RK3588: How the Strongest Domestic AIoT Chip Was Developed

1. RK3288: A Flagship Chip No One Wanted, Unexpectedly Found Its Way into AIoT

To understand how the RK3588 came to be, we should first look at where it “curved around and rolled out from”.

This journey begins with the fateful chip from 2014 — RK3288.

At that time, Rockchip was still fighting in the white-label tablet market, competing fiercely with Allwinner and MTK. Tablets were evolving rapidly, and whoever could cut a few dimes off the BOM could survive a little longer. Allwinner used Cortex-A7 for cost-effectiveness, MTK used 4G SoCs for a combination punch, while RK3288, a “true flagship” with Cortex-A17 + Mali-T764, was immediately rejected by the market: too expensive, too complex, not cost-effective.

Unable to compete in tablets, it was also tough in set-top boxes. Amlogic had already secured TV manufacturers and OTT platforms, with a stable ecosystem and competitive pricing. RK3288 found almost no place in the mainstream consumer market, and Rockchip was anxiously watching as its main market was squeezed out one by one.

But the industry has its quirks: you accidentally create a chip that no one wants, and it turns out to be the preferred choice in a niche market.

During those years, the explosion of commercial displays and large-screen interactive devices occurred. Various bus stop signs, electronic blackboards, interactive advertising screens, hospital guidance desks, and building terminals emerged, all requiring a main control chip that could run high-resolution screens, support dual displays, and had strong multimedia capabilities, all at a competitive price.

RK3288 happened to hit this point perfectly.

Its video capabilities, interface configuration, and local rendering capabilities were overkill for the tablet market but just right for these embedded scenarios. The key was that customers saw:this chip has impressive performance, is mass-produced, and has complete supporting resources. Thus, RK3288 established a foothold in the “commercial display” niche market, unexpectedly becoming a “high-performance SoC affordable for industrial use”.

Moreover, during those years, Google approached to promote Chromebooks, and RK3288 was packaged into several educational devices. Rockchip even sent engineers to Mountain View to work on the Veyron platform, collaborating on debugging Chromium OS and doing upstream work. Although the Chromebook line did not become a major success later, for Rockchip, this was the first time they realized that “SoCs can not only be rebranded but can also build ecosystems”.

Ultimately, RK3288 was not born for AIoT; it simply ended up in a place that others overlooked after being pushed back from the main battlefield.

But it was precisely in this gray area that Rockchip first hit the windfall and discovered:

Creating an SoC does not necessarily mean you have to win the consumer market; you can also become a necessity in a specific scenario.

This approach of “performance first + engineering adaptation + embedded longevity + development board testing” later became the foundation for the rise of RK3588.

2. RK3399: Google’s Endorsement, Intel’s Suppression, Ultimately Saved by Development Boards

After RK3288 made its initial mark in commercial displays and Chromebooks, Rockchip began to truly realize the potential of platform-type SoCs and further bet on RK3399.

This was Rockchip’s first SoC with “systematic ambition”: a big.LITTLE architecture with dual-core Cortex-A72 + quad-core A53, Mali-T864 GPU, built-in PCIe, USB 3.0, dual display outputs, Type-C, and high-definition video acceleration, packed with high-end configurations aimed not just at tablets but also at light office, edge computing, and even ARM PCs.

The key was,this time Google gave the nod.

RK3399 was selected by Google, branded as “OP1”, and became one of the core chips for Chromebooks. In 2017, Samsung launched the Chromebook Plus equipped with RK3399, a beautifully designed industrial device that supported 360° rotation, stylus input, and ran Android applications. This was the first time a domestic SoC officially appeared in an international brand’s flagship notebook.

At that moment, RK3399 shone brightly. But it didn’t last long.

Intel soon intervened.

While Google promoted Chromebooks, OEMs still had to heed Intel’s wishes. The ARM platform was already squeezing the low-end x86 business, and RK3399’s partnership with Samsung made Intel uneasy. Consequently, they exerted pressure at the OEM level, leading to the quiet termination of the ARM Chromebook project. Subsequent new models all reverted to Intel platforms, abruptly halting RK3399’s Chromebook journey.

With the Google connection severed, Rockchip was once again brought back to reality.

It wasn’t that they couldn’t compete with Intel; it was that they weren’t even at the table.

However, this time was different; Rockchip did not turn back to compete in tablets but gritted their teeth and continued to treat RK3399 as a “platform chip”. They steadily released SDK versions, supplemented documentation, and gradually provided mainline support — slow but steady.

Yet what truly resurrected RK3399 and brought it back into the developer’s view was not Google or any industrial client, but a small team called Rock Pi. Yes, the name includes Rock, and this small team had genuine faith in Rockchip.

In 2018, before the Raspberry Pi 4 was released, the development board market was dominated by the Raspberry Pi 3B+. At that time, whoever could offer USB 3.0, M.2 SSD, dual displays, and still keep the price reasonable could attract the geek traffic of those wanting to turn their Raspberry Pi into a server.

The Rock Pi team launched ROCK Pi 4 using RK3399 at this time, with all interfaces exposed, a compact board design, high openness, and a price low enough to make one question life.

For the first time, individual users could access “RK3399 performance” at “Raspberry Pi prices”.

ROCK Pi 4 exploded in the development board community, with benchmarks, open-source support, and tutorials all emerging. Armbian, Ubuntu, Debian, Android, Manjaro… various community systems were adapted, and RK3399 became a favorite among DIY enthusiasts and professional developers.

This wave of “wild ecology” directly elevated RK3399 to a competitive player in the domestic SoC ecosystem. Rockchip finally realized that building an ecosystem does not necessarily rely on big clients; it can also depend on communities, developers, and a group of “demanding, tinkering, and word-of-mouth spreading” tech enthusiasts.

From then on, Rockchip established the strategy for the subsequent RK series:

First sample with development boards, let the community tinker, run the ecosystem, and then replicate to industry clients.

RK3399 may not have brought Rockchip onto the Chromebook stage, but it thrived in seemingly “unsexy” scenarios like edge computing, industrial control, and AI terminals. It didn’t rely on press conferences or explosive points but on generations of engineers using it for projects, tuning devices, and mass production.

Did you say it “failed”? No, it was the most grounded generation.

And all of this laid the groundwork for RK3588.

The next chip would truly be a heavyweight contender.

3. The Launch of the RK3588 Project: The Inside Story of a “Heavyweight Platform Chip” Initiative

After RK3399, Rockchip’s engineers felt a subtle emotion — they were not incapable of making chips, but they were a bit late, a bit slow, and a bit frustrated.

It wasn’t that no one used RK3399, but it was often compared to Raspberry Pi, Qualcomm, and Intel. The performance was sufficient, and the ecosystem was emerging, but it always felt lacking: missing a definitive specification, missing an opportunity for a direct confrontation.

Thus,they decided to create a chip that could truly bridge the gap and connect the ecosystem — RK3588.

At the end of 2019, RK3588 was first revealed at Rockchip’s developer conference. At that time, RK1808 and RK3399Pro had just been launched, AI was hot, and edge computing was beginning to surface, but no one dared to place heavy bets on this line.

But Rockchip dared.

They defined the RK3588 project directly at the “flagship platform” level, targeting not a specific application but “eight-pronged attacks”:

• High-performance tablets
• ARM PCs
• Smart cockpits
• Multi-camera systems
• Smart NVRs
• Smart large screens / multi-screen applications
• Cloud-edge collaboration
• VR / AR

At first glance, these eight directions seem quite mixed. But the core message is:RK no longer wants to exist solely in niche markets; it aims to create a general-purpose SoC that can serve as a foundation.

To achieve this goal, RK3588’s material strategy was aggressive:

• The CPU uses a quad-core Cortex-A76 + quad-core A55, built on a 8nm process, directly outperforming the previous A72 + A53 combination.
• The GPU has been upgraded to Mali-G610, supporting OpenCL and Vulkan, along with a 6T NPU, making AI computing power no longer just a “can be used” setup but capable of running models and being truly production-ready.
• The video codec module is nearly industry-leading: it can decode 8K@60fps under H.265 / VP9, and 8K@30fps under H.264, capable of processing up to 16 channels of 1080p video streams simultaneously, fully sufficient for multi-camera and NVR scenarios.
• It supports dual MIPI CSI, dual HDMI, dual MIPI DSI, eDP, LVDS, dual Type-C, PCIe 3.0 x4, SATA3, USB3.1 Gen1… all high-speed interfaces available are included.

But the most audacious move was that RK knew this chip would be expensive, yet they decided to proceed.

At that time, the entire industry was shouting that “mid-range SoCs have no profit, and high-end ones cannot be made”, and the only sellable ones were those “good enough and cheap” SKUs. The initial design BOM for RK3588 clearly indicated it was not a cheap product.

To make matters worse,the pandemic had just begun, wafer foundry was tight, and chip costs skyrocketed.

The initial pricing for RK3588 was intended to be controlled around $35, targeting NXP’s high-performance industrial SoC. However, as wafer prices soared, the final retail price approached $60. Given Rockchip’s usual image as the “king of cost-performance”, this price could be considered “out of line”.

However, the project did not stop, nor were functionalities cut; instead, they kept piling on more.

Because RK was not customizing for a specific client, nor was it temporarily creating a SKU to catch the trend; they genuinely wanted to create a “strategic chip” that could be used for five years, produced in millions, and establish an ecological foundation.

They were not unaware of the high cost; rather, they wanted to try: “If we push the chip to the extreme, will it bring volume and ecosystem in return?”

This is the background of RK3588’s birth: it is built on the accumulation of the previous generation of chips and the frustration of being blocked by Intel and overlooked by the market —

We will not rely on trends; this time, we will fight a real battle based on strength.

Thus, a “domestically produced heavyweight platform chip” was truly initiated in the era of edge computing.

The next question is — who will be the first to use it? Who can bring it into the public eye?

4. From PPT to Desktop: How Various Pi Boards Exploded RK3588

The initiative for RK3588 was correct, and the specifications were fierce, but placing a $60 cost SoC on the table does not guarantee someone will dare to use it. Especially at that time, the pandemic was still ongoing, foundries were still lacking, and upstream prices were rising; the industry looked at the parameters of this chip and said “impressive”, but when they saw the price, they said “let’s wait and see”.

As a result, the first to rush out was not a solution company or an industry client, but those who care the least about ROI —the development board makers.

In 2022,the Rock Pi team launched Rock 5B, a super compact SBC (Single Board Computer) equipped with RK3588, which truly turned this SoC that had only appeared in developer conferences and PDFs into a “plug-and-play, fully equipped USB peripherals, M.2 slot” personal development board.

You might think development boards are common, as there are many Raspberry Pis. But Rock 5B is different.

It was not made as a sample to showcase SoC functions but was built to the standard of a “true product-level engineering board”:

• Its size is controlled to be as small as a Raspberry Pi, but the interfaces are fully loaded: dual HDMI, M.2 NVMe, PCIe x4, 2.5G Ethernet, USB3.1, RTC, Debug UART,everything that can be exposed is exposed.
• The board comes with SPI flash and eMMC slots,making system boot configuration developer-friendly.
• It supports Type-C power supply, PD fast charging protocol, and the board’s stability and power consumption scheduling are optimized well,not just “lighting up the screen counts as winning”, but truly capable of running long-term tasks.

More importantly —the Rock Pi team directly set the retail price of this board below $100.

At that time, RK3588’s chip cost alone was around $60; how could they dare to retail it? The Rock Pi team managed to push RK3588 into the sub-$100 market through tight BOM control and modular design.

RK3588 quickly entered the global developer discussion circle thanks to Rock 5B, from Reddit to Telegram, from benchmarks to system porting, the developer community formed a collective interest in this SoC for the first time.

Rock 5B brought three direct impacts:

1. RK3588 began to have a real developer ecosystem. Various Linux distributions began to adapt, the open-source community started creating firmware, documentation, and BSP debugging, transforming RK3588 from a SoC in PDF files to a practical platform with “real users, tutorials available, and someone answering your questions”.

2. Industry clients began to “bypass solution companies and first buy development boards” for project evaluation. Originally hesitant security manufacturers, edge box manufacturers, and industrial display manufacturers saw Rock 5B performing excellently, with quick system adaptation, and could prototype without needing to create a BSP, immediately began considering direct evaluation for integration.

3. The community became an amplifier for chip commercialization. The Rock Pi team did not rely on traditional agents for distribution but leveraged forums, GitHub, Telegram, and YouTube to maximize the influence of Rock 5B, directly boosting RK3588’s brand heat and attention.

RK3588’s engineering foundation was solid, and the popularity of Rock 5B allowed these foundations to reach a broader developer community for the first time.

Interestingly, after this wave of ignition by Rock Pi, Firefly and other Pi boards quickly followed, launching lower-priced versions of RK3588S boards, bringing the development board price directly down to the $60 range, making RK3588 the only “affordable high-performance SoC” of that era.

From then on, developers no longer only looked at Raspberry Pi, NXP, or TI for chips but began to say:

“RK3588? Just buy a board, burn a system, and it runs well, with good support.”

This is the result of Rockchip’s years of hard work on SDKs, documentation, upstream contributions, and demos, blossoming in various Pi boards.

It is not a flower nurtured by capital but one that has been burned, soldered, and emerged from the pitfalls.

5. The Engineering Battle Behind “Open Source Popularization”: From Wiring Barriers to Establishing an Industrial Collaborative Network

RK3588 is a “good chip”, but its “goodness” does not suit everyone.

The chip can run fast, IO can be fully utilized, video can decode to 8K, while PCIe, MIPI, SATA, USB3.1 interfaces are all fully loaded — this looks great in the datasheet, but in reality, it presents a problem:

You simply cannot lay it out.

When Rockchip initially developed RK3588, it was not aimed at mass production of development boards but was targeting “industrial main control + high-end custom projects”, so they never considered that someone would use it to create low-cost boards. Thus, the chip’s BGA packaging was tight, the wiring was complex, power consumption requirements were high, and thermal design requirements were stringent. The first version of the reference design directly specified the PCB standard as:

10-layer board, 2-stage HDI.

This is acceptable for traditional motherboard manufacturers, but for small and medium-sized solution providers accustomed to making 4-layer or 6-layer boards, this was a nightmare start.

Before cooling was laid out, the power supply was already conflicting, MIPI wiring caused crosstalk, and PCIe wiring was non-functional, leading to signal integrity and wiring process issues that discouraged many. Many clients received the chip not because they couldn’t run it, but because they couldn’t even prototype it.

What to do?

Rockchip did not say, “If you are incompetent, don’t blame me”; instead, they did something very uncharacteristic for a chip company —they organized training.

Collaborating with PCB design companies, HDI board manufacturers, and technical media, Rockchip began to conduct “wiring literacy” in the industry chain:

• Publishing complete multi-layer board reference designs and providing reusable modular design guidelines;
• Conducting live teaching and offline training sessions, turning RK3588’s wiring and power design experience into courses for solution companies;
• Bridging resources for HDI board manufacturers that can prototype with just one piece, allowing companies that originally could not afford 10-layer boards to quickly get started with board verification;
• Establishing an internal FAE network to assist solution companies in reviewing board layouts, tuning hardware, selecting components, and often even helping to produce a demo board.

In short,they did not raise the bar to “filter clients” but actively lowered the bar to “support the ecosystem”.

The core idea of this entire collaborative system can be summarized in one sentence:

This is not just a battle for our chip company; it is a collective bet on whether we can bring a domestic platform out of the industry circle.

This is the key to RK3588 truly transitioning from “usable” to “implementable”. You might think Rock 5B became popular, but behind it is a whole engineering service network supporting it:

• PCB processes have been demystified;
• Power design has been standardized;
• BSP and SDK have been tested by the community;
• Board manufacturers are willing to prototype, and engineers are willing to assist;
• Industrial clients, AI clients, and commercial display clients first ran demos on Rock 5B before transitioning to solutions.

This promotion method does not rely on “channels” or “agents” but on a complete “top-down engineering collaborative ecosystem”.

We have seen too many domestic SoCs die in the pitfalls of “boards not being able to be designed, documentation lacking, and manufacturers not supporting”. RK3588 did not avoid these pitfalls; it walked through them and emerged while fixing issues along the way.

This is the most authentic path to popularizing domestic chips: it is not a market built on slogans but an engineering trust built on every layer of the board, every line of code, and every manual.

Because of this layer of “technical foundation”,RK3588 has achieved real shipments in industrial markets, AI boxes, edge computing, and smart cockpits.

From Rock Pi to other Pi boards, to various custom modules, to vehicle surround control, RK3588’s volume is built on “engineering capability popularization”, not on “PPT performance” hype.

This is the most touching part of the Chinese chip industry:it is not about winning on heat but about surviving when the heat does not kill you, while others are still building infrastructure.

6. Ecosystem and Rhythm: Why RK3588 is Irreplaceable in the Short Term

In the chip industry, performance is not the only factor, and it may not even be the most important.The true determinant of an SoC’s fate is its ecosystem.

This statement may sound trivial, but when applied to RK3588, it becomes extremely relevant.

You say it has strong performance? Indeed, with a quad-core A76 + quad-core A55, equipped with G610 GPU, 6T NPU, it runs edge AI without hesitation, decodes 8K without gasping, and pulls interfaces to the point of driving hardware engineers crazy.

But if this chip only had performance and no ecosystem, it would have long been lying in some factory’s datasheet, yellowing.

Yet RK3588 has survived, and it has become the “first choice for developers” in the domestic AIoT field.

Why? It is not because it has some secret weapon, but because Rockchip has been doing one “invisible but crucial thing” for ten years since RK3288:

Fostering a developer ecosystem, making each generation of chips into a platform that can be used.

While other chip manufacturers release SoCs, they first ask about market positioning, target customers, and shipment rhythm; Rockchip, when releasing SoCs, first checks whether the SDK is complete, whether it has gone upstream, whether it can run Debian, and whether developers can find drivers.

Thus, you will see that as soon as RK3588 was launched, the community immediately had:

• Multiple distributions like Armbian, Debian, Ubuntu, Android TV, Manjaro adapted.
• A complete SDK release mechanism, with kernel, drivers, and BSP organized, not relying on “manual group sending firmware”;
• Documentation systems that keep up, not only datasheets but also development guides, IO allocation diagrams, reference board layouts;
• Active updates on Github / Gitee, with issues being addressed and pull requests being genuinely merged, buildroot being gradually built up;
• A large number of hardware support modules that have already “been through the pits”, with USB3.1 / PCIe / MIPI CSI / SATA running stably, not just “able to power on” but “able to go into projects”.

All of this ecosystem feedback is backed by Rockchip’s years of “non-standard efforts” in interacting with the community:

• Starting from RK3399, a large investment in upstream submissions;
• Collaborating with domestic distributions (like FydeOS, KylinOS) to provide experience optimizations;
• Investing in documentation systems, even co-writing routines with development board manufacturers;
• Pairing with FAE systems to guide solution providers and industrial clients in prototyping;
• Providing substantial support to some core developers and community maintainers.

While other chip manufacturers are still focused on “finding a big client for mass production”, Rockchip has built a foundational capability that allows anyone to get started with development within three days, even if they have never seen it before.

This is RK3588’s true moat —

It is not about leading in performance, but about the stability of “engineers being able to find boards, documentation, code, and tutorials when they want to use it”.

Once this ecosystem starts running, it will form a flywheel:

• The more developers use it, the more feedback there is;
• The more feedback there is, the more stable the SDK becomes;
• The more stable the SDK is, the more clients dare to use it;
• The more clients dare to use it, the more shipments there are;
• The more shipments there are, the more development board manufacturers and module vendors are willing to promote it;
• The more it is promoted, the higher the community heat, and developers come back again…

This is an ecosystem-level positive feedback loop, not a pretty number built on one or two big projects.

“easiest to use” SoC with the best performance.

You can make an educational machine today, build a robot tomorrow, and switch to a multi-camera box the day after; the community experience, underlying compatibility, and board-level products of RK3588 are all there,it has become a kind of “general capability”, rather than a customized solution.

This kind of chip is not produced every year, nor can every company make it.

It is a true landing of platform-type SoCs, not a marketing illusion of “universal SoCs”.

While RK3588 may not be revolutionary in architecture, in the history of domestic SoC ecosystem construction, it is a rare victory that has truly been “used by developers”.

7. A Victory Against the Odds: 4 Million Units Shipped and 1.3 Billion Revenue in 2024

By the end of 2024, a set of data quietly circulated in the domestic SoC circle:

RK3588 series shipped over 4 million units in the year, with cumulative revenue exceeding 1.3 billion RMB.

In an industry where everyone is used to “shipping without profit, and profit without shipping”, this data is like a splash in a muddy pond — not a huge wave, but it catches the eye.

You must understand that the domestic chip industry seems lively in recent years, with financing in the billions, news flooding in, and more shouting slogans than producing products. But to achieve:

• High-performance architecture
• Mid-range chip pricing
• Multi-industry adaptation
• Open ecosystem leadership
• No major client binding
• And still achieve millions in annual shipments, truly returning cash flow

Very few can achieve this. Yet RK3588 did.

Moreover, what is even more challenging is —it did not rely on a single client for explosive volume but gradually rolled out from eight directions.

Looking back at the application paths publicly announced by RK3588: tablets, ARM PCs, smart cockpits, multi-camera systems, smart NVRs, smart large screens, edge computing, VR/AR — these directions sound quite “general”, but in reality, it means that none of them can earn money easily; each relies on penetrating one industry at a time.

Some started with development boards, some with embedded modules, some solution companies took a year to evaluate before launching, and some clients verified in small batches for two generations before scaling up —none of these orders were easy to earn.

But precisely because of this, RK3588’s revenue istruly diversified, truly stable, and genuinely market-derived.

1.3 billion RMB does not rely on subsidies, “strategic cooperation”, or collective procurement; it relies on development boards like Rock 5B shipping hundreds of thousands of units, on 5,000 solution companies across the country pulling hundreds or thousands of units each month, on batches of educational terminals entering campuses, on edge computing landing one by one, and on vehicle surround systems gradually passing certification.

This is the true business model that allows domestic chips to “sustain themselves”:

It is not about creating a hit product and exiting, but about maintaining stable revenue over three years, with clients renewing orders, and accumulating developer reputation, until you look back and realize: oh, we have already shipped 4 million units.

The success of RK3588 also brings a more realistic industrial consequence:

• More module manufacturers are willing to develop around it;
• More motherboard manufacturers are embedding it as a “usable platform” in their product lines;
• More clients are willing to “upgrade from RK3399” rather than restart evaluations;
• Rockchip can also develop a family of platform-type chip products centered around RK3588 (like RK3576, RK3582, RK3588J)…

It has become a true “chip cornerstone that can be commercially produced and has an inheritable ecosystem”.

Many domestic chips rely on financing to survive, on custom orders to stay alive, and on subsidies to inflate scale. RK3588 is the kind thattruly occupies every small scene in desktops, displays, cameras, and main control boards, gradually piecing together an industrial map.

It is not about winning once in a trend but about creating a closed loop.

This round, RK got it right and held on. Next, it will be up to the entire industry to decide whether it can continue to run.

However, when a chip sits on the table for too long, it becomes the target everyone wants to move off.

8. But the Jianghu is not as Stable as You Think: MTK, Qualcomm, and Domestic Newcomers Enter the Arena

In April 2024, a small YouTube video unexpectedly pulled Rockchip into an international risk whirlpool.

The video content was not complicated: a drone wreckage shot down in Ukraine revealed a certain domestic development board, which was based on RK3588S. This development board was not shipped by Rockchip but was a product of a certain manufacturer, yet this incident quickly sparked speculation in the European and American developer circles:Does Rockchip pose a military risk? Is it involved with Russia? Does it violate export controls?

For a domestic fabless company positioned on open-source, civilian use, and independent research and development, this kind of “borderline” incident is tantamount to stepping on the edge of an international red line. Even if there is no subjective involvement, and even if the chip is a publicly sold commercial product, if the situation escalates, it could be pulled into the “entity list” blacklist.

Rockchip’s response was very swift —they immediately announced the termination of cooperation with the manufacturer and prohibited them from continuing to purchase Rockchip chips.

This move received mixed opinions in the industry, but it conveyed a very clear signal:

Rockchip knows this matter cannot be gambled with, cannot be delayed, and must be cut off quickly to stop losses.

Before this crisis had subsided, in April 2025, another wave of more bizarre winds blew in.

Multiple industry self-media began to spread wildly: “Samsung’s wafer foundry will completely halt its foundry business for Chinese customers”. This news had neither source nor details, yet it spread vividly in WeChat groups and semiconductor circles, causing panic.

Then came a scene that made all chip people feel “a bit absurd” —Rockchip personally stood up to refute the rumors about Samsung.

“Today, there are rumors in the market claiming that ‘Samsung’s wafer foundry has suspended all business with China’. After confirming with Samsung’s wafer foundry, this is false news… All products that Rockchip cooperates with Samsung are progressing normally.”

As a fabless company, Rockchip should not have to clarify for a foundry, especially since this rumor did not directly attack them.

So why the urgency?

The reason is simple:they are indeed heavily reliant on Samsung for some high-end product foundries, especially for flagship chips like RK3588 that use 8nm processes.

This reaction actually exposed three layers of reality:

1. Rockchip’s high-end products are seriously dependent on Samsung’s wafer foundry, especially for flagship chips like RK3588 that use 8nm processes;
2. Competitors may be “leaking information” behind the scenes, creating customer panic and shifting market trust.
3. Rockchip understands that this is not a matter of “just talking it over”; it could directly shake customer confidence, investment judgments, and the rhythm of industry cooperation.

In short,the fire has spread from the market to the edge of the supply chain and geopolitical issues.

RK3588 is no longer an ordinary SoC; it has become the ecological cornerstone of the domestic AIoT track, as well as a “target chip” for foreign scrutiny, peer monitoring, and upstream and downstream bets.

And the real siege is happening simultaneously.

International Giants Closing In: MTK, Qualcomm, Combining Ecosystem and Compliance to Strike

First, let’s talk about MTK.

MediaTek launched the Genio series, specifically targeting RK3588. Especially the G1200, which focuses on high-performance edge computing and industrial AI scenarios, with performance specifications that do not fall short of RK, and it has directly collaborated with Ubuntu for the official distribution, inviting open-source giant BayLibre to refine the Linux SDK.

That lineup, to be honest, is not inferior to RK.

Moreover, Taiwanese industrial control manufacturers tend to use MTK, having project experience and certification channels, leading many to feel that RK’s position might be at risk.

However, to be honest,MTK’s biggest problem is not the product but the pricing.

The G1200, with such specifications, dares to sell for $80, and when you add support costs, reference designs, and solution services, the total is significantly more expensive than RK3588. In the hands of domestic AIoT clients seeking “plug-and-play, mass production”, this essentially gets labeled as a “project-type chip”, making it difficult to adopt RK’s lightweight approach of “universal, small batches, development boards first”.

Now let’s talk about Qualcomm.

Qualcomm has created a series called Dragonwing, specifically targeting AIoT, with complete SDKs, LE Linux without walls, comprehensive documentation, and well-made development boards.

It sounds like “American Rockchip”, and even the release rhythm is becoming increasingly similar.

However, there are three issues:

• High entry barriers, as truly running Qualcomm chips requires familiarity with a not-so-friendly toolchain, boot flow, and system structure; new developers may take three days to get accustomed, and project managers may lose weight;
• Commercial support relies entirely on contracts or fees, while community users can use it, if problems arise, no one will respond, which is not at the same level as RK’s “emails are answered, QQ groups have responses”;
• The ecosystem inertia has not been established, RK built its momentum through development boards + forums + open-source kits, while Qualcomm, although more standardized, is still not “hot” or “sticky” enough.

Thus, while MTK and Qualcomm are strong competitors, they currently pose more of a threat to RK’s “strategic space” rather than directly shaking its main market territory.

The Dilemma of Challengers: Rolling Out Specifications Does Not Equal Rolling Out Ecosystems

Domestic Players Charge Forward

Of course, what truly puts pressure on RK3588 is the collective charge of both new and old domestic players.

In recent years, Allwinner, SigmaStar, Amlogic, Realtek, Synaptics have all ramped up efforts in ARM application SoCs, with high-performance AIoT chips flooding the market:

• Allwinner focuses on cost advantages + voice control scenarios;
• SigmaStar targets multi-channel video input, aiming at dedicated security main control;
• Amlogic has long been unwilling to just make boxes and is also pushing towards the edge;
• New players like This Chip and Horizon have launched new platform chips with ARMv9 architecture and NPU acceleration, directly targeting AI application scenarios.

These chips have some impressive specifications and competitive pricing. But the problem is —just making a chip does not mean the product will roll out.

If the ecosystem is not established, the SDK is incomplete, the boards are difficult to design, the documentation is not open, and developers face crashes every time they use it. The most fatal issue is: no one will use it a second time. You lack repeat purchases, tutorials, issue follow-ups, and GitHub submissions. You can run benchmarks quickly, but projects die just as fast.

This is like engaging RK in a protracted battle while only bringing the supplies for a blitzkrieg.

So Will RK3588 Be Ousted? Yes, but Not Now

RK’s ecological moat is not built on any single project, client, or technical advantage, but on:

• Development board manufacturers being trustworthy;
• Embedded engineers being able to find resources;
• System distributions adapting quickly;
• Solution companies finding it easy to tune;
• Client support keeping pace.

It is built by thousands of small projects “using it out”, not by a single super project “blowing it up”.

Thus, it is not irreplaceable, but to replace it, you must first start from RK3288 and spend ten years.

This is an issue of ecological inertia, not something that can be solved by press conferences, PPTs, and benchmarks.

In this long-term battle interwoven with technical density, supply chain risks, geopolitical issues, and developer communities —

RK3588 may not be alone anymore, but it still stands at the center of the domestic SoC ecosystem.

9. In Conclusion: This is Not the Strongest SoC, but It Represents a Way of Life

Is RK3588 the strongest SoC? Perhaps not. But it lives most like a “usable Chinese chip”.

When you compare it to Apple’s M series, it lags behind in an entire ecological generation; compared to Qualcomm, it falls short in IP stacking and brand halo; compared to MTK, it lacks in process stack and capital scale.

But it wins in one thing:

It is the first chip made in China that can be sold, supported by a community, trusted by clients for mass production, repeatable in projects, and can run in desktops and vehicles.

It does not crush any single parameter, but every single metric isjust right, and you can truly use it.

You are not limited to using it; it is the most convenient to use.

More importantly, it did not succeed by luck. It rolled down from RK3288, from no one daring to design boards to boards not needing to be designed, from no one writing drivers to today where GitHub has a pile of issues being answered, from “where is this chip’s documentation” to “just search the forum for this problem”.

The path to RK3588’s success is a standard, myth-free, hard-working approach to domestic chips.

It is not a smooth ride; it has experienced “not being able to sell, not being able to scale, and not being able to cut costs”, faced “clients not using it, channels not recognizing it, and engineers not trusting it”, and has also been pressed by Raspberry Pi, kicked out of Chromebooks, chased by external rumors, and had competitors leaking information behind its back, all while skirting the edge of international export regulations.

It is not perfect; it has survived due to the accumulation of SDKs, FAEs, and demo boards over the years, bit by bit.

And its next generation is already brewing.

In 2025, RK has already announced the next generation of automotive-grade flagship SoC — RK3688M, targeting high-end smart cockpits:

• Using 4nm process
• CPU performance reaching 300K DMIPS
• GPU performance reaching 2 TFLOPS
• NPU computing power reaching 32 TOPS
• Supporting up to 12 screens for simultaneous output
• Supporting 6 channels of 4K resolution for parallel display
• Scheduled for official launch in 2026

This RK3688M will no longer follow the “grassroots army development board first” approach but will take the high-end automotive route, targeting pre-installed and customized for car manufacturers, aiming directly at the main battlefield of Qualcomm SA8295P and MTK Dimensity Auto levels.

The ecosystem construction, process platform, software middleware, certification processes, and client systems behind it will no longer follow the old RK3288 logic of “let’s see if it can be used” but will be a complete, frontal, heavyweight breakthrough for domestic chips.

Will it succeed? No one can say for sure now.

But if RK3688M truly achieves automotive-grade mass production, pre-installation, and a smooth ecosystem in 2026, it will undoubtedly walk the path paved by RK3588.

Without RK3588, the term “platform chip” would not exist in domestic SoCs;

Without Rock 5B, Orange Pi, those forum posts from developers, those makeshift solutions on prototypes, and those last-minute fixes to SDK issues, RK would not be able to claim to be a “chip manufacturer with an ecosystem”.

So, is it the strongest? It doesn’t matter.

What matters is that it represents a way of life —

not a flagship dream built on financing, but a survival inertia built by generations of engineers.

And in this era, being able to survive is already a victory.

This is not the end, but the true starting point for the domestic SoC ecosystem.