AI Intelligence // signal over noise
← back to feed
TSPA Semiconductor 7/10 signal

Why TLVR Will Not Kill MLCCs—and May Accelerate the Shift to VPD

infrastructure
Summary
A recent stock selloff in passive-component makers has been attributed to the idea that Trans-Inductor Voltage Regulators (TLVR) in new NVIDIA AI GPUs will eliminate the need for Multilayer Ceramic Capacitors (MLCCs). This analysis argues the opposite: rising AI chip power demands (from hundreds of watts to kilowatts) will necessitate a more complex, distributed power delivery network, changing but not reducing the overall need for advanced capacitors and accelerating the shift to Vertical Power Delivery (VPD).
Context
Passive-component stocks, including Murata Manufacturing, Samsung Electro-Mechanics, Taiyo Yuden, Yageo, and Walsin Technology, recently experienced a sharp correction. The market narrative quickly formed that NVIDIA's adoption of TLVR technology would improve power supply transient response so effectively that large numbers of high-end MLCCs could be removed from AI server designs, collapsing demand. This article directly challenges that simplistic, component-level view. It reframes the issue within the broader context of rapidly escalating AI accelerator power consumption, where total power is rising to kilowatts while core voltages remain low (0.6-1.0V), pushing currents into the kiloampere range and making power delivery a critical system-level challenge.
Details

The Market Narrative vs. System-Level Reality:

A sharp decline in the stock prices of passive-component suppliers (Murata, Samsung Electro-Mechanics, Taiyo Yuden, Yageo, Walsin Technology, Holy Stone, Fenghua) was widely blamed on the adoption of TLVR in AI GPUs. The prevailing argument is that TLVR's improved transient response allows for the removal of many board-level MLCCs, thus reducing demand. The author contends this overlooks the simultaneous and more powerful trend of increasing AI chip power requirements.

Escalating Power Delivery Challenges:

  • Power & Current Growth: AI GPUs are moving from consuming hundreds of watts to several kilowatts.
  • Low Voltage Operation: Core operating voltages remain in the 0.6V to 1.0V range.
  • Resulting Effects: This combination drives current into the kiloampere range, which severely worsens issues like parasitic resistance, parasitic inductance, voltage droop, and load transients across the entire power path.

The Shift to Distributed Decoupling:

The author argues the core architectural shift is not about TLVR replacing MLCCs, but about the power delivery network (PDN) evolving from a centralized, board-level decoupling model to a distributed one. In this new model, capacitance is strategically placed across multiple locations:

  • The main board
  • The package underside
  • The interposer
  • The voltage regulators
  • The backside of the chip itself

Redefining Technology Roles:

  • TLVR: Positioned as an enhancement to the existing lateral power-delivery architecture, not a revolutionary replacement for capacitors. It may reduce some board-level output capacitance but does not eliminate the need for decoupling capacitors system-wide.
  • MLCCs: The demand for MLCCs will not disappear. Instead, it will shift from a simple component count to a focus on structural upgrades and performance metrics like higher capacitance, specific voltage ratings, low ESL (Equivalent Series Inductance), low ESR (Equivalent Series Resistance), wider bandwidth, and improved thermal endurance and reliability.
  • Future Architectures: The next significant architectural evolution in power delivery includes in-package Integrated Voltage Regulators (IVR), embedded Deep Trench Capacitors (eDTC), backside power delivery, and ultimately Vertical Power Delivery (VPD).

Stock Market vs. Technical Demand:

The analysis separates stock market performance from fundamental demand. The stock correction is presented as a reflection of changing growth expectations or capital allocation, not a direct indicator of a collapse in MLCC consumption. The affected companies have diverse product portfolios (automotive, industrial, consumer electronics) and different exposures to various capacitor types, making TLVR an overly simplistic explanation for a market-wide selloff.

What's new
The analysis reframes the TLVR vs. MLCC debate from a simple component substitution narrative to a broader, system-level architectural evolution. It posits that TLVR is a transitional enhancement to lateral power delivery, not a disruptive replacement for capacitors. The core new idea is that escalating power demands in AI chips will force a move to distributed decoupling architectures, ultimately accelerating the transition to more advanced solutions like Vertical Power Delivery (VPD), which will change, not eliminate, the role of high-performance MLCCs.
Limitations
This is an analysis and opinion piece from a research firm, not a presentation of new experimental data. It argues against a market narrative without providing its own quantitative forecasts. The article also explicitly states that it is an incomplete analysis due to space limitations and promotes the author's paid subscription services for more in-depth coverage.
The take

This is a sharp, necessary correction to a simplistic market narrative. The idea that a single technology (TLVR) would simply erase demand for a foundational component like MLCCs ignores the history of high-performance computing, where efficiency gains are immediately spent on pushing performance. As AI chips become power-hungry beasts demanding kilowatts at sub-1V levels, the entire power delivery network becomes a first-order design problem. The real story isn't component count reduction; it's an architectural war against parasitic inductance and voltage droop. This means pushing capacitance closer to the silicon—on-package, on-interposer, and eventually on-chip. TLVR is a step in that direction, but it's part of a larger trend that will ultimately drive demand for more specialized, higher-performance, and strategically placed capacitors, not fewer.

Read the source →

Don't read this site daily. Get it in your inbox.

The daily brief and Sunday deep dive — distilled, scored, and opinionated. For builders only.