← Back to blog
Essay No. 034  ·  AI Infrastructure  ·  Melbourne, Australia
AI Infrastructure MediaTek Dimensity 9500 TSMC Qualcomm Snapdragon On-Device AI Smart Edge Automotive ASIC 2nm N2P Edge AI Mobile AI

The Android Flagship Breakout.Original analysisNot investment advice

How MediaTek moved from cheap-chip stigma to TSMC-node flagship and edge-to-cloud AI platform.
PM
Pugalenthi Magendran
April 2026  ·  Melbourne, Australia
12 min read

MediaTek’s 2021 TSMC 4nm flagship bet was not a one-off. It was the start of a structural climb. By 2026, MediaTek has flagship Android credibility, early TSMC 2nm access, on-device AI ambitions, Smart Edge growth, automotive traction, and a real data-center ASIC story. The question is no longer whether MediaTek can enter flagship. It is whether it can turn flagship credibility into durable premium pricing, ecosystem trust, and platform power.

In 2021, MediaTek had a brand problem. Not a volume problem. A brand problem.

MediaTek already shipped huge numbers of chips. It had become powerful in midrange Android. It had benefited from Huawei’s collapse and the rise of affordable 5G phones. But in flagship Android, the mental map was still simple. Qualcomm was premium. Samsung was trying. Apple was separate. MediaTek was cheaper.

The uploaded SemiAnalysis article argued that this was starting to change. MediaTek had climbed from Dimensity 1000 to Dimensity 1100 and 1200, used TSMC 6nm early, and was preparing a flagship 5G SoC on TSMC 4nm.1

In 2026, that thesis looks much bigger. MediaTek did not merely enter flagship. It stayed.

Diagram · MediaTek — from cheap-chip stigma to platform breakout
Stage 01

Volume player

Strong in low-end and midrange Android, weak in flagship perception.
Stage 02

Premium challenger

Dimensity 1000 / 1100 / 1200 entered premium/high-end Android.1
Stage 03

Flagship platform

Dimensity 9400 + 9500 made MediaTek a real Android flagship competitor.23
Stage 04

Edge-to-cloud platform

Smart Edge, automotive, AI PCs, and cloud ASICs.78
A simplified, original four-stage arc of MediaTek’s climb. Stage labels are this essay’s framing, not a MediaTek diagram.
Key idea

MediaTek’s 2021 TSMC 4nm flagship bet evolved into a broader platform breakout. Dimensity 9500 shows MediaTek can compete in flagship Android performance, gaming, imaging, and on-device AI. Its TSMC N2P chip shows it remains early on leading nodes. Its automotive, Smart Edge, connectivity, and AI ASIC push shows the company is no longer only a smartphone SoC vendor. The risk is that premium smartphone demand is cyclical, Qualcomm remains strong, and MediaTek must prove it can convert technical credibility into durable premium margins and ecosystem trust.


I. The 2021 thesis was about credibility

In July 2021, Dylan Patel published a SemiAnalysis piece arguing that MediaTek already had Android volume but needed premium credibility. The Dimensity 1000 / 1100 / 1200 family pushed MediaTek into the upper-midrange and premium tiers, often in the roughly $300–$600 phone band. A flagship 5G SoC on TSMC 4nm was in the pipeline, with an expected CPU layout of one X2 + three A710 + four A510 cores and target flagship pricing around RMB 4,000 / about $615 and above. The piece flagged that MediaTek still needed mmWave and a more competitive GPU architecture to fully enter the $800+ segment, and argued that MediaTek’s special relationship with TSMC — early on 6nm, early on 4nm — was the lever it was using to break out of the cheap-chip stigma.1

2021 thesis

MediaTek already had Android volume. The real question was whether early TSMC node access could help it break into flagship credibility.


II. TSMC was the ladder

The most important character in MediaTek’s rise is not only MediaTek. It is TSMC. MediaTek’s premium credibility has been tied to early access to TSMC’s most competitive nodes — 6nm in 2021, 4nm shortly after, then a first flagship Dimensity on TSMC 3nm with volume production expected in 2024, and a flagship SoC tape-out on TSMC N2P with volume production expected by late 2026.145

Diagram · The TSMC ladder MediaTek climbed
6nm
Upper-midrange credibility· early TSMC 6nm
20211
4nm
Flagship entry· first flagship 5G SoC
~20221
3nm
Flagship continuity· first Dimensity on N3, ~2024 vol prod
20244
N2P
Early-node leadership attempt· nanosheet 2nm family
late 20265
A simplified, original ladder visual. Node years are based on MediaTek and TSMC public statements; not a MediaTek or TSMC chart.

MediaTek and TSMC framed the partnership in concrete process terms. TSMC said N3 offered up to 18% speed improvement at the same power, 32% power reduction at the same speed, and roughly 60% logic density increase versus N5. TSMC and MediaTek frame N2P as a nanosheet 2nm family option offering up to 18% performance at the same power, about 36% power reduction at the same speed, and 1.2x logic density versus N3E.45

MediaTek’s rise is the story of a fabless company turning TSMC access into brand elevation.


III. Dimensity 9400 proved the architecture

MediaTek’s fourth-generation flagship Dimensity 9400 was built on TSMC’s second-generation 3nm process and a second-generation All Big Core design with one Cortex-X925 above 3.62GHz, three Cortex-X4 cores, and four Cortex-A720 cores. MediaTek claimed up to 35% faster single-core, 28% faster multi-core, and up to 40% better power efficiency versus Dimensity 9300.3

Dimensity 9400 mattered because it showed MediaTek was not relying only on node advantage. The architecture decisions — All Big Core, larger caches, advanced GPU, and updated NPU — were the kind of choices flagship SoCs are judged by.

All Big Core was not a cheap-chip move. It was a flagship sustained-performance move.


IV. Dimensity 9500 makes the AI angle unavoidable

MediaTek’s Dimensity 9500 launch frames the chip as a third-generation All Big Core CPU with one ultra core at 4.21GHz, three premium cores, and four performance cores, claiming up to 32% higher single-core and 17% higher multi-core performance versus the previous generation, and up to 55% lower ultra-core power at peak. MediaTek also frames the Arm G1-Ultra GPU with up to 33% peak performance gain and 42% improved power efficiency. NPU 990 with Generative AI Engine 2.0 doubles compute power per MediaTek; BitNet 1.58-bit large-model processing is supported with up to 33% lower power consumption for large-model processing; four-channel UFS 4.1 storage is included, with large AI model loading reported as 40% faster, and 128K-token long-text processing supported.2

Reading the claims

These are MediaTek claims, not independent benchmarks.

Dimensity 9500 performance and efficiency figures are useful direction-of-travel signals from MediaTek’s own materials. They depend on the test conditions, comparison points, and workloads MediaTek picked. Independent reviews and broader OEM device data are the bar for converting these into real-world expectations.2

Diagram · Dimensity 9500 architecture stack, simplified
01
CPU 3rd-gen All Big Core, 4.21GHz ultra core2
Compute
02
GPU Arm G1-Ultra; +33% peak, +42% efficiency2
Graphics
03
NPU 990 Generative AI Engine 2.0; 2x compute2
AI
04
Memory + storage 4-channel UFS 4.1; 40% faster AI model loading2
Memory
05
On-device AI BitNet 1.58-bit, 128K context, agentic AI2
Workloads
A simplified, original stack based on MediaTek’s Dimensity 9500 launch framing. Not a MediaTek chart.

The flagship phone is becoming an edge-AI computer. MediaTek wants Dimensity to be one of its main engines.


V. The financial signal matters

MediaTek’s Q4 2025 results frame revenue at NT$150.188B — up 5.7% QoQ and 8.8% YoY — with the sequential increase mainly driven by strong Dimensity 9500 ramp and a favourable exchange rate, and full-year 2025 revenue at NT$595.966B, up 12.3% YoY.6 A flagship SoC only matters if it shows up in revenue quality. In 2021 the question was whether MediaTek could climb ASP. By the end of 2025 the ramp was visible enough for management to call it out in earnings.

Card · MediaTek Q4 2025 / FY2025, simplified
NT$150.2B
Q4 2025 revenue6
+5.7%
QoQ growth6
+8.8%
YoY growth6
NT$596B
FY2025 revenue (+12.3% YoY)6
Figures as reported by MediaTek. Treated as company disclosures; this essay is not investment advice.

Dimensity 9500 is not just a spec sheet. It is part of MediaTek’s revenue upgrade story.


VI. But smartphones are not enough anymore

MediaTek’s Q1 2026 framing makes the limit clear. Mobile Phone revenue fell 17% QoQ and 15% YoY, accounting for 49% of total revenue. MediaTek attributed the pressure to industry resource concentration on data centers, which raised smartphone component costs, and said it expected global smartphone shipments to decline about 15% in 2026.7

MediaTek’s flagship phone success gives it credibility, but its future growth needs Smart Edge, automotive, connectivity, and data-center ASICs.


VII. Smart Edge is becoming almost as important as mobile

The same Q1 2026 framing puts Smart Edge Platforms at +23% QoQ and +13% YoY, with the segment reaching 46% of total revenue. MediaTek attributes growth to connectivity, computing, automotive, and projects with top-tier consumer brands, notebook makers, telecom operators, carmakers, and cloud service providers, with Wi-Fi 8, 5G NR-NTN, 10G PON, and agentic-AI cockpit programs called out.7

Diagram · Q1 2026 revenue mix, simplified
Mobile Phone
49%7
Smart Edge
46%7
Other
~5%
A simplified, original visual of MediaTek’s Q1 2026 revenue mix. Percentages as reported by MediaTek.7

The Dimensity story proved MediaTek could climb the stack. Smart Edge shows it wants to leave the smartphone box.


VIII. The data-center ASIC story is now real

MediaTek’s Q1 2026 transcript describes a first AI accelerator ASIC project for a U.S. hyperscale customer progressing well, expected to contribute around $2B in Q4 2026 revenue and to scale to multiple billions of dollars in 2027. A second AI accelerator ASIC project is targeted for mass production by the end of 2027.7 Reuters reports that MediaTek expects roughly $1B in cloud AI chip revenue in 2026 and multiple billions from its first AI accelerator ASIC project in 2027, with the company targeting a 10–15% share of an estimated $50B data-center ASIC TAM.8

The separate Nvidia GB10 Grace Blackwell Superchip in DGX Spark was co-designed with MediaTek; DGX Spark includes 128GB unified memory and up to 1 PFLOP of FP4 AI performance per MediaTek’s framing.9 The point is not that MediaTek competes directly with Nvidia. The point is that MediaTek now has design credibility in AI compute subsystems.

Diagram · MediaTek AI ASIC opportunity, simplified
Project 01

First hyperscale ASIC

~$2B Q4 2026, scaling to multiple billions in 2027.7
Project 02

Second ASIC

Mass production targeted by end of 2027.7
TAM share

10–15% of ~$50B

MediaTek-targeted share of the data-center ASIC TAM.8
A simplified, original framing of MediaTek’s data-center ASIC opportunity based on company guidance and Reuters reporting.78

The AI ASIC story is where MediaTek’s power-efficient SoC DNA meets the data-center custom-silicon wave.

Diagram · MediaTek edge-to-cloud platform map, simplified
Device
Dimensity
flagship + premium phones2
Smart Edge
AI PCs + IoT
edge compute + agents7
Automotive
Cockpit + ADAS-adj.
agentic AI cockpit7
Connectivity
Wi-Fi 8 / NTN / PON
5G NR-NTN, 10G PON7
Cloud
AI ASICs
hyperscaler accelerator design8
A simplified, original platform map. Each tile represents a category MediaTek now competes in.

IX. Why ASML and TSMC matter here

MediaTek sits at the collision point between mobile AI, smart edge, automotive compute, connectivity, and cloud ASIC demand — all of which depend on leading-node manufacturing. ASML’s 2025 Annual Report frames AI as requiring leading-edge high-performance processors and a significant increase in DRAM relative to traditional compute architectures.12 TSMC’s 2026 North America Technology Symposium positions advanced nodes as relevant for next-generation AI, HPC, and mobile applications.11

MediaTek is no longer only riding the smartphone cycle. It is riding the edge-AI and custom-silicon cycle.


X. Where Qualcomm still matters

Be fair to Qualcomm. The Snapdragon 8 Elite Gen 5 launch frames the chip around a third-generation Oryon CPU with a claimed 20% CPU performance improvement, 23% GPU improvement, and 37% faster Hexagon NPU performance, with major premium OEM support.10 Qualcomm continues to bring custom Oryon CPU, Adreno GPU, modem-RF depth, carrier relationships, U.S. premium Android strength, and a long OEM familiarity advantage.

Dimension
Qualcomm today
MediaTek today
CPU
Third-gen Oryon, +20% perf claim.10
Third-gen All Big Core, 4.21GHz ultra core.2
GPU
Adreno, mature drivers and OEM tuning.10
Arm G1-Ultra, +33% peak / +42% efficiency claim.2
NPU
Hexagon NPU, +37% perf claim.10
NPU 990 with Generative AI Engine 2.0.2
Modem-RF
Deep modem-RF integration, broad carrier support.
Strong modem; mmWave still a relative gap historically.1
OEM mix
U.S. premium Android, Samsung flagships, broad reach.10
Strong China flagships; selective global premium wins.2
Beyond phones
Auto, PC, XR, IoT, modem-RF infrastructure.
Smart Edge, automotive, AI PCs, AI ASICs.78

The correct claim is not that MediaTek has beaten Qualcomm. The correct claim is that MediaTek has made the Android flagship fight real again.


XI. The in-house silicon risk

Large Android OEMs may experiment with more in-house silicon over time. That can pressure merchant SoC vendors regardless of architectural strength. The risk is structural: premium brands want differentiation, custom silicon gives leverage, and the most ambitious OEMs may build internal chips for flagship control. MediaTek is fighting Qualcomm, but it is also competing with the desire of major OEMs to own more of the silicon stack.

MediaTek is fighting Qualcomm, but it is also fighting the desire of major OEMs to own more of the silicon stack.


XII. The actual 2026 thesis

The correct claim is not “MediaTek is no longer a cheap-chip company.” That is too simple.

2026 thesis

MediaTek’s 2021 TSMC 4nm flagship bet evolved into a broader platform breakout. Dimensity 9500 shows that MediaTek can compete in flagship Android performance, gaming, imaging, and on-device AI. Its TSMC N2P 2nm chip shows it remains early on leading nodes. Its automotive, Smart Edge, connectivity, and AI ASIC push shows the company is no longer only a smartphone SoC vendor. The risk is that premium smartphone demand is cyclical, Qualcomm remains strong, and MediaTek must prove it can convert technical credibility into durable premium margins and ecosystem trust.


XIII. What could break the thesis?

MediaTek can win specs and still struggle to win durable premium pricing outside selected Android OEMs.

Bear case · what could break the thesis
  1. Smartphone demand. MediaTek itself expects global smartphone shipments to decline ~15% in 2026.7
  2. Component costs. Higher DRAM and component costs can force OEMs to raise prices.
  3. Qualcomm strength. Qualcomm remains very strong in premium Android.10
  4. Apple separate. Apple is unreachable for merchant Android SoCs.
  5. Samsung mix. Samsung may continue using Qualcomm / Exynos mix.
  6. OEM in-house silicon. Chinese and global OEMs may experiment with internal chips.
  7. Geographic concentration. Premium wins may remain stronger in China than U.S. flagship channels.
  8. ASIC lumpiness. Data-center ASIC revenue may be large but lumpy.7
  9. Customer concentration. AI ASIC projects can depend on one or two customers.
  10. 2nm execution. Yield, cost, timing, and wafer allocation all matter.5
  11. Software depth. MediaTek must maintain drivers, AI tooling, and OEM support at flagship level.
  12. GPU and modem perception. Some markets still favour Qualcomm on graphics and connectivity.

XIV. What could break the bear case?

MediaTek now has the ingredients Qualcomm never wanted it to have.

Bull case · what could break the bear
  1. Leading-node access. Early TSMC node access on 3nm and N2P.45
  2. Flagship CPU/GPU/NPU. Dimensity 9400 and 9500 are credible flagship platforms.23
  3. On-device AI. NPUs and memory systems gain value as AI moves to the device.2
  4. Smart Edge. Now nearly as important as mobile in revenue mix.7
  5. Automotive + connectivity. Cockpit, Wi-Fi 8, 5G NR-NTN, 10G PON create new lanes.7
  6. AI ASIC pool. Targeting 10–15% of a large data-center ASIC TAM.8
  7. TSMC partnership. Process credibility from a long-running relationship.45
  8. Customer demand for alternatives. OEMs want options to Qualcomm.
  9. Power-efficient DNA. Useful from phone to edge to cloud.
  10. Flagship halo. Premium phones validate the broader silicon platform.

MediaTek now has the ingredients Qualcomm never wanted it to have: leading-node access, flagship CPU/GPU/NPU credibility, automotive and edge expansion, and a credible path into data-center ASICs.


XV. What to watch

What to watch
  • Dimensity 9500 phone adoption.2
  • Dimensity 9600 / next flagship timing.
  • 2nm / N2P production timing.5
  • TSMC wafer allocation.
  • Sustained performance versus Snapdragon.10
  • GPU driver maturity.
  • On-device AI developer support.
  • Premium ASP and margin trend.
  • U.S. premium Android penetration.
  • Chinese OEM adoption.
  • Vivo / Oppo / Xiaomi flagship wins.
  • Smart Edge revenue share.7
  • Automotive cockpit design wins.7
  • Wi-Fi 8 / 5G NR-NTN / 10G PON traction.7
  • Cloud AI ASIC revenue.8
  • Customer concentration in ASIC projects.
  • GB10 / Nvidia collaboration follow-through.9
  • Qualcomm Oryon / Adreno response.10
  • OEM in-house silicon risk.
  • Smartphone unit demand.7

Glossary

A short reference for the vocabulary used above. Definitions are simplified.

Glossary
SoC
System-on-chip, combining CPU, GPU, modem, NPU, memory controllers, and other blocks.
Dimensity
MediaTek’s smartphone SoC brand.
Snapdragon
Qualcomm’s smartphone SoC brand.
TSMC
Taiwan Semiconductor Manufacturing Company, MediaTek’s key foundry partner.
Node
Manufacturing process generation such as 4nm, 3nm, or 2nm.
N2P
Enhanced TSMC 2nm-family process.
All Big Core
CPU design using only high-performance-class cores rather than traditional little efficiency cores.
NPU
Neural processing unit for AI workloads.
UFS 4.1
High-speed storage standard for mobile devices.
mmWave
High-frequency 5G spectrum used for very high-speed connectivity in some markets.
Smart Edge
MediaTek segment covering connectivity, computing, automotive, and other non-phone platforms.
ASIC
Application-specific integrated circuit, often custom-designed for a customer or workload.
Edge AI
AI running close to the user or device rather than entirely in the cloud.
ASP
Average selling price.

XVI. The Android flagship breakout

MediaTek’s 2021 TSMC 4nm flagship bet was not a one-off. It was the start of a structural climb.

By 2026, MediaTek has flagship Android credibility, early TSMC 2nm access, on-device AI ambitions, automotive cockpit traction, Smart Edge growth, and a real data-center ASIC story.

The question is no longer whether MediaTek can enter flagship. It is whether MediaTek can turn flagship credibility into durable premium pricing, ecosystem trust, and edge-to-cloud platform power.

Qualcomm still has a strong premium Android position. Apple is still its own world. Samsung still mixes its own silicon and merchant SoCs. Smartphone demand is weaker than it was. ASIC revenue is large but lumpy. None of those facts disappear because MediaTek shipped a great chip.

But the structural picture has changed. MediaTek climbed the TSMC ladder. MediaTek made architecture choices that flagship buyers respect. MediaTek made Smart Edge nearly as large as mobile. MediaTek won a real data-center ASIC engagement. And MediaTek co-designed a piece of an Nvidia AI compute platform. None of that was true in 2021.

That is the Android flagship breakout.


1 Patel, D. (Jul 2021). MediaTek Officially Enters The Flagship Mobile Race With The First TSMC 4nm SOC. SemiAnalysis. Historical anchor for the 2021 framing, including MediaTek’s cheap-chip stigma vs Qualcomm and Samsung, the Dimensity 1000 / 1100 / 1200 premium push, early TSMC 6nm use, the planned flagship 5G SoC on TSMC 4nm, the X2 + 3 A710 + 4 A510 CPU layout, RMB 4,000 / ~$615 flagship pricing target, mmWave and GPU gaps, and the framing of MediaTek’s special TSMC relationship. Used as inspiration only. No content, structure, or charts reproduced.

2 MediaTek (2025). Dimensity 9500 launch. Third-generation All Big Core CPU, 4.21GHz ultra core, performance and efficiency claims, Arm G1-Ultra GPU, NPU 990 with Generative AI Engine 2.0, BitNet 1.58-bit large-model support, UFS 4.1, AI model loading and long-context framing. Treated as MediaTek claims.

3 MediaTek (2024). Dimensity 9400 launch. TSMC 3nm process, second-generation All Big Core, Cortex-X925 / X4 / A720 layout, performance and efficiency claims versus Dimensity 9300.

4 MediaTek (2023). MediaTek + TSMC 3nm. First MediaTek flagship Dimensity using TSMC 3nm, volume production expected in 2024, TSMC’s claimed N3 vs N5 PPA framing.

5 MediaTek (2025). MediaTek + TSMC 2nm / N2P. Flagship SoC tape-out on TSMC enhanced N2P process, late 2026 expected production timing, N2P PPA framing versus N3E.

6 MediaTek. Quarterly earnings — Q4 2025. Q4 2025 revenue NT$150.188B, +5.7% QoQ, +8.8% YoY; Dimensity 9500 ramp framing; FY2025 revenue NT$595.966B, +12.3% YoY.

7 MediaTek. Quarterly earnings — Q1 2026 transcript. Mobile Phone revenue −17% QoQ, −15% YoY, ~49% of total revenue; global smartphone shipment ~15% decline expectation; Smart Edge +23% QoQ, +13% YoY, ~46% of total revenue; Wi-Fi 8, 5G NR-NTN, 10G PON, agentic-AI cockpit framing; first AI accelerator ASIC project ~$2B Q4 2026 framing; second AI accelerator ASIC mass production by end of 2027.

8 Reuters (Oct 2025). Taiwan’s MediaTek targets billions in AI accelerator chip revenue by 2027. MediaTek expected ~$1B in cloud AI chip revenue in 2026, multiple billions from first AI accelerator ASIC project in 2027, and a 10–15% share target of a ~$50B data-center ASIC TAM.

9 MediaTek (2025). Nvidia DGX Spark with GB10 Superchip co-designed by MediaTek. GB10 Grace Blackwell Superchip co-design framing, MediaTek power-efficient CPU / memory / high-speed interface design experience, DGX Spark with 128GB unified memory and up to 1 PFLOP FP4 AI performance.

10 Qualcomm (Sep 2025). Snapdragon 8 Elite Gen 5 launch. Third-generation Oryon CPU framing, claimed 20% CPU performance improvement, 23% GPU improvement, 37% faster Hexagon NPU performance, OEM support.

11 TSMC (2026). 2026 North America Technology Symposium. Leading-node roadmap, A13 / A12 / N2P positioning, AI / HPC / mobile relevance.

12 ASML (2025). 2025 Annual Report, strategic report section. AI requires leading-edge high-performance processors and a significant increase in DRAM relative to traditional compute architectures.

13 Market context (Counterpoint, Canalys, IDC, Omdia, or Gartner) is referenced in this essay only at the level the cited MediaTek, Qualcomm, and Reuters materials already disclose. Market-share figures are not stated unless attached to a verified source.

Further reading
*   *   *

This is Essay No. 034. The topics: intelligence, AI, systems, knowledge, and the questions underneath the questions everyone else is asking. If you read this far and disagreed with any part of it, write to me. I read everything.

Pugalenthi Magendran