Renesas Electronics Corporation (6723.T): Deep Dive Research Report

Research as of 12 October 2026. Renesas has a December fiscal year and reports every quarter. Q2 FY2026 (April to June) was reported on 31 July 2026, inside the last 90 days. The company's financial calendar schedules Q3 FY2026 for 29 October 2026, which is positive evidence that the next period has not been released yet.

The six results events used in this report: Q1 FY2025 (call held 24 April 2025), Q2 FY2025 (25 July 2025), Q3 FY2025 (30 October 2025), Q4/FY2025 (5 February 2026), Q1 FY2026 (24 April 2026) and Q2 FY2026 (31 July 2026). The Capital Market Day of 25 June 2026 supplements them.

1. What the Company Does

Renesas designs the small chips that make electronic things behave: when to fire a fuel injector, how hard to brake a wheel, how much power to send to a GPU, when a washing machine should spin. Its core product is the microcontroller (MCU), a single chip that combines a processor, memory and input/output and runs one control job reliably for 15 years. Around the MCU it sells the analog and power chips that sit beside it: voltage regulators, power stages, battery managers, sensor interfaces and memory-interface chips. In cars it also sells larger "system-on-chip" processors (the R-Car family) that run driver-assistance cameras and the digital cockpit. Since 2024 it has also sold the software engineers use to design circuit boards (Altium).

The business is roughly half automotive and half everything else, and the "everything else" is quickly becoming AI data-centre power and memory chips.

How it came to be. Renesas is the product of a consolidation the Japanese chip industry was forced into. Renesas Technology was formed in 2003 from the logic and microcontroller operations of Hitachi and Mitsubishi Electric. It merged with NEC Electronics in 2010 to create Renesas Electronics, putting three national champions' MCU franchises under one roof. The merged company was too big to fail and too inefficient to profit. The March 2011 Tohoku earthquake knocked out its Naka fab, and the global car industry discovered how many of its engine and chassis controllers came from one building in Ibaraki. In 2013 the state-backed Innovation Network Corporation of Japan (INCJ) led a rescue recapitalisation alongside Japanese automakers and suppliers.

The rescue also brought the man who rebuilt the company. Per Renesas's own executive biography, Hidetoshi Shibata was Executive Managing Director at INCJ before joining Renesas in November 2013 as Executive Vice President and CFO. He served on the board from November 2013 to May 2016, rejoined it in March 2018 and became CEO in July 2019. Earlier he was Managing Director of Global Private Equity at Merrill Lynch, a partner at Japanese private-equity firm MKS Partners and an employee of Central Japan Railway. He holds an MBA from Harvard Business School and a BE from the University of Tokyo.

Under Shibata, Renesas ran a private-equity playbook on a semiconductor company: it closed fabs, cut headcount, and then bought its way out of a pure-Japanese, pure-MCU identity:

• Intersil (2017): US power management and analog
• Integrated Device Technology (IDT, 2019): US timing, memory interface and data-centre connectivity
• Dialog Semiconductor (2021): UK/German power-management ICs and Bluetooth
• Transphorm (2024): gallium-nitride power
• Altium (completed 1 August 2024, about A$9.1bn in cash): PCB design software

The biography credits Shibata with leading the Intersil, IDT and Dialog deals. These acquisitions are why a company once synonymous with Japanese automotive MCUs is now a credible supplier of GPU power stages and DDR5 memory chips to American hyperscalers. Two later reversals are also part of the history. Renesas abandoned its silicon-carbide ambitions in 2025 after a prepaid wafer deal with Wolfspeed went bad (the capital-allocation lesson is in Section 8). In February 2026 it agreed to sell the IDT-derived timing business to SiTime, and that sale closed on 1 July 2026.

The value proposition. A car maker or industrial OEM needs a control chip that works the same way in 2040 as it did in 2026, from minus 40 to plus 150 degrees Celsius, inside a functional-safety certified software stack, with a supplier that will still ship it in 15 years. Renesas sells that certainty, together with the power, analog and connectivity parts that let the customer design a whole board around one vendor (Renesas calls these "winning combinations"). In the data centre the proposition is different: it delivers clean, fast-responding power to a 1,000-watt-plus accelerator, and buffers and manages the signal between CPU and DRAM, on the customer's tape-out schedule.

What makes it hard to replicate (headline). Automotive MCUs require decades of qualified process technology, embedded flash at automotive temperature grades, ISO 26262 safety documentation, and an installed base of customer software written for Renesas's instruction sets. Data-centre power requires analog design talent and a slot on each GPU or ASIC reference design. Section 3 has the detail.

A concrete example. Take an electric-vehicle inverter project at a Tier-1 supplier. Renesas supplies:

• the RH850 or 28nm-class MCU that runs the motor-control loop
• the gate drivers and power-management ICs around it
• reference software and a safety manual

The supplier's engineers start from a Renesas evaluation board and design their PCB, increasingly in Altium. The chip is qualified with the car maker, and production starts two to three years later. Renesas then ships that part for the life of the vehicle platform plus spare-parts obligations. Once the motor-control software is written for that MCU, the socket is effectively Renesas's for the next generation too, unless something goes badly wrong.

2. Business Segments

Renesas reports two operating segments: Automotive and Industrial/Infrastructure/IoT (IIoT). A small "Other" line now also carries the divested timing business, which was retroactively moved out of both segments for comparison purposes after the sale closed (Q2 FY2026 call, 31 July 2026). Altium's design software sits within the infrastructure side of the business.

Automotive

What it does. Automotive sells the chips that control a vehicle, mostly to Tier-1 suppliers (who build the electronic control units) and increasingly directly to car makers designing their own electronics architectures. There are three product groups:

• MCUs for powertrain, chassis, body, battery management and inverters
• R-Car SoCs for driver-assistance cameras, the digital cockpit and, increasingly, the central computers of "software-defined vehicles"
• Analog and power devices that support both

Core capability. Renesas has decades of embedded-flash process technology qualified for automotive grades. It also has one of the largest installed bases of automotive control software written for its RH850 and earlier instruction-set architectures, especially at Japanese OEMs and Tier-1s. Replicating that installed base would take a rival a full vehicle-platform cycle per customer.

Why it is separate. Automotive has its own qualification regime, longer design cycles, its own sales force aligned to Tier-1s and OEMs, and economics where volume and longevity matter more than peak price.

Role in the group. This is the historical franchise and the cash anchor. Management describes it plainly as volatile ("sometimes it goes up, sometimes it goes down", Q4 FY2025 call, 5 February 2026). The near-term growth story here is new content per vehicle (the R-Car Gen 4 SoC ramp and 28nm MCUs) rather than vehicle volume.

Main rivals: Infineon, NXP, STMicroelectronics, Texas Instruments, Microchip.

Mix. In FY2025 the two segments were close to an even split, after an automotive downturn during the year was offset by IIoT growth. By Q2 FY2026 Automotive was about 46% of revenue.

Industrial / Infrastructure / IoT (IIoT)

What it does. This is three businesses wearing one segment label:

• Infrastructure: AI data-centre power delivery (multiphase controllers and smart power stages for GPUs and custom ASICs), DDR5 memory-interface chips (registered clock drivers, data buffers, power-management ICs and SPD hubs on server memory modules), and power and control for communications equipment
• Industrial: general-purpose MCUs (the RA, RX and RL78 families) and microprocessors (RZ), sold largely through distribution into factory automation, building control, medical devices, meters and appliances
• IoT/consumer: Dialog-derived power-management ICs, battery chargers and Bluetooth Low Energy for phones, PCs, wearables and appliances

Altium's PCB design software also reports here.

Core capability. The data-centre business relies on analog and mixed-signal design inherited from Intersil and IDT, and above all on having a seat at the design table when Nvidia or a hyperscaler finalises a board. The industrial business relies on breadth: thousands of catalogue parts, development tools and reference designs, so an engineer at a small OEM finds what they need without a sales call.

Why it is separate. It has different customers (hyperscalers, memory module makers, thousands of industrial OEMs through distributors), shorter design cycles, and in the data centre a much faster product cadence than automotive.

Role in the group. This is now the growth engine. IIoT revenue grew about 40% year on year in Q2 FY2026 (Q2 FY2026 call), carried by data-centre demand management called "exceptionally strong." At the June 2026 Capital Market Day management said AI and IT infrastructure should account for about 40% of total revenue within three to four years. Its long-term target mix for 2035 is roughly 35% automotive, 50% infrastructure/IoT and 15% software/digital.

Main rivals: Monolithic Power Systems, Infineon, Analog Devices, Texas Instruments (power); Rambus and Montage Technology (memory interface); STMicroelectronics, NXP, Microchip, Texas Instruments (industrial MCUs); Cadence, Siemens and Zuken (PCB design software).

Mix. About 53% of revenue in Q2 FY2026, and the mix is moving toward IIoT every quarter.

Geographic mix

Renesas sells mostly in Asia. For FY2025 the company's results show:

• China: about 31%
• Asia excluding Japan and China: about 23%
• Japan: about 20%
• Europe: about 15%
• North America: about 10%

These figures reflect where chips are shipped (largely to contract manufacturers, module makers and Tier-1 plants), not where the end product is sold. Much of the data-centre business designed in by US companies, for example, ships to Taiwanese and Southeast Asian board and module assemblers.

Each region behaves differently:

• China is the largest single market. In automotive it is driven by domestic EV makers, where management saw strength in mid-2025 and then a slowdown as subsidies faded (Q2 and Q3 FY2025 calls). Chinese domestic MCU vendors are also increasingly the alternative (Section 5).
• Japan is the legacy home market, built on decades of relationships with Toyota, Denso, Honda and others. It moves with Japanese vehicle production schedules.
• Europe is mainly automotive Tier-1s. It was weak through 2025 and showed signs of inventory rebuilding in early 2026 (Q4 FY2025 call).
• North America is small by shipment but large by influence, because that is where GPU, ASIC and server design decisions are made.

3. Products and Operations

The product lines that matter

1. Automotive MCUs (RH850 and the new 28nm generation). These are the workhorse controllers in engine, chassis, airbag, body and battery ECUs. The 28nm generation replaces the 40nm parts. Management has cited it repeatedly as a content driver: it was gaining traction in Chinese EVs, ADAS and powertrain by Q1 FY2025, and showed "solid growth, particularly in China" by Q1 FY2026. The technically hard part is embedded non-volatile memory at advanced nodes under automotive temperature and reliability grades, plus the ISO 26262 functional-safety documentation every part must carry. 28nm parts are made at a foundry (TSMC), while older generations run in Renesas's own Japanese fabs.

2. R-Car SoCs. These are high-performance automotive processors for camera-based driver assistance, cockpit infotainment and, from Gen 5, centralised "zonal" vehicle computers. Gen 3 parts remain in production with lifecycles that have "extended beyond original expectations" (Q1 FY2026 call). Gen 4 (including the V4H for ADAS) began to ramp at the end of 2025. Management had flagged it in July 2025 ("fourth generation ADAS-related SoC will start to pick up in Q4") and called it "ramping up very smoothly" in April 2026. These chips are fully outsourced to leading-edge foundry nodes. Renesas's value is the architecture, the safety concept and the automotive software ecosystem around it.

3. Data-centre digital power. This is the fastest-growing line. It includes multiphase voltage-regulator controllers and smart power stages that convert 48V or 12V rack power into the sub-1V, very high current an accelerator needs. Management described its share of GPU power sockets as "one-third to 50%" (Q4 FY2025 call). The line is broadening into power for hyperscalers' custom ASICs and into the 800V DC rack architecture: Nvidia named Renesas among 14 silicon partners for 800V DC power at the October 2025 OCP Global Summit. The hard part is transient response, meaning a regulator that holds voltage within millivolts while a GPU's load swings by hundreds of amps in microseconds, plus the firmware and telemetry hyperscalers demand. Renesas is also using the GaN technology it acquired with Transphorm for the higher-voltage conversion stages.

4. Memory interface. These chips sit on every server DDR5 memory module: the registered clock driver (RCD) that buffers commands from the CPU, data buffers, the on-module power-management IC, SPD hubs and temperature sensors. Each new DDR5 speed grade and the new multiplexed-rank (MRDIMM) modules need a new, faster chip, which reopens share at every transition. Management described its memory-interface backlog as showing a "step increase like a staircase" (Q3 FY2025 call).

5. General-purpose MCUs and MPUs (RA, RX, RL78, RZ). These are catalogue parts for industrial and consumer control, with the RZ microprocessors moving into edge-AI vision. They sell mainly through distribution, so the support burden is carried by development tools and reference designs.

6. Power-management and connectivity ICs for consumer and IoT (Dialog-derived). PMICs, chargers and Bluetooth LE for phones, PCs and wearables. The mobile business is seasonal. Management grouped it with data centre as a driver of growth in Q3 FY2025.

7. Altium and "Renesas 365". Altium Designer is PCB design software, and Altium 365 is its cloud collaboration platform. Octopart is a component search engine used by electronics engineers. Renesas 365, announced in March 2025, layers Renesas silicon selection, software and lifecycle management on top of Altium 365. Renesas's strategic bet is that owning the tool where boards are designed lets it influence which components are chosen.

Timing products (clocks and oscillators) were sold to SiTime on 1 July 2026. Renesas received about US$1.5bn in cash and 3,558,691 SiTime shares (SiTime 8-K, July 2026), plus a partnership memorandum covering SiTime MEMS resonators inside Renesas embedded products.

How the chips are made

Renesas runs a hybrid "fab-lite" model:

• Own front-end fabs: mature and specialty nodes in Japan, mostly analog, power and older-generation MCUs
• Foundries: advanced-node MCUs and every R-Car SoC go to TSMC, and some analog and power to other foundries
• Back end: its own assembly and test plants, plus subcontractors in Southeast Asia and China

The process runs from wafer fabrication (in-house or foundry), to wafer probe test, to a "die bank" of tested wafers held as strategic inventory, to packaging and final test, then to distributors or direct customers.

The Japanese front-end network includes:

• Naka (Ibaraki): 300mm, MCUs
• Kofu (Yamanashi): a 300mm power-device line reopened in 2024
• Takasaki (Gunma): legacy 6-inch line, now being closed
• Kawashiri and Nishiki (Kumamoto)
• Saijo (Ehime) and Yonezawa (Yamagata): 200mm

Recent capacity and plant events shape the investment case:

• Utilisation. Front-end utilisation was "slightly below 50%" in mid-2025 (Q2 FY2025 call), the bottom of the inventory correction. Input utilisation had risen to 58% by Q2 FY2026, with a slight further rise expected in Q3.
• Bottlenecks have moved. In Q1 FY2026 management named three: testers limiting how many wafers could be turned into sellable parts, wafer supply from a Taiwanese production partner hit by a late-December 2025 earthquake and power disruption, and the partner's general capacity.
• New capacity investment. In Q1 FY2026 management approved roughly ¥94bn of investment, mostly capacity for AI data-centre digital power. On the same call it referenced a new 300mm line targeted to start production in 2028. By Q2 FY2026 it was promising "step-function" supply increases from late 2026 into 2027 (Section 9 tracks this).
• Kumamoto earthquake. A magnitude 7.1 earthquake struck Kumamoto on 28 July 2026 and halted both Kumamoto fabs. Per Renesas's update of 31 July 2026, Nishiki was back to pre-quake capacity that morning. Kawashiri, with wall cracks, water leaks and an interrupted pure-water supply, was due to restart in phases from 5 August and reach full capacity in about three weeks. Shibata said the earnings impact "should be limited."
• Takasaki closure. The 6-inch Takasaki plant is being wound down. Management says employment will be kept and staff redeployed toward R&D (Q2 FY2026 call). This is the same site where Renesas had planned SiC production before exiting SiC in 2025.
• Inventory policy. Management raised its internal days-of-inventory target from 120 to 150 days (Q4 FY2025 call), explicitly to support AI and data-centre demand and to buffer supply-chain shocks. The die bank is the physical form of that policy.

4. Customers

Renesas has three very different customer types, and the buying relationship with each runs on different logic.

1. Automotive Tier-1 suppliers and car makers. Buyers include Japanese Tier-1s (Denso, Hitachi Astemo and others), European Tier-1s, Chinese EV makers designing their own electronics, and Japanese OEMs. Toyota is also a shareholder with 4.09% at 30 September 2026 (Section 10).

• Who decides: a cross-functional team of the ECU's system architect, software lead, functional-safety engineers and purchasing, often with the car maker's electronics organisation approving or mandating the platform.
• Criteria, in rough order: software compatibility with existing code, safety certification, roadmap continuity (will there be a pin-compatible next generation?), supply assurance (sharply elevated after the 2021 shortage), then price.
• Sales cycle: two to four years from first engagement to start of production. A won socket then ships for five to seven years of platform life, plus aftermarket obligations.
• Switching costs are very high. Changing MCU vendor means rewriting and re-certifying safety-critical software, requalifying the ECU and often re-sourcing companion chips. Customers switch at a platform change, not mid-life. In practice Renesas's automotive revenue in any year is mostly decided by design wins three to six years earlier, multiplied by vehicle production and channel inventory swings.

2. AI data-centre customers. These are GPU vendors (Nvidia is the reference customer for the power business), hyperscalers building custom ASICs (management names hyperscaler ASICs as "the main driver" of AI power growth, Q4 FY2025 call), the ODMs who build their boards, and the DRAM makers and module houses that buy memory-interface chips (Samsung is named in the Korean antitrust probe in Section 8).

• Who decides: the accelerator's board and power architects (for power) and the memory module or platform qualification team (for memory interface).
• Criteria: electrical performance at the bleeding edge, ability to co-design on the customer's schedule, telemetry and firmware, and above all the ability to ship in volume on day one of a ramp.
• Sales cycle: short and repeating. Every GPU generation (now roughly annual) and every DDR5 speed grade is a fresh competition.
• Switching costs are low to moderate. They exist within a generation, because the reference design is locked, but disappear at the next one. Shibata's own framing of the data-centre position on the Q2 FY2026 call was that strength comes from customers "we have been supplying quite exclusively," and that holding that position requires "very strong execution." Supply reliability has become a selection criterion in its own right: in Q1 FY2026 management said "demand is extremely strong, but supply is not keeping pace."

3. Industrial, IoT and consumer OEMs. Thousands of companies, from factory-automation majors to small device makers, plus phone and PC brands for PMICs. Most volume moves through global distributors, so distributors are among Renesas's largest invoiced customers. The engineer chooses the part and the distributor fulfils the order. Criteria are breadth of portfolio, development tools, availability and price. Switching costs are moderate: once firmware is written on a Renesas MCU family, staying is the path of least resistance, but new designs are contestable. This is the channel where Altium and Renesas 365 are meant to influence part selection at the design stage.

Concentration. No single customer dependence is disclosed in the materials reviewed, and the diversified base across thousands of OEMs limits concentration risk. The real concentration is in the data-centre business, where a handful of GPU and ASIC programmes and a few memory makers account for the growth. That is a reflection of quality (Renesas holds sockets on the most demanding boards) but also a risk (Section 8.1).

Contract structures. There are no take-or-pay contracts at scale.

• Automotive runs on platform design wins with annual price negotiations and multi-year supply expectations. After 2021 many OEMs sought longer-horizon supply commitments, which Renesas supports partly through its die bank.
• Data centre runs on purchase orders against a programme ramp, so revenue is visible for a quarter or two.
• Industrial runs through distribution, where the gap between distributor buying (sell-in) and distributor selling (sell-through) creates its own cycle. Renesas tracks channel inventory in weeks: it fell from 8.9 to 8.1 weeks in Q3 FY2025, and management has repeatedly said it will stay cautious, citing lessons from the 2021-2023 cycle.
• Pricing is now an active lever again. Renesas raised prices on selected products from 1 July 2026, though Shibata said it was "not like a significant price hike that becomes a news headline" (Q2 FY2026 call).

5. Competitive Landscape and Moat

Renesas competes in four arenas that share almost no competitors with one another.

Automotive MCUs and SoCs. This is an oligopoly of incumbents with decades of qualification history. Infineon is the clear leader: TechInsights data cited by Infineon shows a 36.0% share of automotive microcontrollers in 2025, up 3.9 points in a year, inside a US$74.4bn automotive semiconductor market. Infineon's AURIX family dominates European powertrain and safety, and it has been gaining share. Renesas is one of the next tier alongside NXP, which is strong in Europe and in vehicle networking and processors, with STMicroelectronics, Texas Instruments and Microchip behind.

• Where Renesas wins: Japanese OEMs and Tier-1s, where its installed software base is deepest, and specific Chinese EV programmes with its 28nm parts.
• Where it loses: European safety and powertrain, where AURIX is entrenched. Infineon's share gain in 2025 implies at least some of it came from Renesas and NXP.
• In SoCs: R-Car competes with Nvidia, Qualcomm and Mobileye for ADAS and cockpit compute. Renesas wins on cost, power and safety heritage in mid-range ADAS and loses at the high end, where Nvidia and Qualcomm offer far more AI compute.

AI data-centre power. This is an engineering race with fast share swings. Monolithic Power Systems (MPS) is the most aggressive rival in GPU power stages, and Infineon, Analog Devices, Texas Instruments and Vicor all compete for accelerator and rack-level power. Renesas's "one-third to 50%" milestone on GPU power sockets (Q4 FY2025) shows it is a co-leader, not a monopolist. The 800V transition widens the field: Nvidia's October 2025 partner list included 14 silicon vendors.

Memory interface. This is a near-triopoly. Montage Technology (China) and Rambus (US) are each estimated at around 40% or more of DDR5 RCDs, with Renesas at roughly 10-15% (trade-press estimates). Together the three hold well over 90% of supply, and that concentration is exactly what drew the South Korean price-fixing raid in July 2026 (Section 8.2). Renesas's angle is to sell the whole module chipset (RCD, data buffer, PMIC, SPD hub, sensors) and to gain share at each speed transition.

General-purpose MCUs. This is a fragmented, distributor-driven market with STMicroelectronics (the STM32 ecosystem is the default for many engineers), NXP, Microchip, Texas Instruments and fast-rising Chinese domestic vendors (GigaDevice and others) competing on price inside China.

Design software (Altium). Cadence (Allegro/OrCAD) and Siemens EDA (Xpedition/PADS) dominate high-end enterprise PCB design. Altium leads among small and mid-sized teams, and Zuken is strong in Japan. Altium's competition is less about features than about whether a chip vendor can own a "neutral" design tool without driving rivals' parts out of it.

Barriers to entry. They are high in automotive and lower everywhere else.

• Automotive: a newcomer needs qualified embedded-flash process technology, safety certification and, hardest of all, a reason for a Tier-1 to rewrite validated software. Chinese domestic vendors are scaling that wall from below with state and customer support, but mainly inside China.
• Data-centre power: the barrier is analog talent and a design-in relationship. MPS proved over a few years that a determined entrant can take major share, so the barrier is real but not durable.
• Memory interface: the barrier is the JEDEC-speed-grade engineering treadmill and memory-maker qualification. It is high enough that only three firms remain.

Structural shifts.

1. Centralisation in cars. Software-defined vehicles consolidate dozens of small MCUs into a few zonal and central computers. That shifts value from small MCUs toward high-end SoCs, where Nvidia and Qualcomm are stronger and Renesas must win with R-Car Gen 5.
2. China localisation. Policy and customer preference are pushing Chinese automotive and industrial MCU demand toward domestic suppliers.
3. AI power architecture churn. Each change (48V, 800V DC, vertical power delivery, GaN) resets the competitive field.
4. Vertical integration by chip vendors into tools, which Renesas pioneered with Altium.

Moat: Narrow - switching costs (automotive installed software base and safety qualification), with intangible assets in analog/power design

Section 4 shows that an automotive MCU socket is re-competed only at platform boundaries and that leaving Renesas means rewriting certified software. That advantage held through the 2011 earthquake, the 2021 Naka fire and the 2023-2025 inventory correction, without the Japanese installed base defecting. The moat is not wide, for two reasons. Infineon gained substantial automotive MCU share in 2025, showing that new platforms are contestable. And the fastest-growing businesses (data-centre power and memory interface) are re-won generation by generation, with switching costs that last only a product cycle. The single thing most likely to erode the moat is vehicle centralisation moving the high-value control function onto SoCs and zonal controllers, where Renesas's legacy software base counts for less.

6. Industry

What drives demand. Renesas sits at the junction of three demand engines with different rhythms:

• Vehicle electronics content. Car volumes grow slowly. Semiconductor content per vehicle grows faster through electrification, driver assistance and centralised compute. TechInsights put the 2025 automotive semiconductor market at US$74.4bn (as cited by Infineon, April 2026).
• AI data-centre infrastructure. Power demand per rack is rising from tens of kilowatts toward a megawatt. Renesas's EVP Zaher Baidas framed the shift at the Capital Market Day as "workloads are shifting from training models on massive datasets to AI inference," which spreads AI hardware demand from GPU training clusters to a broader inference build-out. Every watt delivered to an accelerator passes through power conversion stages, and every server memory channel needs interface chips, so this demand is proportional to installed AI compute rather than to GPU unit counts alone.
• Industrial and consumer electronics. These are cyclical and tied to capital spending, factory automation, appliances and handsets.

Where Renesas sits in the supply chain. It is a chip designer and part-manufacturer between the foundries and equipment makers (TSMC, Japanese tool makers, tester suppliers) and the module makers, Tier-1s and server ODMs. It does not sell finished systems. It depends on TSMC for leading-edge parts and on Japanese fabs, and Japan's earthquake geography, for mature-node parts.

Regulation and policy.

• Automotive safety standards (ISO 26262, AEC-Q100) are the effective licence to sell in automotive.
• Export controls and tariffs shape China exposure. US tariffs in 2025 caused demand-volume worries more than direct cost, and the eventual 15% US-Japan tariff settlement removed the worst case (Q2 FY2025 call).
• Japanese industrial policy subsidises domestic semiconductor capacity (the TSMC-backed JASM fab in Kumamoto, and Rapidus). That strengthens the local supply ecosystem around Renesas.
• Chinese localisation policy works the other way, encouraging domestic MCU and power substitution.
• Antitrust attention has arrived in concentrated niches such as memory-interface chips.

Cyclicality. Automotive and industrial semiconductors are highly cyclical through inventory, not end demand. The 2021-2022 shortage triggered double-ordering, which unwound into a long correction through 2023-2025. Renesas's own utilisation fell below 50% in mid-2025 and channel inventories were drained. The pattern is now reversing. Supply is tight "everywhere you look" (Shibata, Q2 FY2026), lead times are lengthening, price increases are back, and the Nexperia supply dispute in late 2025 raised the risk of customers stockpiling (Q3 FY2025 call). This is the early part of the cycle, when booking strength can overstate end demand.

Tailwinds: AI inference build-out and the 800V/GaN power transition; vehicle centralisation and ADAS adoption; recovery in industrial inventories; mature-node supply tightening that restores pricing power.

Headwinds: China localisation; slow Western EV adoption; the shift of value in cars toward high-end compute; a much faster product cadence in data centres that raises R&D intensity; and the end of the AI capex boom whenever it comes, which would hit the newest, fastest-growing revenue first.

7. Growth Triggers

Ranked by potential to change the business:

• AI infrastructure power, broadening from GPUs to hyperscaler ASICs and 800V architectures. Management said the AI business would roughly double in FY2026, with "ASICs for hyperscalers" as "the main driver" and power growing faster than memory (Q4 FY2025 call, 5 February 2026). It called data-centre demand "exceptionally strong" and expected strength to continue into Q3 (Q2 FY2026 call, 31 July 2026). At the Capital Market Day (25 June 2026) it set the goal that AI and IT infrastructure reach about 40% of revenue within three to four years, citing 800V DC distribution, vertical power delivery and GaN switches. This has been repeated on the Q3 FY2025, Q4 FY2025, Q1 FY2026 and Q2 FY2026 calls.

• Step-function supply increases from late 2026 into 2027. About ¥94bn of investment, mostly for AI data-centre digital power, was approved, with a new 300mm line targeted for 2028 (Q1 FY2026 call, 24 April 2026). Management said "the step function type of increase is something we are working on right now" for the end of 2026 and next year, alongside expanded capacity from Taiwanese foundry partners in H2 2026 and 2027 (Q2 FY2026 call). This is repeated across both 2026 calls.

• Revenue doubling by 2030 and tripling by 2035 versus 2025. This is the "three-stage rocket" of AI Infrastructure & Compute, Physical AI & Software-Defined Vehicle, and Intelligence at the Edge, with a target mix of 35% automotive, 50% infrastructure/IoT and 15% software/digital (Capital Market Day, 25 June 2026).

• R-Car Gen 4 ramp into mass production and the Gen 5 / software-defined-vehicle generation. Management said "fourth generation ADAS-related SoC will start to pick up in Q4" (Q2 FY2025 call, 25 July 2025), that Gen 4 was "ramping up very smoothly" (Q1 FY2026 call), and that new products are driving content growth (Q2 FY2026 call). SDV was positioned as one of the three growth stages at the Capital Market Day. Repeated across three calls.

• 28nm automotive MCU adoption, led by China. The 28nm MCU replacing 40nm was gaining traction in China EVs, ADAS and powertrain (Q1 FY2025 call, 24 April 2025), was expected to be a growth driver in 2026 (Q4 FY2025 call), and was growing particularly in China (Q1 FY2026 call). Repeated across three calls.

• Memory-interface share gains at DDR5 speed transitions. Management expected share recovery and a backlog rising "like a staircase" (Q3 FY2025 call, 30 October 2025), and described its memory-interface position as "a very good position" (Q2 FY2026 call).

• Selective price increases. Price revisions on certain products took effect 1 July 2026, with the effect to come through gradually in second-half results (Q2 FY2026 call), after management signalled willingness to raise prices in April (Q1 FY2026 call).

8. Key Risks

8.1 Data-centre socket loss at a generation change. The fastest-growing revenue rests on a handful of GPU and ASIC programmes, each re-competed at the next generation (Section 4). The mechanism is a lost reference design, typically to MPS or Infineon, on a single major accelerator platform. That would remove a large share of incremental growth within a year, and there is no installed-base cushion as in automotive. Management has itself tied the outcome to execution: "It's all up to us whether we can execute" (Q3 FY2025 call). Supply shortfalls in 2026 (Section 3) add a twist, because a customer that cannot get enough parts qualifies a second source and that second source may keep share. Calibration: moderate-to-high probability of some share loss at some point, moderate impact on the business, high impact on the growth narrative.

8.2 Antitrust exposure in memory interface. On 15 July 2026, Seoul Central District Prosecutors' Office investigators raided Renesas, Rambus and Montage on suspicion of coordinating memory-interface chip prices supplied to Samsung Electronics and others, in violation of Korea's Fair Trade Act. The three are said to control more than 93% of global supply. The mechanism runs through fines, follow-on probes in other jurisdictions, and customer leverage in price negotiations even without a finding. No wrongdoing has been established, and the materials reviewed include no Renesas statement on the substance. Calibration: low-to-moderate probability of a material penalty, but a real near-term overhang on one of the two AI-linked product lines.

8.3 China automotive and localisation. China is the largest shipment region (Section 2), and Chinese EV demand has already swung Renesas's quarters, stronger than expected in Q2 2025 and then a "reactionary decrease" as subsidies ended (Shibata, Q2 FY2025 call). The structural version of the risk is Chinese OEMs switching new platforms to domestic MCU vendors under policy pressure. Because those switches happen at platform boundaries (Section 4), the damage would be slow and invisible for two to three years and then show up as missing growth. Calibration: high probability of gradual erosion; moderate impact, partly offset by 28nm wins in the same market.

8.4 Manufacturing concentration in seismic zones. Renesas's own fabs are in Japan and its advanced parts come from Taiwan. Within seven months the business absorbed a Taiwanese earthquake hitting a foundry partner (December 2025) and the Kumamoto earthquake halting two of its own fabs (July 2026). Earlier, the 2011 earthquake and the 2021 Naka fire showed how one site can stop the car industry. The 150-day inventory target is a partial hedge. Calibration: low probability per year of a long outage, but potentially severe impact. In a tight market, an outage now costs share as well as revenue (8.1).

8.5 Capital allocation and the M&A track record. Renesas's transformation was bought, and not every purchase worked. The Wolfspeed wafer-supply deposit, made to secure silicon carbide for a SiC business Renesas later abandoned, ended in a ¥235bn valuation loss in Q2 FY2025 and a conversion into Wolfspeed equity, notes and warrants after Wolfspeed's Chapter 11. Altium, the largest acquisition, is a strategic bet that a chip vendor can own a neutral design tool. Its recurring-revenue growth slowed between the Q3 FY2025 call and Q1 FY2026 (per Capital Market Day coverage). The company now holds the SiTime proceeds and stake, the Wolfspeed position and rising cash flow, and the CFO said at the Capital Market Day: "We do not intend to make a rushed decision." Calibration: moderate probability that a future large deal destroys value, given the record.

8.6 Cycle overshoot. Section 6 shows the industry entering the phase where customers stockpile. Renesas is deliberately building inventory toward a higher target, raising prices and adding capacity at the same time. If AI capex pauses or automotive customers find they over-ordered after the Nexperia scare, the cycle could reverse as it did after 2022: channel inventory piles up, utilisation falls, and the new capacity depreciates over a weaker base. Management has said it will stay cautious on channel inventory, but the 150-day target is an explicit decision to carry more risk on its own balance sheet. Calibration: moderate probability within two to three years, moderate-to-high impact.

9. Walk the Talk

Results events used: Q1 FY2025 (24 April 2025), Q2 FY2025 (25 July 2025), Q3 FY2025 (30 October 2025), Q4/FY2025 (5 February 2026), Q1 FY2026 (24 April 2026) and Q2 FY2026 (31 July 2026, within 90 days of this report).

The tariff haircut shows how this management guides. In April 2025 Shibata openly called the reduction in guidance:

"It's like a midstream growth percentage point haircut... very emotional and psychological." - Hidetoshi Shibata, Q1 FY2025 call, 24 April 2025

Three months later the tariff hit had not materialised and the quarter beat. The same pattern repeated in Q1 FY2026: management guided to a sharp profitability step-down for Q2 based on energy costs, Golden Week maintenance, salary increases and FX, and the quarter came in far ahead.

Supply is where the record is weakest. In April 2026 management promised incremental relief through Q2 and a larger step-up from Q3. By July the language had moved:

"From here on to the end of the year and next year, the step function type of increase is something we are working on right now." - Hidetoshi Shibata, Q2 FY2026 call, 31 July 2026

Some of the delay is beyond management's control (a Taiwanese partner's earthquake, then Kumamoto). Even so, the timetable for meeting demand has moved out by about two quarters, and in a business where Section 8.1 says supply reliability decides socket wins, this is the commitment to watch on 29 October.

A quietly dropped strategy. Silicon carbide was announced in July 2023 as a production start for early 2025 at Takasaki. It was abandoned in 2025, the team was disbanded, the Wolfspeed deposit was written down, and the Takasaki 6-inch plant is now closing. Management has been candid about the exit, but it remains the clearest instance of a long-range promise reversed.

Assessment. On near-term operating guidance, this management is consistently conservative: across six calls it guided below the outcome at least three times and never meaningfully missed a quarterly guide in the period reviewed. On multi-year strategic bets (SiC, Wolfspeed and, so far, the pace of Altium's contribution) its record is mixed. The current capacity timetable is slipping rather than broken.

10. Ownership, Governance and Shareholder Friendliness

Part 1 - Ownership and control. Renesas no longer has a controlling shareholder. The state fund that rescued it in 2013 has sold down and no longer appears among the top ten holders. Per the company's shareholder distribution page as of 30 September 2026, the largest registered holders are:

• The Master Trust Bank of Japan (trust account): 15.47%
• Custody Bank of Japan (trust account): 6.03%
• Toyota Motor Corporation: 4.09%
• Foreign custodian accounts (State Street, BNY Mellon, Goldman Sachs, Clearstream, JPMorgan Chase nominee accounts): roughly 1.4-3.7% each

The top ten together hold 39.12%, and 40,410,807 treasury shares are excluded from the percentages. The trust-bank and custodian accounts are nominee holders for domestic and foreign index, pension and active funds. Large-shareholding (5%) reports have been filed by asset managers including BlackRock Japan, Fidelity and T. Rowe Price (Section 11). There is one share class, no golden share and a wide float. For a minority shareholder, this structure means voting control sits with dispersed institutions. The only strategic holder of note, Toyota, is also a major end customer, which aligns it with supply continuity rather than any particular capital policy.

Part 2 - Governance. Renesas runs a three-committee board structure.

• As of March 2026, six of the directors are independent outside directors, led by Lead Independent Outside Director Jiro Iwasaki. They include Selena Loh LaCroix, Noboru Yamamoto, Takuya Hirano, Tomoko Mizuno and Kimberly Mathisen, who joined effective 25 March 2026 (company notices). Shibata is the executive director.
• The Nomination Committee and the Audit & Sustainability Committee consist entirely of independent outside directors. The Compensation Committee has an outside chair (Corporate Governance Report).
• Executive pay relies heavily on equity: restricted stock units and performance share units are settled from treasury shares (disposal notice of 14 February 2025; shelf registrations of 18 July 2025 and 10 April 2026), so part of pay is linked to measured performance.
• No material related-party transaction with a controlling holder exists, because there is no controlling holder. No auditor change or qualified opinion surfaced in the materials reviewed.
• Governance is unremarkable for a large-cap Japanese company, and better than most on independence.

Part 3 - Capital returns.

Dividends. Renesas resumed dividends with a year-end payment of ¥28 per share for FY2023, after a long suspension that dated from the rescue years (company notice on the dividend resumption). It paid ¥28 again for FY2024 and FY2025, and the CFO said on the Q4 FY2025 call: "We will continue to pay dividends worth JPY 28 a share." The dividend has been held flat for three years, a token payout for a business of this cash generation.

Buybacks and dilution. The company's dividend-resumption notice describes two buyback rounds totalling about ¥250bn, in June 2022 and April 2023. The second falls inside the three-year window. No new repurchase programme was found in company announcements for FY2024, FY2025 or FY2026 to date. Treasury-share activity in that period was disposals to settle stock compensation. MoatMap's database records no buybacks in the ~90 days since 14 July 2026. Cash went instead to the Altium acquisition (August 2024) and the debt that funded it. Management said the SiTime proceeds would go first to debt repayment and shareholder returns absent major M&A (Q4 FY2025 call), but at the June 2026 Capital Market Day the CFO deferred any decision. The precise net change in shares outstanding over three years could not be verified. Its direction is mild dilution from stock-compensation disposals, with no offsetting repurchase since 2023.

Verdict: Neutral. The dividend is maintained but flat and small. The last buyback was in April 2023, and since then cash has gone to acquisitions and deleveraging, with a decision on the SiTime proceeds explicitly deferred.

11. Insider Activities

Over the past 12 months no director or executive-officer dealing appears in MoatMap's Japan disclosure data (current to 12 October 2026). The only filings are three large-shareholding (5%-rule) reports by institutional asset managers:

• T. Rowe Price Associates reported a reduction of 26,542,826 shares, about 1.45% of shares outstanding (EDINET Large Shareholding Report, 2026-05-12).
• BlackRock Japan reported an increase of 19,831,191 shares, about 1.09% (EDINET Large Shareholding Report, 2026-06-05).
• Fidelity Management & Research filed a special-exemption large-shareholding report indicating an increase, with no share count disclosed in the data (EDINET Large Shareholding Report, 2026-06-09).

Japanese 5% reports publish no transaction price, so values cannot be stated. These are portfolio adjustments by fund managers acting for clients, not purchases by people running the company. T. Rowe Price's reduction has no disclosed reason, which is normal for a fund manager. The net direction is slight institutional accumulation. With no insider buying or selling by management, the read is neutral.

12. Scenarios

Bull case. The step-function capacity arrives on time in late 2026 and 2027, so Renesas ships into the AI inference build-out instead of rationing it. Hyperscaler ASIC power wins add a second leg beside GPU sockets, Renesas holds its co-leadership through the 800V and GaN transition, and memory interface gains share at the next DDR5 and MRDIMM transitions with the Korean probe ending without material findings. In cars, R-Car Gen 4 becomes a volume platform, Gen 5 lands central-compute wins at Japanese and Chinese OEMs, and 28nm MCUs replace the 40nm installed base. Two or three years out, Renesas looks like a data-centre power and memory company with a large, stable automotive franchise beneath it. The "three-stage rocket" is visibly on its first stage, and the SiTime and Wolfspeed proceeds have gone to debt reduction and returns rather than another large purchase.

Base case. Management keeps doing what its record says it does: guiding conservatively and beating quarters while the supply timetable slips a quarter or two. AI infrastructure grows strongly but lumpily, winning some sockets and losing some at each generation as MPS and Infineon compete hard. Automotive drifts upward on new content while China volumes and localisation offset part of the gain, and Infineon keeps a lead in European safety MCUs. Selective price increases stick while supply stays tight. Altium contributes steadily but does not change component selection at scale. The dividend stays flat and capital returns stay modest until management decides what to do with the SiTime cash.

Bear case. 8.6 and 8.1 compound. The current tightness turns out to be partly customers stockpiling after the Nexperia scare and the earthquakes, and when AI capex digests, Renesas is left with a 150-day inventory target, new capacity and lengthening channel weeks just as a hyperscaler or GPU generation moves to MPS. 8.2 adds a fine and harder negotiations with memory makers. 8.3 plays out slowly: Chinese platforms launched in 2026-2027 choose domestic MCUs, so automotive growth disappears in 2028-2029 even as vehicle electronics content rises. If management answers with another large acquisition (8.5), the business ends up more complex and more leveraged at the bottom of the next cycle, repeating the pattern of its first decade.