Last week we covered etching equipment. The week before, gas turbines. This week the bottleneck is about analog semiconductors.
When you think of chip shortages, your mind immediately goes to the leading edge. NVIDIA's Blackwell and Rubin GPUs, and the HBM memory stacked beside them. Those chips are genuinely scarce, and incredibly critical to the AI buildout.
But another set of chips, far less talked about but still important to the buildout, is quietly becoming a chokepoint of its own.
Prices are climbing. Lead times just stretched to six months. And the demand driving both keeps exploding.
The chokepoint is analog semiconductors.
The chips between the world and the computer
As the name suggests, analog semiconductors deal with analog signals, which are the continuous physical quantities of the real world. Voltage. Current. Temperature. Light. Sound.
A GPU lives in the clean digital world of ones and zeros. Analog chips, on the other hand, are the layer between that world and physical reality. Sensing signals, keeping time, and above all managing power. And in a data center, power is the job that dominates
The power a GPU consumes keeps climbing. Racks that drew tens of kilowatts a few years ago now draw well over a hundred, and megawatt-class designs are already on the roadmap. As the wattage grows, so does the difficulty of the journey that power has to make from the grid, through the building, into the rack, and finally down to the chip.
Source: generativevalue.com
Two closely related families of semiconductors run that journey together. Power semiconductors (which we covered in an earlier edition) are the switches that physically carry and convert the electricity.
Analog semiconductors are the control and sensing layer wrapped around them: the chips that drive those switches, hold their output steady, protect the board when something goes wrong, and measure what’s flowing at every point.
If power semis are the muscle, analog is the nervous system that coordinates it. You'll find different types of analog chips at every stage of the journey:
From the grid to the rack, electricity arrives as high-voltage AC and gets converted down in stages. Analog chips control and supervise those conversions, and they stand guard at the rack door. Hot-swap controllers let a server be plugged into a live rack without an arc or a surge, and electronic fuses cut a fault before it spreads.
On the server board, the rack's voltage has to be stepped down again and split across every component. This is the home turf of PMICs (power management ICs), the analog chips that regulate and distribute power on a board, along with sequencers that turn dozens of power rails on in precisely the right order, and the temperature sensors watching the board's thermals.
At the chip itself comes the hardest step. A modern GPU runs at under one volt while drawing more than a kilowatt, which means over a thousand amps flowing into a small chip. Here, analog control chips work together with power semiconductor stages right beside the GPU. The analog controllers manage dozens of parallel phases and hold the voltage extremely steady, while the power stages handle the actual high-current switching.
Every generation of AI hardware asks more of this layer. Higher rack power means more power rails and every new rail needs its own regulation, protection, and sensing.
And hyperscalers increasingly demand real-time telemetry on all of it, because at these power levels you don't run what you can't measure. So analog content per server is climbing steadily and PMICs especially, the workhorses of the board level, are the fastest-tightening slice of it.
Who makes all this? A short and familiar list:
Texas Instruments (TI)- the world's largest analog maker at roughly 18% share. The volume player, with a catalog of ~80,000 parts sold into everything (not just data centers).
Analog Devices (ADI)- #2 at about 13.5%. The precision house, with over 40% share in data converters and amplifiers, and the fastest-growing data center franchise in the group.
The specialists led by Monolithic Power Systems, estimated to hold around 70% of the voltage-regulator sockets on NVIDIA's newest GPU generation.
The top five hold just under 60% of a fragmented market.
The demand evidence
The tightening is showing up everywhere you'd expect. In the earnings, in the price lists, and in letters to customers:
ADI's data center business is compounding at AI speed: Fiscal Q2 revenue hit a record $3.62B, up 37% year over year company-wide. The data center slice grew more than 90%, and is now over three-quarters of the Communications segment.
Lead times are stretching across the group: On July 3, ADI told customers that supply is tightening and lead times on part of its portfolio now run up to six months. Distributors report lead times extending on select parts across TI, Microchip, ST, NXP, and Infineon too with PMIC quotes running 35–40 weeks, up from 21–26.
Prices are going up and sticking: ADI raised prices February 1, following TI's August hike, reported as its largest ever. Management says the pricing holds because substitution is effectively zero once a part is designed in.
Six-month lead times and 30% price hikes on forty-year-old technology. Something upstream is stuck.
ADI’s Revenue Growth from Q3 2024 - Q2 2026 (Source: Tessara)
The fab nobody builds
So the demand is there and the prices are moving. Why can't the suppliers just catch up?
Because of how and where analog chips are made. Analog doesn't need the newest manufacturing. It runs on mature process nodes, using a workhorse process called BCD (Bipolar-CMOS-DMOS) that puts precision circuits, logic, and power handling on a single die. It's the standard way to make a PMIC.
Most of this production still takes place on older 8-inch (200mm) wafers, which is the same wafer size the industry relied on heavily in early 2000s and 2010s. While it’s possible to move these mature processes onto larger 12-inch (300mm) wafers, which can produce more than twice as many chips per wafer, very few companies have invested in doing so. As a result, most analog capacity remains stuck on these older 8-inch production lines.
Who owns those fabs matters just as much. TI makes the vast majority of its chips in-house and for PMICs, ADI runs roughly half in its own fabs, half at external foundries. Much of the rest of the industry outsources.
Texas Instruments: $274B market cap, Fwd PE of 40 (8 Jul 2026)
So when capacity tightens, everyone without their own line queues at the same handful of mature-node foundries (UMC, Hua Hong, SMIC, etc) for the same 8-inch BCD wafers.
And that capacity is tightening from every direction at once:
Nobody builds new 8-inch fabs: A leading-edge fab earns multiples more per wafer than a mature-node line, so that's where the capex goes. Equipment makers barely produce new 200mm tools anymore, so expansion means hunting for secondhand machines. And even a committed new fab takes 3–5 years to build, plus months of customer qualification.
Existing capacity is shrinking: Samsung plans to shut its S7 8-inch fab, per TrendForce, just as demand inflects. Global 8-inch utilization is projected to climb from 75–80% in 2025 to 85–90% in 2026, levels where pricing power sits with whoever owns the line.
AI isn't the only demand: The same mature-node capacity feeds the EV industry. An electric car carries roughly twice the chip content of a gas one, much of it analog, in battery management, charging, and inverters, plus industrial automation and renewables. AI demand is arriving on top of structurally growing buyers, not instead of them.
The relief valve is geopolitical: Roughly 70% of new mature-node capacity coming in 2026 is at Chinese foundries, by industry estimates. Between tariffs and sourcing restrictions, that concentrates the risk more than it relieves the West.
Mature Node Wafers constraint is tight with no relief visible (Source: Tessara)
One company built its way around all of this.
Texas Instruments (TI) spent fifteen years moving analog onto 300mm wafers, over twice the chips per wafer, roughly 40% lower die cost and all seven fabs in its $60B US buildout are 300mm, while its remaining 200mm lines get closed or sold.
That puts TI in an enviable position. While peers ration allocation and absorb rising foundry prices, it owns cheaper capacity it fully controls, with headroom to take share. The constraint is 8-inch, and TI is the one major player leaving 8-inch.
Who pays, who captures
Pays rent: the hyperscalers and server OEMs whose racks need the content plus everyone else competing for the same wafers like auto OEMs and industrial buyers.
Captures rent: the owners of mature-node capacity. TI structurally, through 300mm cost and control. The 8-inch foundries like UMC, Hua Hong cyclically, as utilization climbs. ADI where design-in makes pricing stick.
We Asked Tessara
We asked Tessara what’s one thing people are missing about Analog semiconductors. Here’s what she said:
ADI extended certain analog IC lead times to six months; a broader read puts power and analog lead times at 26-30 weeks with prices up about 30%.
Board power conversion reads tight at 73/100 and tightening, driven by AI-server VRM demand from MPWR and server PSU revenue running well above trend.
Japan MLCC export value runs well above trend while Samsung Electro-Mechanics capex is pulling back: throughput hot, supply response quiet.
Tessara’s research agent on analog semiconductors
The Week Ahead
Earnings Tab on Tessara
Next Wednesday, July 15
ASML (ASML) — EUV Lithography.
Next Thursday, July 16
TSM (TSMC) — Foundry.
ABB.SW (ABB) — Power and Automation.
Read our post and pre-call briefs here and stay prepared.
See where the AI buildout goes next
This issue, we named 8-inch fab capacity as the chokepoint and Texas Instruments as the one player built around it.
In Tessara terminal, you can track the tightening across analog and power semis, and see which of 400+ public names are most exposed to the squeeze
“From extensive sector and company deep dives to thoughtful market updates, it has become a go-to resource for our investment team” - Portfolio Manager, Titan Global Capital Management
See you next week,
Teng & Arvind
This article is for informational and research purposes only. It is not financial advice, investment advice, or a recommendation to buy or sell any security. Tessara Research does not publish price targets. The views expressed here reflect our analysis at the time of publication and may change as new evidence arrives. Readers should do their own research and consult a qualified financial adviser before making investment decisions.