Substack长文重述onsemi收购Synaptics与Physical AI叙事，但核心交易已在6月25日公开

My two main focal points for my investing are AI infrastructure and Physical AI.

So if I find a company that has exposure to both, I get excited.

Today we are going to dig deep and unpack my latest investment in this space.

*This report is for educational and informational purposes only. It reflects my personal research process, opinions, and interpretation of publicly available information, including company filings, investor presentations, earnings calls, management commentary, press releases, industry reports, and third-party sources believed to be reliable. It should never be treated as individualized financial advice, investment advice, legal advice, tax advice, or a recommendation to buy, sell, hold, short, or avoid any security.*

Today we are talking about . Onsemi.


Here is my TLDR thesis:

Onsemi becomes a full-stack Physical AI systems company. The deal with Synaptics positions onsemi at the intersection of Power, Sense, Connected Compute and Control, which are the four pillars of Physical AI that enable machines to sense, decide, act and adapt in the physical world. onsemi already dominates power (silicon carbide, GaN) and sensing (image sensors, automotive silicon). Synaptics adds a proven Edge AI compute platform (Astra) combining purpose-built AI processors and NPUs with a wireless connectivity portfolio spanning Wi-Fi, Bluetooth and GPS, plus an open-source software stack. That fills onsemi’s biggest gap: compute and connectivity.

The acquisition expands onsemi’s total addressable market by $30 billion to $243 billion by 2030, and management sees a ~$100 billion AI-specific opportunity by 2030 across AI infrastructure and Physical AI. My dual thesis (AI infra + physical AI) maps directly where onsemi’s power devices feed AI data centers, while the combined portfolio targets robots, autonomous vehicles, and industrial automation.

Higher-value system solutions with embedded IP and software support improved mix, margin expansion and durable growth. Selling systems instead of components means more dollar content per platform. The deal is expected to be accretive to non-GAAP EPS within 18 months of closing, with $200 million in annual synergies. Notably, Synaptics’ core IoT revenue grew 53% year-over-year. I’m buying accelerating edge-AI demand.

Now here is the full report!

Most people know ON through the AI infrastructure lens. Especially with all the discussion around 800V architectures that will be needed for massively powerful AI racks. In this lens we are talking about power conversion, power delivery, and energy efficiency.

But lesser known is their Physical AI angle. It has now been brought into the spotlight with their acquisition of Synaptics. We’ll unpack that more in detail.

Before we dive in let’s revisit some of the ground work really quickly.

One way I see Physical AI is the move from AI inside software into machines that operate in the real world.

A Physical AI system has to interact with the world. They have to:

-

Sense what is around them

-

Process information locally

-

Make a decision

-

Move something physically

-

Control motors, power, heat, signals, and safety

-

Adapt as conditions change

And a physical machine has a body that needs power, sensors, control electronics and local compute & connectivity.


ON already had three of those layers.

-

**Power** - the chips that convert, protect, and deliver electricity

-

**Sense** - image sensors, ultrasonic sensors, position sensing, and other related sensing products

-

**Control** - drivers, switches, analog/mixed signal ICs, and other control products

Now with their acquisition of Synaptics, they add a fourth layer.


**Connected compute**.

This is a combination of two different things.

-

local compute (processing happens inside the device, close to the action).

-

connectivity (the device can communicate).

Synaptics provides both.

So now ON has a fuller stack. And is very relevant to Physical AI.

## Digging Deeper

Alright, now that we have the bigger picture, let’s unpack all that ON does.

If we were to define this company it would go something like this: ON is a semiconductor company built around electricity, sensing, and physical control. It sells chips that go inside cars, factories, energy systems, robots, data centers, and industrial machines. These chips help physical systems convert power, protect circuits, sense the environment, and control real-world action.

Let’s break it into three buckets.

1 - **Power Solutions Group**


This is the largest bucket. Q1 2026, $737m in revenue. Up 14% YoY.

This is all about controlling electricity. Every physical system has an energy source like a battery or a wall outlet or a grid connection. Whatever it may be. But the system usually cannot use that raw power directly. It has to be converted, switched, stepped down, stepped up, protected, and delivered at the right voltage and current.

That is what their power chips do.

In an EV, power have to move from the battery into the motor in a controlled way.


In an AI rack, power have to move from the facility into power supplies, rack systems, boards, and eventually the processor.

In a robot, power has to move from the battery to motors, joints, sensors, cameras and control electronics.


ON sells into this through:

-

SiC (Silicon carbide. Material used for high-voltage, high-heat, high-efficiency power conversion)

-

IGBTs (Mature high-power switches used in inverters, often when customers want a cost-effective balance of performance and price)

-

MOSFETs (Fast electronic switches that turn power on and off inside power supplies, converters, chargers, motors, and racks)

-

GaN (Gallium Nitride. Used for very fact, compact, high frequency power conversion.)

-

JFETs (Junction field-effect transistors. Power switches that can improve efficiency, especially when built with SiC for demanding power supplies).

-

Power Modules (Packaged group of power chips placed together so customers can handle high voltage/current more easily.)

-

Smart Power Stages (Integrated power chips that combine switching, control, and monitoring close to the processor or load).

-

Protection (Chips that prevent damage from overloads, short circuits, overheating, voltage spikes. or unsafe power events).

-

Drivers (Chips that tell power switches when and how to turn on or off).

2 - **Analog and Mixed-Signal Group**

Q1 2026, $540m of revenue. Down 5% YoY.

This is the layer that connects the physical world to electronic control.

Digital chips speak in 1s and 0s.

But the physical world is continuous. You have voltages that change gradually, temperature that rises and falls, battery voltage drifts etc.

Analog chips help measure, condition, amplify, regulate, and control those real-world signals.

Mixed-signal chips combine analog and digital functions. So they can read a physical signal, clean it up, translate it, and help the system make a useable decision.

This bucket is especially important for Physical AI because real machines run on thousands of small feedback loops.

-

How much current is this motor pulling?

-

Where is this joint?

-

Should this switch turn on?

This is all analog and mixed-signal territory.

This is where **Treo** comes in. ON describes Treo as an analog and mixed-signal platform. Grew more than 2.5x sequentially in Q1. You can learn more about it here:

[https://www.onsemi.com/solutions/technology/treo-platform](https://www.onsemi.com/solutions/technology/treo-platform)

We will dig into Treo more later in the report as well.

When I first saw that this was business segment was down 5% YoY I didn’t love that. So upon digging more, AMG was down because the old analog/mixed-signal base is still digesting the downturn and ON is cutting non-core revenue, while the new higher-value Treo / AI / zonal / automation products are only beginning to show up. So this is definitely something to watch.

**3 - Intelligent Sensing Group**

Q1 2026, $236m in revenue. Up 1% YoY.

This is the awareness layer. Machines need to detect what is happening around them and inside them. ON’s sensing business includes image sensors and related sensing tech. They talk about industrial autonomy, automotive sensing, and robotics including high-res image sensors, depth, and SWIR.

A car needs to detect other vehicles, or pedestrians, lanes.

A robot needs depth perception and navigation.

An industrial system needs to detect position, presence, motion, and safety conditions.

Again, not much growth YoY. Its still tied to slower-recovering automotive and industrial sensing cycles, and the newer robotics / autonomy wins are design-win momentum rather than current rev.

Now let’s unpack power, control, and sensing a bit more one at a time.

## Power As The Foundation of AI Infra

The main reason that ON is showing up in the AI conversation to begin with is that data centers are increasingly becoming power constrained systems.

This is the flow:

-

Grid power comes to the DC

-

That power gets converted and conditioned.

-

Backup power and battery systems sit around it.

-

Power supplies feed the rack.

-

The rack distributes power to the boards.

-

The board steps power down again.

-

The chip finally receives very low-voltage, very high-current power.

Every step along the way creates losses and those losses become heat.

So the problem is that AI racks are demanding more power, but the data center has limited space, cooling, grid access, and tolerance for wasted energy.

This is where power semis come in.

These are the devices that switch, convert, regulate, and protect electricity. They determine how efficiently power moves through the system.

ON sells into that chain.


Management describes ON’s AI data-center opportunity as running across the ‘power tree’ from high voltage conversion to intelligent power stages, protection, control, and system-level integration from the grid to the processor. In Q1, data center rev grew more than 30% QoQ and it should double year over year. This is the inflection point.

Management uses the phrase ‘from the wall to the core’. Meaning they have power tech from data-center power input all the way down toward the processor. Products across many power-conversion layers.

And as racks get denser, power delivery gets harder.

If power delivery gets harder, better power devices become more valuable.

And if these devices reduce losses, save space, lower heat, and improve reliability, customers can justify paying for them.

This is revenue today. But there is also much deeper upside in future rack architectures. A 120kw rack can represent around $9500 of ON content. In an 800V or high-voltage rack, ON content could reach about $115,000.

Massive jump.

On timing, 800V DC starts moving into production with next-gen NVIDIA rack-scale system in 2027. 2028+ is broader adoption depending on hyperscaler deployments.

## EVs

ON has been working on high-voltage power problems for years through EVs. EVs and racks have a similar kind of physics problem.

How do you move a lot of electricity with less heat, less waste, less space, and higher reliability?

In an EV the battery stores energy. But it can’t simply send raw electricity straight into the motor. It needs controlled power. The car has to accelerate, slow down, manage the heat, protect the battery.

The main bridge between the two is the traction inverter. This takes DC power from the battery and converts it into controlled AC power for the motor.

Inside that inverter are power semiconductors. These chips switch electrictiy on and off very quickly. Which controls how much power reaches the motor and how efficiently the vehicle uses energy.

Many EVs have used 400V systems. More advanced EV platforms are moving to 800V or 900V. For the same amount of power, higher voltage allows lower current. Which usually means less heat, lower loss, thinners cables, better packaging, longer range and better efficiency.

This is all very attractive for EVs.

ON said that Chinese OEMs are leading the move toward 900V architectures and ON is already in production on next-gen platforms that support flash charging and higher efficiency. They have about 55% of silicon carbide share of new EV models shown at the 2026 Biejing Auto show.

That is great proof. It is already in production. It is showing up in high profile EV platforms, especially in China, where a lot of the high-voltage EV architecture work is moving fastest.

So when I think about ON’s EV power position, I think of it like this.

SiC is the premium high-voltage efficiency technology. It is useful when heat, voltage, charging speed, and range are the priority.

IGBTs are the more mature and cost-effective power switch. They can still be the better answer for lower-cost EV platforms where the customer wants a balance between performance and price.

That combination is important, and ON has both.

So if a high-end EV platform wants SiC, ON can play there. If a cost-optimized EV platform wants IGBTs, ON can play there too. Management even called out a new IGBT-based traction inverter win with a North American OEM.

And this is where the EV section connects back to AI infrastructure.

ON has already spent years solving high-voltage power problems inside vehicles.

Now AI data centers are moving into a similar type of high-voltage power problem.

Different end markets, but similar physics.

EVs gave ON credibility in high-voltage power. AI racks may give ON a much larger and faster-growing second market for that same skill set.

## The Control + Sensing Layer / Treo

Control is about making the system behave correctly.

Like I said before, a physical machine has thousands of tiny decisions happening constantly. Motors turn on and off. Switches open and close. Sensors send readings. Actuators move. Signals get routed. Power has to be managed safely.

This is all the control layer.

Let’s unpack Treo.

Treo is ON’s newer analog/mixed-signal platform.


Think of it as the platform ON uses to build smarter control chips for cars, industrial systems, automation, robotics, and AI-related devices.

This is especially useful in software-defined vehicles.

Cars used to have a lot of separate electronic boxes spread across the vehicle. Each box controlled one function. Doors, lights, seats, mirrors, cameras, parking systems, and so on.

Newer architectures are moving toward zonal design.

So instead of a separate box for every function, the car gets organized by physical zones. Front-left. Front-right. Rear-left. Rear-right.

Each zone handles local sensors, power distribution, communication, and control. Then those zones connect back to a more centralized compute system.

That creates new semiconductor content.

This is where ON says Treo is showing up. Management specifically mentioned production shipments of Treo-based and 10BASE-T1S Ethernet solutions for a leading North American customer’s next-generation zonal architecture, with the platform integrating more than 30 Treo devices.


I won’t spend much time on this, but just know that if zonal architecture is the new vehicle layout, 10BASE-T1S is one of the communication layers that helps the zones talk.

ON is selling into that shift.

That’s great and all, but we’re here for Physical AI.

Well, good news!

Robots have a similar control problem.

A robot’s main AI system might decide what task to perform. But the body still needs lower-level electronics to execute that task accurately. That is where driver ICs, SmartFETs, position sensing, power management, and mixed-signal control become useful.

ON specifically called out inductive position sensing for humanoids and advanced automation in Q1.

Position sensing is very simple conceptually. The machine needs to know where something is. A robot that can’t measure position accurately can’t move safely or precisely.

I want to separate something quickly because sensing and control are connected, but they are not the same thing.

Sensing is the input and control is the response.

Position sensing is one kind of sensing that helps control movement. But ON’s sensing portfolio is broader than that. They have a whole awareness layer.

Image sensors help machines see. Ultrasonic sensors help detect nearby objects. Depth sensing helps machines understand distance. SWIR can help in certain conditions where normal cameras may struggle.

And there is already some traction behind the Physical AI evidence here.

In Q1, management said ON won a design with a leading global robotics platform where its high-resolution image sensor and indirect time-of-flight technology were selected for depth perception and navigation in autonomous systems. They also talked about a sensing roadmap that spans high-resolution imaging, depth, and SWIR.

Treo revenue increased more than 2.5x sequentially in Q1, and management said Treo products can carry 60%-70% gross margins.

That is a big deal because ON’s overall gross margin is still recovering.

If more revenue shifts toward higher-value platforms like Treo, they can improve profitability even without needing every end market to boom at once.

So to me, this is the more lowkey part of the ON story.

The market knows the power story.

It is starting to understand the AI data-center story.

But the control and sensing layer is what makes ON more relevant to intelligent physical systems.

And this is what ON already had before Synaptics.

Synaptics adds the next missing piece: connected compute and tactile interface.

## Synaptics

Now we get to the acquisition.


Synaptics gives ON connected compute.

That phrase means two things.

First is **local compute**. Processing happens inside the device, close to where the data is created.

Second is **connectivity**. The device can communicate through Wi-Fi, Bluetooth, Thread, GPS/GNSS, and other wireless links.

Synaptics provides both.

Astra is Synaptics’ edge AI processor family. Think of it as a low-power local brain for intelligent edge devices. These chips are designed to run AI inference close to the device instead of sending every decision back to the cloud or a central processor.

Inference just means using an AI model after it has already been trained. So the device can recognize something, classify something, detect motion, process a sensor input, or make a local decision.


This slide from the acqusition deck describes Astra as a family of low-power edge AI processors, AI-native MPUs, smart MCUs, and a full-stack AI software platform. It also highlights Synaptics’ collaboration with Google on real-world Physical AI and Edge AI product development.

The Google Coral angle is important because edge AI needs more than silicon. Customers need tools that help them move models onto the device. Synaptics has integrated Google Coral NPU technology into parts of the Astra roadmap, which gives the platform a stronger software and developer ecosystem.

Now in regards to connectivity, Synaptics talked about an Astra-enabled connected MCU that integrates Wi-Fi 7, Bluetooth 6, and edge AI compute in a single monolithic SoC. That means more compute and wireless capability can sit inside one tighter chip package. For edge devices, this can reduce board space, power consumption, system cost, and engineering complexity.

A really strong Physical AI hook is tactile sensing.

Synaptics has decades of touch and interface experience. In robotics, that can become tactile sensing. A robot hand needs to understand pressure, slip, contact, proximity, and grip.


Synaptics said customers are adopting its AI-enabled touch controllers for tactile sensing, and management described the technology as detecting force, slip, and proximity so robots can respond in real time.

This is where the old touch expertise turns into a new robotics opportunity.

Synaptics said its robotics pipeline had grown to more than 35 customers globally. On the M&A call, Rahul Patel said that number had already increased, and he framed Synaptics’ humanoid content at roughly $30-$60 per platform before adding ON’s content.

The financial evidence shows why ON wanted this asset.

In Q3 FY26, Synaptics revenue was $294.2M, up 10% YoY. Core IoT grew 31% YoY, and non-GAAP gross margin was 53.6%.

Core IoT is the key piece. That bucket went from $178M in FY24 to $272M in FY25, and management expects $385M+ in FY26. It also moved from 19% of revenue in FY24 to roughly 32% in FY26E.

So Synaptics brings ON three things:

**Astra** for local edge AI compute.

**Wireless connectivity** for connected devices and machines.

**Tactile sensing / interface** for robots, wearables, smart devices, and human-machine interaction.

That is the reason for the acquisition.

Together, the combined company can address more semiconductor content inside intelligent physical systems.

## Combined Financial Profile

ON’s deal deck shows a CY26E pro forma revenue base of $7.8B: $6.5B from ON and $1.3B from Synaptics.


The margin mix improves because Synaptics carries a 54% gross margin profile versus ON at 40%. Pro forma, including Synaptics stock-based compensation and expected synergies, the combined company is shown at 42% gross margin and $1.7B of operating profit, or a 22% operating margin.

The key synergy number is $200M within 18 months after close. Management said 85%-90% of that should come from OpEx, mostly SG&A, with the balance from COGS. So the underwritten financial benefit is primarily cost reduction. The larger upside would come from revenue synerg with ON pushing Synaptics’ edge AI, connectivity, and tactile sensing through its auto and industrial customer base.

The balance sheet stays clean because the deal is all stock. The deck shows pro forma gross debt of $5.4B, cash of $4.2B, net debt of $1.2B, and net debt / LTM adjusted EBITDA of 0.6x. ON gives up equity, but keeps balance-sheet flexibility and keeps its capital return policy alive during the pendency period.

So the financial thesis is that ON gets more scale, a higher-margin asset, $200M of cost synergies, low leverage, and expected non-GAAP EPS accretion within 18 months after close. The real upside comes if Synaptics helps ON sell more complete semiconductor systems into automotive, industrial, robotics, wearables, edge AI, and physical AI machines.

## Competition

ON competes in several different markets, so the competitive set depends on the product.

In power and SiC, the main competitors include Infineon, STMicro, Wolfspeed, Rohm, Mitsubishi, Bosch, BYD Semiconductor, and emerging Chinese SiC suppliers.

In analog and power management, the peer group includes Texas Instruments, Analog Devices, Monolithic Power Systems, Renesas, Infineon, and STMicro.

In AI data-center power, ON runs into companies like Infineon, TI, ADI, MPS, Renesas, STMicro, Navitas, Power Integrations, and others depending on the voltage level and architecture.

In edge AI, MCUs, processors, and connectivity, Synaptics competes with NXP, STMicro, Infineon, Renesas, Microchip, Qualcomm, Broadcom, MediaTek, and Ambarella-type edge AI players depending on the socket.

This is a crowded market for sure but ON’s advantage is breadth.

It can sell high-voltage power, low-voltage power, sensing, analog, mixed-signal control, and now connected compute.

## Risks

1- ON is still tied to automotive and industrial. Those markets have been stabilizing, but a full replenishment cycle has yet to show up. If auto and industrial stay soft, the margin recovery takes longer.

2- SiC is attractive, but competition is intense. China is adding capacity, and substrate supply can pressure parts of the value chain. ON’s argument is that the moat sits in device design, packaging, reliability, qualification, supply assurance, and system-level execution. That has to keep proving itself.

3- The AI data-center power business already has revenue today, but the big content step-up depends on high-voltage architectures scaling. NVIDIA has pointed to 800V DC infrastructure starting around 2027, with broader deployment after that, but the rollout will be phased by customer, facility, rack generation, and safety architecture.

4- Robotics engagement is real. Revenue scale is still early. Design wins, samples, and pilots are different from high-volume production. The humanoid market can create large upside, but it can also take longer than the market wants.

5- The transaction needs Synaptics shareholder approval, regulatory approvals, and other closing conditions. ON also expects China regulatory review.

6- The market already liked ON’s AI data-center power story. Synaptics makes the story broader. Broader can mean more upside, but it also increases execution burden.

## Valuation Framework

At the time I am writing this framework, ON is at $91.22.

Current market cap of $35.5b.

CY26E combined revenue: $7.8B
Combined gross margin: 42%
Combined operating profit: $1.7B
Annual synergies: $200M
Pro forma net debt: $1.2B

So the market is valuing the current ON around a few different questions:

1- How fast does the core business recover?

2- How large does AI data-center power become?

3- How much of the 800V rack opportunity converts into revenue?

4- How much margin upside comes from utilization, Fab Right, Treo, and Synaptics?

5- How much revenue synergy comes from the combined platform?

I want to try to figure out what the upside could really look like here.

First, AI data-center power can become bigger than expected.

ON said AI data-center revenue was around $250M last year and should double in 2026.

That would put 2026 AI data-center revenue around $500M.

If this business grows 40% per year after that, it can approach $1B by 2028.

That is before assuming any major 800V step-change. ON’s deck shows content moving from roughly $15K per current 120kW rack to about $115K per 600kW-1MW high-voltage rack. Even partial adoption can move the model.

If ON captures only a few thousand high-voltage racks over time, the revenue contribution can reach hundreds of millions.

Second, Treo can lift the mix.

Treo revenue grew more than 2.5x sequentially in Q1, and management has said Treo products can carry 60%-70% gross margins. The whole company is still below 40% gross margin today.

That means every incremental dollar of higher-margin Treo revenue helps the model.

Third, Synaptics can scale through ON’s channel.

Synaptics Core IoT is expected to reach $385M+ in FY26, up from $178M in FY24.

If ON can use its automotive and industrial reach to accelerate Astra, connectivity, tactile sensing, and HMI, Core IoT can become a more meaningful growth engine by 2028.

A simple base case would be Synaptics Core IoT moving toward $600M-$700M by 2028.

A bull case would be $800M+, driven by Astra ramps, wireless attach, semi-custom programs, and robotics design conversion.

Fourth, robotics is the free option.

The humanoid revenue curve could be slow at first, then very nonlinear.

At $100-$250 of combined semiconductor content per humanoid, 1 million units equals $100M-$250M of revenue. At 5 million units, it becomes $500M-$1.25B.

That is far from today’s revenue base, but it explains why ON wanted this optionality.

### 12-month framework

I am using three scenarios.

Bear case: $70-$80

This assumes auto and industrial stay sluggish, the market questions the Synaptics deal, 800V timing gets pushed further out, and ON trades closer to a cyclical power semi multiple.

In this scenario, I would use roughly $3.20-$3.50 of forward EPS and a 22x-23x multiple.

Base case: $115-$120

This assumes the recovery continues, AI data-center revenue doubles in 2026, gross margin starts moving into the low 40s, and the Synaptics deal moves toward close without major issues.

In this scenario, I would use roughly $4.30-$4.40 of forward EPS and a 26x-27x multiple.

Bull case: $130-$150

This assumes the market starts underwriting ON as more than auto/industrial power. AI data-center power becomes a visible growth engine, 800V confidence improves, Treo mix keeps building, and the market gives ON some Physical AI platform credit.

In this scenario, I would use roughly $4.75-$5.00 of forward EPS and a 28x-30x multiple.

### 24-month framework

This is where the Synaptics deal becomes more important.

Bear case: $75-$90

This assumes the deal closes, but integration is slow, AI data-center growth decelerates, 800V adoption is pushed out, and margins stay in the low 40s.

That would imply combined revenue around $8B-$8.5B, operating margin in the low 20s, and EPS around $3.50-$4.00. At 21x-23x, the stock stays range-bound.

Base case: $150-$165

This assumes the combined company reaches roughly $9.5B-$10B of revenue by 2028.

That requires ON’s base business to recover, AI data-center power to keep compounding, Synaptics to grow Core IoT, and the $200M synergy plan to land.

With gross margin in the mid-40s and operating margin around 28%-30%, EPS can move toward $5.50-$6.00. At a 27x-28x multiple, that supports a stock in the $150-$165 range.

Bull case: $190-$230

This is the version where the market starts valuing ON as a true AI power + Physical AI semiconductor platform.

Revenue moves toward $10.5B-$11.5B. AI data-center power approaches or passes $1B. 800V racks begin contributing. Treo becomes a visible margin driver. Synaptics Core IoT scales faster through ON’s channel. Robotics and humanoid content starts showing up in backlog or early production revenue.

If operating margin moves toward the low-to-mid 30s, EPS can reach $6.75-$7.50. At 28x-32x, the stock can reach the $190-$230 range.

At around $90, I think the risk/reward is attractive because the market is still debating the acquisition while the core recovery is already forming.

The stock has three ways to work:

1- Cyclical recovery.

2- AI data-center power growth.

3- Physical AI platform expansion.

The 12-month setup depends mostly on recovery and AI power.

The 24-month setup depends on whether Synaptics turns ON into a more complete semiconductor platform for intelligent physical systems.

I am paying for a recovering power semiconductor company and getting a long-duration Physical AI option layered on top.

## *Financial Disclosure and Disclaimer*

*At the time of publication, I own shares of onsemi ($ON) in my Physical AI portfolio. My position reflects my personal view that ON gives me exposure to the semiconductor layer of Physical AI, including power, sensing, control, AI data-center power, and connected edge compute through the pending Synaptics acquisition.*

*I may add to, trim, sell, hedge, or change my view on $ON at any time based on valuation, earnings results, guidance, price action, portfolio construction, risk management, deal developments, customer traction, margin trends, AI data-center demand, 800V architecture adoption, Synaptics integration updates, or changes in the broader semiconductor cycle.*

*This report is for educational and informational purposes only. It reflects my personal research process, opinions, and interpretation of publicly available information, including company filings, investor presentations, earnings calls, management commentary, press releases, industry reports, and third-party sources believed to be reliable. It should never be treated as individualized financial advice, investment advice, legal advice, tax advice, or a recommendation to buy, sell, hold, short, or avoid any security.*

*The valuation framework and price scenarios discussed in this report are my own underwriting ranges, based on the information available at the time of writing. They are scenario frameworks rather than formal price targets. These ranges can move higher or lower as new information becomes available.*

*onsemi carries meaningful risk. These risks include automotive and industrial cyclicality, semiconductor inventory corrections, silicon carbide pricing pressure, Chinese competition, customer concentration, execution risk, manufacturing utilization risk, gross-margin risk, AI data-center power adoption risk, 800V DC timing risk, EV demand volatility, geopolitical risk, tariff risk, supply-chain risk, capital allocation risk, and stock-price volatility.*

*The pending Synaptics acquisition adds additional risks. These include shareholder approval risk, regulatory approval risk, integration risk, talent-retention risk, cultural fit risk, product-roadmap risk, stock-for-stock deal risk, and the possibility that expected cost synergies, revenue synergies, or strategic benefits fail to materialize. Synaptics’ edge AI, robotics, tactile sensing, and connected-compute opportunities are still developing, and design activity may take longer than expected to convert into meaningful revenue.*

*Any discussion of AI data-center power, 800V rack architectures, robotics, humanoids, Physical AI, edge AI, tactile sensing, TAM estimates, content-per-system assumptions, long-term margins, revenue scenarios, or future earnings power involves uncertainty. Company-provided targets, management commentary, market projections, and my own assumptions may prove inaccurate or incomplete. Future results may differ materially from current expectations.*

*I receive zero compensation from onsemi, Synaptics, or any related party for this report. I have zero business relationship with onsemi, Synaptics, their management teams, or their investor relations teams.*

*Readers should conduct their own due diligence, review all relevant SEC filings, earnings calls, investor presentations, balance sheets, cash-flow statements, competitive risks, valuation, and personal financial circumstances before making any investment decision. Consider consulting a qualified financial adviser, tax professional, or legal adviser for guidance specific to your situation.*
A one-two punch combo