly impossible targets that force “the A game out of everyone.” They hit it — more efficient motor, slipperier car, denser Panasonic cells — and needed all of it when EPA range methodology later changed unfavorably. At the time: “I never wanna be in a meeting with him again.”

3. A long Tesla tenure: thick skin, ego off — and earnings-call terror

• Lukas’s framing of the standard Elon-alumni story — “I told him the truth, he didn’t listen, then he fired me when it turned out to be the truth” — draws Baglino’s counter: “You need to decouple your work product from your ego,” accepting three months of work getting torched “for really good reason.” He credits early managers JB Straubel and Craig Carlson for teaching him to check ego at the door — “true in business generally, not just in an Elon company.”
• After JB left, he absorbed the cell roadmap, battery packs, propulsion, power electronics, the energy business, and became a company officer. The earnings calls: Elon arriving from Neuralink or “maybe even OpenAI at the time” with five minutes in the room before recording started. “You have the people that are gonna move the stock listening on every word… if you put your foot in your mouth, you’re probably not gonna be at the company the next day.” His verdict: “probably some of the more stressful moments of my life.”

4. Production hell and the flufferbot: delete the robot, then the part

• On Model 3 “production hell,” Baglino offers context, not defense: scaling from tens of thousands — perhaps ~100,000 Model S/X per year — toward millions meant “every aspect of the company needed to grow… orders of magnitude.” When an executive on the line says “just delete that thing,” the real message is holistic — “how do we make more product” — and over-indexing on one interaction misreads it.
• The specimen: a robot (“we called it the flufferbot or something”) installing a flimsy desiccant bag atop the battery pack, constantly erroring and stopping the line. In two or three days the logic cascaded — a person can fix it, don’t stop the line; if a person is fixing the robot, delete the robot; wait, delete the part entirely. The team then found “seven other versions” of the same opportunity. The painful edge, acknowledged: the engineer who spent two weeks coding the robot and defended it was “maybe not there the next day.”

5. Heron’s culture is explicitly separate from Elon’s: ego off

• Building Heron, Baglino values humility, transparent communication, and failing fast — ego attached to work makes you take “the positive side of any finding” (“this test result, it’s a little funny… maybe we need to retest it”) instead of hunting failures.
• Lukas’s pushback — worth keeping: the Elon of Twitter hardly looks like a zero-ego dispassionate observer. Baglino’s clean separation: “I’m not talking about Elon.” He’s describing what he asks of his own team, not his old boss’s style.
• The second pillar: hire better than you. People habitually “throw their weight over their reports” by needing to feel smarter; do the opposite — hire people smarter and more experienced, fill your blind spots, elevate them, “because that’s how you grow too.”

6. Schedule compression: keep cutting the long poles

• From his father, a Teradyne engineering manager: reconcile bottoms-up estimates with a top-down business view and push for something in between, “because if you give people 12 months, they’ll take 15 months… that’s just human nature.” But pragmatically — a 40-week stamping tool is “probably not that compressible,” so soft-tool around it. Teams play “long pole in the tent Whac-A-Mole,” erecting new poles to justify the schedule they want; the executive’s job is “to just keep cutting the poles down” — and when a long pole lands, everyone else rests. No: get your four-week part in, perfect it, “and go help the guy with the 36-week lead time.”
• Lukas’s software counter — especially for junior teams: they wildly underestimate (“the first 90%, the second 90%, the third 90%”), so the job is padding, not compression. Baglino agrees software differs: it’s about integration and functionality-demonstration milestones, because “you can basically fill time endlessly with more code or re-architecting over and over again.”
• The incompressible-deadline specimen: a 400-person Roadster event at an LA airport (“Arnold went”), every celebrity doing a four-second 0-60. Technicians rigged a dry-ice rapid chiller behind a curtain to survive back-to-back launches; the magnesium motor mount cracked at around run 300, leaving the motor thumping between battery pack and trunk on launch and regen — “but everything was fine.”

7. Luxury is mass

• What makes a luxury car: an NVH (noise, vibration, harshness) package — $500 of sound-absorbing material versus $50 in a budget sedan is “the difference between a quiet cabin and an acceptably noisy one.” The satisfying door thunk is door stiffness and seal lips: “you’re trading some mass… and some cost for these creature comforts.”
• He’s not a car guy — “I will never drive a non-electric car, or at least not happily” — but his track experience taught him “where you are relative to the edge of stability… just from the feedback you get through the steering wheel.” He says he drives faster mainly because “I’m always chronically late.”

8. Autopilot’s origin: a Mobileye demo and “Can it be next week?”

• In 2012 or 2013, JB tasked him with making Model S competitive on adaptive cruise and Euro NCAP active safety; his physics-first, control-theory approach worked for ACC but “pretty quickly became clear that wasn’t gonna work well all the way to the end, even back then.”
• The pivot moment, early 2014: Mobileye’s demonstrator — an Audi A8 or similar — drove itself off a single monocular camera, with Baglino recalling possible radar fusion as well. Elon came out of the demo declaring “We’re gonna have the car drive itself” — on the existing hardware. “That’s when the people-have-two-eyes, why can’t we drive the car with just vision — that’s when that sort of started.”
• The weekly meeting question thereafter: “When’s the car gonna drive itself from California to Boston?… Can it be next week?” — Baglino calls it the “amazing value of the reality distortion field.” Eventually: “I’m an energy nerd” — he decamped to Tesla Energy when the Powerwall 1 team was three engineers, leaving Autopilot to “a series of leaders that tried to work with Elon through the why-can’t-it-be-tomorrow reality.”
• On FSD, genuine humility: the team went “from heuristics and physics-based optimal control planning to nothing but nets, like pixels to actuators… Could I have predicted that? No. I just didn’t have the context at all.”

9. The calcified grid: why Heron Power exists

• The structural setup: from the ’80s through the 2010s, US load growth ran 1% or less — efficiency gains in lighting, air conditioning, and compute meant data centers’ share of end-use electricity barely rose from the ’90s to the 2010s. Talent and technology development migrated to growing markets: Japan, Korea, China, India.
• The incentive kicker: utilities earn a guaranteed rate of return on deployed CapEx, not on electricity sold. With no growth, suppliers shipping flat volumes got “fat and happy, charging more for that same unit over time, because that’s actually what the utility wants to do — spend more money on it.” His Tesla-era name for it: “towing a dinosaur… the supply base is almost like a boat anchor on the path toward electrification.”
• The demand side: electricity is only about a third of end-use energy today; a fully electric sustainable economy — the subject of “Master Plan Part 3,” which he worked on — means producing three times as much electricity. “Short answer, yes, you can”: the resources and technologies exist and it’s economical, but with electrification growing at 3-5% per year, “we need new suppliers.”

10. California rates: the shrinking denominator, now reversing

• His opening curveball: California retail rates are very high, but the average household bill is “middle of the pack” — mild climate, little air conditioning, low usage. The rate drivers: wildfire abatement and prior-fire liability (“probably the single individual adder relative to other places”), inflating component costs, a rural distribution footprint (serving northern Mendocino County as fully as downtown SF — “compare that to Delaware”), and aging infrastructure needing replacement.
• The controversial one: net energy metering compensated rooftop solar at retail price rather than the spot price the utility could pay in the market — and, in his fixed-cost framing, every rooftop-solar kWh leaves the denominator over which grid costs amortize, so per-kWh costs must rise. Rooftop solar is popular enough in California to move that math materially.
• The reversal: EVs, heat pumps, and data centers — “maybe even in homes, like Xfra [as heard] is this idea of data centers in homes” — have restored load growth, and Baglino says the equation’s numerator — total system costs — finally went up. “If you listen to Patty Poppe, CEO of PG&E… if we keep having this load growth and the regulatory structure doesn’t change, we’re actually gonna reduce rates in California.”

11. Heron Link: a 10,000-volt transistor smaller than a GPU

• The incumbent hardware Heron targets: switchgear — motorized “blade switches from Frankenstein” in vacuum or inert gas, taking hundreds of milliseconds or seconds to open, and among the longest-lead-time items in a data-center or factory build — plus passive steel-oil-and-copper transformers, sold into a balkanized market of 3,000 US utilities each with its own custom requirements.
• The physics: wide-bandgap materials (silicon carbide, GaN) block more voltage per micron with less parasitic capacitance, so they switch far faster at the same loss. A GPU has a “bajillion” transistors — “I don’t even know how many, trillions or whatever” — while “a power device has one… but it’s one highly engineered device.” Silicon carbide lets you make “a 10,000-volt transistor… actually smaller than the GPU.”
• The product consequence: you still need galvanic isolation from grid to computer, but doing it at hundreds of kilohertz instead of 60Hz — “at 60 times a second you can move a packet of energy through, versus 200,000 times a second” — makes the transformer 100x smaller volumetrically per unit of power. The power electronics on either side can natively interrupt current, eliminating the additional mechanical switch in this architecture, and there’s no flammable oil (he watched more than one transformer catch fire down the road in Bernal). It also collapses cascading conversion stages (300kV → 13/34kV → 480V → 208V) into one stage with multiple output voltages. His immediate focus is new solar, battery, factory, and data-center builds; older passive infrastructure can be replaced over time.
• The efficiency math, in his words: “You take a gigawatt of power and you convert it into 700 megawatts of heat. You spend 300 megawatts to get that heat into the atmosphere. That’s actually what a data center is” — with perhaps 8-10% lost in conversion en route to the chip. Heron halves grid-to-chip loss: “for somebody building a gigawatt data center, it means they can get 35 megawatts more useful compute out.”

12. Data centers as the grid’s best customer

• The core argument: a home with 200-amp (~40kW) service averages one or two kilowatts; a data center with gigawatt interconnection averages 800 megawatts. “They’re basically like aluminum smelters or steel mills — the customers that make the grid affordable for everybody else.” Empirically, states with the highest data-center penetration “overwhelmingly have had the lowest electricity rates and actually have had rates reduced.” Lukas presses — “So you actually think data centers will cause electricity rates to go down?” — “I do.”
• The stated conditions: it fails if utilities’ CapEx incentives lead them to overbuild, or if interconnection costs get amortized onto residential ratepayers. His fix: “the data center should just pay for the infrastructure that comes to them, because the return on investment of these data centers is so high, and that infrastructure actually isn’t that expensive compared to the GPUs.”
• The stability angle: training load requires on-site storage to absorb “training ripple,” and once it’s there, data centers can support peaks and stabilize frequency and voltage instead of disconnecting. Historically they’ve been a liability — rectifier flicker utilities had to clean up, and “just this week: three gigawatts of data centers all turned off at the same time” during a grid wobble.

13. “Down for nukes,” but geothermal may beat them

• No aversion to nuclear: glad Diablo Canyon was extended, wants a real US waste plan, and points to Asia’s learning curve — build costs per kilowatt “have progressively come down” with reps, “without any safety problems… it’s not like you’re trading safety for cost.” The US stopped building and lost the trend, but he’s “confident if we start building nuclear again, we can find ways to do it more affordably.”
• The LCOE ladder he lays out: 100% solar-plus-storage — which Tesla used to make islands in American Samoa fully renewable — still runs “basically 10 cents,” maybe seven or eight for an incredible desert asset like the Middle East. Fervo-style geothermal is “trying to get down to six or five”; fully depreciated existing nukes are “like two or three cents”; the just-completed Georgia facility “was, I think, more than 10.” His call: “geothermal is gonna give nuclear a run for its money in terms of affordable baseload 24-by-7 renewable power.”
• On whether China’s cheap nuclear is subsidized — his honest non-answer: “whether they directly subsidize the nuke or not, they’ve subsidized everything that went into the nuke in some way or the other” — concrete, steel, fancy metals.

14. From 70-watt GaN bricks to five-megawatt links — and the parking-lot battery

• The consumer analogy he reaches for, pointing at a laptop charger: a MacBook adapter is already a solid-state transformer “probably switching 800,000 times a second,” and a tiny 70W dual-output GaN plug contains a pinky-sized isolation transformer with probably ~5mm GaN dies switching a million times a second. “Basically what we’re doing at Heron is… doing that but for industrial electronics. Five million watts is the rating of our product instead of 70 watts. Same concept, though.”
• Closing on why charging slows: lithium-ion batteries are “a giant parking lot at a stadium” — the first spots are easy to find, the last ones hard, and a linear parking lot with much more surface area is “super expensive,” so you’re always trading the 2D/3D problem. Lukas’s reaction stands: “that’s a great analogy” — Baglino: “It’s the best analogy for lithium-ion batteries.”