The NextSilicon Valley

I've always been fascinated by the history of "Silicon Valley." It's so familiar now that it barely registers as a description of anything. To most people, it's just a place — known for ambition and disruption and enormous wealth. But the name is literal. It came from a material. And that material changed everything.

The term was coined in January 1971, by a journalist named Don Hoefler, writing for a trade magazine called Electronic News. The headline: "Silicon Valley U.S.A." He was writing about the semiconductor companies clustered around San Jose, California — companies that were doing something extraordinary with a humble element found in sand. Silicon. Cheap, abundant, and, it turned out, uniquely suited for making the transistors that would eventually power every computer, phone, and data center on earth.

What followed is the greatest technology origin story ever told. William Shockley set up his semiconductor lab in Mountain View in 1956. Eight of his engineers, fed up with his management style, quit to found Fairchild Semiconductor in 1957. Two of them — Robert Noyce and Gordon Moore — left to start Intel in 1968. From that single act of rebellion, you can trace a direct line to Apple, Google, Nvidia, and nearly every other company that defines modern life. Silicon didn't just name a valley. It built a civilization.

Now, roughly 70 years later, silicon is showing its age. It's not failing everywhere — but in the hardware that converts, manages, and delivers electricity, it has hit a wall. Push silicon too hard and electrons start leaking where they shouldn't. Heat builds. Energy is wasted.

The timing couldn't be worse — or better, depending on where you're standing. AI data centers are devouring electricity at a scale that's reshaping national power grids. Electric vehicles need smaller, lighter, more efficient charging systems. Solar farms need to waste as little energy as possible at every step. The world is asking silicon to do more than it physically can. And that opens a door for whatever comes next.

So the industry is replacing silicon in power applications. Not gradually, not reluctantly — but with real urgency, because the economics now demand it. The replacement material is called gallium nitride, or GaN. And a small company listed on the Nasdaq, founded in 2014, has spent the last decade quietly building the most advanced GaN power chips in the world. They've already won a design partnership with Nvidia. The revenue looks terrible right now. And I think this might be one of the most important semiconductor investments of the next decade.

It helps to understand, briefly, why silicon is struggling. The semiconductor material in a chip determines how much voltage it can handle before breaking down, how fast it can switch on and off, and how much heat it generates doing so. Silicon has a bandgap — a measure of these properties — of about 1.1 electron volts. For decades, that was more than enough. But the applications driving growth today operate at voltages and temperatures that push silicon past its comfortable limits.

At high voltages above roughly 600 volts, silicon transistors start leaking electrons. Heat builds faster than it can be dissipated. Switching speeds plateau. Engineers compensate with bigger heat sinks, more complex circuit designs, and heavier components — all of which add cost, weight, and energy loss.

Gallium nitride has a bandgap of about 3.4 electron volts — more than three times wider than silicon. That wider bandgap means it can handle higher voltages without breaking down, operate at higher temperatures without leaking, and switch on and off far faster, losing less energy as heat in the process. GaN power systems can be smaller, lighter, and more efficient than silicon equivalents. In fast chargers, this is why your new laptop brick is a fraction of the size of the one from five years ago.

Navitas Semiconductor is headquartered in Torrance, California, and trades on the Nasdaq under the ticker NVTS. It was founded in 2014 by a team of power semiconductor veterans, and it describes itself — accurately — as the only pure-play, next-generation power semiconductor company publicly traded anywhere in the world. Every other GaN player is a division inside a much larger company. Infineon, Texas Instruments, STMicroelectronics — they all have GaN programs, but it's one product line among dozens. For Navitas, this is everything.

What specifically caught my attention is not just the material — it's the integration. Navitas builds what they call GaNFast power ICs. The key word is integrated. Traditional GaN approaches stack a GaN power transistor next to a silicon controller chip, connecting them externally. Navitas integrates the power device, the driver, the control logic, the sensing, and the protection circuitry all on a single monolithic chip. Fewer components. Smaller footprint. Less energy lost between stages. It's the kind of architectural advantage that's hard for a larger company to copy quickly.

Then, in October 2025, the confirmation that changed the market's opinion of this company: Nvidia named Navitas as a power semiconductor partner for its next-generation 800-volt DC architecture for AI factory computing. That's not a press release. That's an engineering partnership. Navitas GaN and SiC chips are designed into the power delivery systems for the AI racks that hyperscalers are building right now.

Navitas designs power semiconductor chips. It is fabless — meaning it doesn't own its own factories. It designs the chips, then contracts manufacturing to partners. This is the same model as Qualcomm, Nvidia, and AMD. It keeps capital requirements lower and lets the company focus on what it's actually good at: the design and the IP.

Here is the honest part: the revenue chart is a mess. Navitas peaked at $83 million in 2024, then deliberately shrank to $46 million in 2025, and Q1 2026 came in at just $8.6 million. That looks catastrophic until you understand what's actually happening. The company made a deliberate decision to exit its low-margin consumer charging business in China — a market flooded with cheap competitors and thin margins — in order to redirect the entire organization toward AI data centers, grid infrastructure, and EVs. They are burning down the old business to build the new one. The mess is intentional.

The forward indicators tell a different story. Navitas closed 2024 with $450 million in customer design-wins — the engineering commitments that precede volume production orders — and a total customer pipeline that grew 92% to $2.4 billion. Design-wins in semiconductors are real commitments. They take 18 to 36 months to convert to revenue, but they convert. The revenue trough is now. The inflection is coming.

Let's be direct. Navitas is not profitable. It is not close to profitable. The GAAP operating loss in 2025 was $41.4 million on $45.9 million of revenue. Non-GAAP operating loss was a smaller but still real $12.1 million. The company has been burning cash to fund the pivot — R&D, restructuring, and the deliberate exit from legacy markets. This is a bet on future revenue, not a bet on today's earnings.

What the balance sheet does have is enough runway to survive the transition. Cash and equivalents were approximately $150 million at last report, following a $100 million capital raise in mid-2025. Q1 2026 showed sequential revenue growth of 18% — the first green shoots of the turn. Management is guiding for continued sequential improvement through 2026, with the more meaningful revenue inflection expected in 2027 as Nvidia's next-gen AI rack systems ramp into volume production.

The valuation is where it gets complicated. The stock has run hard — from under $4 in early 2025 to over $22 today, a move driven by the Nvidia partnership announcement and renewed investor interest in AI infrastructure plays. At current prices, you are paying a significant premium to current revenues on the expectation of a 2027 inflection. That is a faith-based multiple, which is exactly what early-stage deep-tech investing looks like.

Here is what I keep coming back to. Every major technological transition of the last century has required a parallel revolution in how electricity is handled. Electrification of homes and factories required new transformers and switchgear. The rise of computing required new power supplies. The mobile revolution required the miniaturized chargers and power management chips inside every phone. Now, three simultaneous megatrends — AI infrastructure, EV adoption, and renewable energy — are converging on the same bottleneck: power conversion efficiency.

AI data centers are already the single fastest-growing electricity consumer on the planet. Every percentage point of efficiency gained in the power delivery chain — from the grid, through the building infrastructure, to the server rack — translates directly into lower operating costs and fewer data centers needed to hit the same compute target. Nvidia's move to 800-volt DC architectures is specifically designed to reduce those losses. And Navitas makes the chips that make that architecture work.

I own this. A small position — sized for what it is, which is a high-conviction speculative bet on a pre-profit company in the early innings of a structural material transition. I am not betting on a good quarter. I am betting on a decade-long shift in how the world handles electricity, and on one of the few pure-play companies positioned at the center of it. The mess in the numbers right now is the price of admission.

Fifty years ago, a journalist named Don Hoefler wrote about a cluster of companies doing something extraordinary with silicon. They named a valley after it. I don't know what they'll name the next era. But I have a sense of the material it will be built from.