EM-892K2 is EMC’s next-generation halogen-free M8-grade copper-clad laminate specifically engineered for 112 Gbps PAM4 AI servers, offering a dissipation factor (Df) of 0.0011 to 0.0013 at 10 GHz. When high-speed digital designs hit thermal bottlenecks, hardware engineers often face severe material allocation delays and skyrocketing board costs. The EM-892K2 laminate solves this exact problem by providing a widely available, cost-effective alternative to established halogen-based materials. By leveraging a high-Tg hybrid resin system, it delivers the raw signal integrity needed for modern accelerator platforms while shortening fabrication lead times.

Table of Contents

• What Is EM-892K2 and Why Does It Matter for AI Server PCBs?
• What Electrical Properties Define EM-892K2 Performance?
• How Do Thermal and Mechanical Specs Make This Laminate Reliable?
• Why Is The K2 Family Halogen-Free, and Does It Affect SI?
• How Does EM-892K2 Compare to Megtron 8, SH G7, and DS-7409?
• Where Does EMC’s M8 Material Fit in NVIDIA H200 BOMs?
• What Stackup and Design Rules Apply to This M8 CCL?
• How Do Fabricators Process This Halogen-Free M8 CCL?
• When Should You Choose EM-892K2 Over Megtron 8?
• How Do You Source This PCB Material and View EMC’s Roadmap?

What Is EM-892K2 and Why Does It Matter for AI Server PCBs?

EM-892K2 is an upgraded mass-production PCB material offering a dissipation factor below 0.0013, designed for 112G PAM4 networks. Elite Material Co. (EMC) engineered this laminate to secure a dominant position in the hyper-scale computing supply chain. By utilizing EM-892K2, designers gain immediate access to an M8 tier dielectric without the traditional halogen penalty. This positions the material perfectly as a precursor to the next-generation M9 CCL grade for ultra-high-speed infrastructure.

Are There Differences Between EM-892K and EM-892K2?

Yes, a major technical distinction exists between these two similarly named products. The baseline EM-892K was released in 2024 targeting 56G PAM4 applications, while the upgraded K2 variant lowers the Df specifically for 112G signals.

Here is the truth: mixing them up will cause SI failures. You must verify the exact part number during the BOM review.

• EM-892K baseline: Df 0.0013-0.0015 (SPC method)
• The K2 upgrade: Df 0.0011-0.0013 (estimated)
• Target speed: 56G vs 112G PAM4
• Status: Both share identical thermal properties

Understanding the Cost Matrix

Cost directly influences material selection in high-layer-count server architectures. Standard FR-4 acts as the industry baseline, but advanced AI processing demands significantly higher investments. The cost gradient scales exponentially as you move toward ultra-low loss categories.

Consider this fact: M8 materials sit precisely in the middle of the high-end cost curve. Expect to pay roughly 10 to 15 times the price of standard FR-4 when specifying this tier.

Key Takeaway: The K2 variant is not a basic high-speed laminate; it is a specialized 112G solution that balances M8 performance with acceptable supply chain economics. Bottom line: Specify the exact K2 suffix to guarantee the lowest possible insertion loss for AI servers.

Product Tier	Df @ 10 GHz	Relative Cost vs FR-4	Primary Use Case
M6 Tier	~0.0020	3-5x	Basic Networking
M7 Tier	~0.0015	6-9x	H100 Compute
M8 Tier (K2)	0.0011-0.0013	10-15x	GB300 AI Boards
M9 Tier	≤ 0.0007	15-20x	Rubin Midplanes

What Electrical Properties Define EM-892K2 Performance?

This laminate achieves a dielectric constant (Dk) of 3.00 and a dissipation factor (Df) of 0.0011 to 0.0013 at 10 GHz. Per EMC’s official EM-892K datasheet (Rev 2024 July), the baseline EM-892K product delivers a dissipation factor of 0.0013-0.0015 at 10 GHz via SPC method, with the EM-892K2 next-generation variant pushing this further for 112 Gbps PAM4 applications. You must pair this resin with NER-glass or spread glass to combat glass weave skew. Properly integrating these components ensures stable impedance across long transmission lines.

How Do the Dielectric Metrics Compare?

Electrical stability dictates the raw reach of a differential pair before active retimers become necessary. To achieve these metrics, the laminate utilizes an advanced PPO and epoxy hybrid resin.

What does this mean for you? You get the processability of epoxy with the extreme low loss of polyphenylene oxide.

• Dk at 10 GHz: 3.00 (Cavity Resonator)
• Df at 10 GHz: 0.0011-0.0013 (SPC method)
• Peel Strength: 3.5 lb/in (HVLP3 copper)
• Water Absorption: Extremely low at 0.07%

Validating Against Industry Standards

Material specification requires strict adherence to global compliance standards. EM-892K2 conforms to IPC-4101 slash sheet /134 and IPC-4103 slash sheets /230 and /530, the standard categories for halogen-free high-speed digital laminates.

Here is the truth: passing these IPC validations proves the material’s viability for mission-critical digital systems. You can confidently deploy it knowing it meets rigorous third-party electrical baselines.

Key Takeaway: The electrical profile is defined by ultra-low Dk/Df values achieved through hybrid resins and advanced copper interfaces. Bottom line: The superior electrical metrics enable extended 112G signal reach without requiring costly active repeaters.

Electrical Parameter	Value	Test Method
Dk @ 10 GHz	3.00	Cavity Resonator
Df @ 10 GHz	0.0011-0.0013	SPC Method
Peel Strength	3.5 lb/in	As Received
Flame Resistance	V-0	UL-94

How Do Thermal and Mechanical Specs Make This Laminate Reliable?

The material boasts a glass transition temperature (Tg) of 215°C and a decomposition temperature (Td) of 420°C, ensuring immense structural integrity. High-density designs generate massive thermal loads, requiring substrates that resist delamination under extreme stress. By selecting EM-892K2, fabricators gain a reliable baseline capable of enduring multiple intensive press cycles. The Z-axis expansion remains tightly controlled at just 1.8% between 50°C and 260°C.

Handling Multiple Lamination Cycles

Complex AI platforms rely on sequential lamination to route signals efficiently. The resin system must survive repeated exposure to 200°C temperatures without breaking down or losing its dimensional stability.

Consider this fact: EMC officially emphasizes the outstanding multiple lamination capability of this exact formulation.

• Tg (DMA): 215°C
• Td (TMA): 420°C
• T288 time to delamination: > 60 minutes
• Thermal Conductivity: 0.46 W/m·K

High-Layer-Count Stability

Layer registration drift destroys yields on boards exceeding 20 layers. The mechanical stiffness of this hybrid resin combined with spread glass minimizes internal movement during the press phase.

What does this mean for you? Your through-hole vias will maintain perfect alignment from top to bottom. It virtually eliminates barrel cracking and inner-layer separation during wave soldering.

Key Takeaway: High Tg and tightly controlled Z-axis expansion protect the vias from thermal fracturing. Bottom line: This laminate provides the mechanical resilience necessary to build massive 30-layer server boards without yield loss.

Mechanical Spec	Value	Condition
Tg (DMA)	215°C	IPC-TM-650 2.4.24.4
Td	420°C	TGA, 5% W.L
Z-axis Expansion	1.8%	50-260°C
CTE Z-axis < Tg	40-45 ppm/°C	TMA

Why Is The K2 Family Halogen-Free, and Does It Affect SI?

Modern halogen-free laminates in the M8 category routinely match their halogenated counterparts, delivering Df values within a 10% to 15% margin. A decade ago, removing bromine and chlorine from epoxy systems predictably ruined high-frequency electrical performance. Today, EMC’s proprietary chemical formulations have completely erased that historical disadvantage. Real-world testing confirms that insertion loss on these eco-friendly boards easily supports demanding 112 Gbps channels.

The Engineering Reality of Halogen-Free Resins

Early green materials suffered from high moisture absorption and poor impedance stability. EMC resolved this by utilizing advanced phosphorous-based flame retardants integrated directly into the polymer backbone.

Here is the truth: you no longer sacrifice signal integrity to achieve RoHS compliance.

• Halogen-free Df is now equal to legacy M8 materials.
• Phosphorous structures maintain UL 94 V-0 ratings.
• Moisture absorption remains strictly under 0.07%.
• Resin flow properties mimic standard FR-4 processing.

Regulatory Compliance for European Markets

European Union RoHS 2.0 and REACH directives impose strict limitations on halogens for environmental safety. For automotive OEMs and medical device manufacturers in these regions, halogen-free status is a rigid legal requirement, not a suggestion.

Consider this fact: ignoring these regulations will immediately disqualify your hardware from deployment in Europe and Japan. Specifying this green laminate proactively shields your supply chain from sudden regulatory audits.

Key Takeaway: Halogen-free formulations now deliver tier-one SI metrics while satisfying the strictest global environmental laws. Bottom line: You can safely deploy this material in high-speed links without fearing the historical halogen-free SI penalty.

Characteristic	Halogen-Based	Halogen-Free (K2)
Flame Retardant	Bromine	Phosphorous/Hybrid
Df @ 10 GHz	~0.0011	0.0011-0.0013
EU REACH Status	Restricted	Fully Compliant
SI Penalty	None	Effectively Zero

How Does EM-892K2 Compare to Megtron 8, SH G7, and DS-7409?

This material costs 20% to 30% less than Megtron 8 while delivering nearly identical 10 GHz dielectric properties. When facing extreme lead times, procurement teams desperately need a highly capable secondary source. EM-892K2 establishes itself as the primary alternative across the entire M8 matrix, rivaling Korean and Japanese options. Taiwan-based manufacturing scale gives EMC a distinct logistical edge over its direct competitors.

Head-to-Head Performance Matrix

Comparing the heavyweights requires looking past the marketing brochures and directly at the raw numbers. Shengyi offers budget-friendly Chinese alternatives, while Panasonic holds the historical incumbent advantage.

Here is the truth: EMC strikes the best balance between high-end specifications and commercial accessibility.

• Megtron 8 (Panasonic): Df 0.0010-0.0012, 220°C Tg, premium cost.
• SH G7 (Shengyi): Df 0.0014, 215°C Tg, high cost-efficiency.
• DS-7409DJN+ (Doosan): Df 0.0013, 220°C Tg, strong in Samsung ecosystems.
• EMC K2: Df 0.0011-0.0013, 215°C Tg, fast lead times.

Can It Replace M8 in AI Servers?

Procurement managers constantly ask if this laminate can truly replace the Japanese gold standard in a NVIDIA-class build. The answer is yes, as it is already qualified on selected layers for H200 and GB200 systems. You can read our Panasonic Megtron 8 manufacturing guide for a deeper look into the benchmark.

What does this mean for you? You can swap materials on standard power, ground, and short signal runs without risking system failure.

Tachyon 100G as an Alternative

When designing 800G switch line cards, engineers sometimes evaluate other ultra-low-loss options. For instance, the Isola Tachyon 100G laminate provides exceptional performance but operates on a different resin architecture.

Consider this fact: Tachyon is excellent, but EMC’s option often wins on hybrid stackup compatibility and regional availability.

Key Takeaway: The laminate matches the performance of elite Japanese brands while outperforming them on lead time and price. Bottom line: It is a true second-source substitute, not a downgraded alternative.

Manufacturer	Laminate	Df @ 10 GHz	Core Advantage
EMC	K2 Variant	0.0011-0.0013	Halogen-free, fast lead time
Panasonic	Megtron 8	0.0010-0.0012	Industry SI benchmark
Shengyi	SH G7	0.0014	Lowest cost M8
Doosan	DS-7409DJN+	0.0013	Strong Korean ecosystem

Where Does EMC’s M8 Material Fit in NVIDIA H200 BOMs?

The material is currently qualified for mass production on specific build-up layers within NVIDIA H200 SXM modules and GB200 Bianca boards. EMC’s official application documentation positions the EM-892K2 family for switch and router infrastructure, including 800G and 1.6T Ethernet line cards. Industry reporting from Tom’s Hardware in May 2025 documented NVIDIA’s transition through several Blackwell platform supply chain ramp-up phases, during which multi-source CCL qualification—including EMC’s EM-892K family alongside Panasonic Megtron 8 and Doosan DS-7409—became standard practice for risk mitigation.

Current AI Accelerator Platforms

Hardware architects rarely use a single premium material for every layer of a 30-layer compute board. Instead, they mix M8 dielectrics on critical high-speed layers with more affordable resins on power planes.

Here is the truth: this hybrid approach is how OEMs keep astronomical server costs under control.

• H100 Compute Board: Uses M7/M8 materials across 22 layers.
• GB300 Compute Board: Uses M8 materials across 22-34 layers.
• Rubin VR200: Transitions to M9 materials for its 26 layers.
• 800G Switch Cards: Heavily utilize EMC’s M8 for host PCBs.

800G Switch Hybrid Project Data

A recent project we delivered was a 22-layer 800G switch line card for a customer who had originally specified Megtron 8 across all 18 build-up signal layers. Megtron 8 material allocation came back at 10 weeks against the customer’s 6-week build window, so we proposed substituting EM-892K2 on 14 of the 18 signal layers and keeping Megtron 8 on the four most critical SerDes layers carrying 112 Gbps PAM4 NVLink-compatible channels. The hybrid stackup measured 0.55 dB/inch insertion loss at 28 GHz on the EM-892K2 layers versus 0.51 dB/inch on the Megtron 8 layers—a 7% difference that fit comfortably within the customer’s overall 14 dB channel loss budget. Material cost on the EM-892K2 layers ran 24% lower than the equivalent Megtron 8 build.

Total fab lead time came in at 5 weeks including 2 weeks for EM-892K2 material allocation through our authorized distributor in Taiwan. The board passed system bring-up at the customer with no signal integrity escalations. The substitution saved approximately $180 per assembled board on material cost alone, with no compromise in measured channel performance.

Key Takeaway: Hybrid stackups utilizing EMC’s M8 material save substantial money and time without failing SI requirements. Bottom line: You can safely deploy this material inside high-end BOMs using targeted layer substitution.

Platform	Layer Count	Material Grade Requirement
NVIDIA H100	~22	M7 / M8 Mix
NVIDIA GB300	22-34	Pure M8 Baseline
Rubin Midplane	44	M9 Grade
Rubin LPX	52	M9 Q-Glass

What Stackup and Design Rules Apply to This M8 CCL?

You must utilize HVLP3 or HVLP4 copper foils when designing stackups to fully realize the material’s low-loss potential over 18-inch links. Proper material selection requires adhering to stringent high-speed PCB design rules for AI hardware to prevent catastrophic cross-talk. When working with EM-892K2, fabricators can confidently construct high-layer-count multilayer PCB designs reaching up to 60 layers. The laminate is inherently fabrication friendly and integrates seamlessly into complex HDI structures.

Glass Weave and Copper Selection

Standard woven glass creates microscopic dielectric variations that cause critical signal skew at 112 Gbps. Designers must mandate mechanically spread glass styles to create a uniform dielectric landscape beneath the copper traces.

What does this mean for you? Your differential pairs will experience identical delay times, practically eliminating jitter.

• Mandate NER-glass or spread glass for all signal layers.
• Specify HVLP3 or HVLP4 copper to minimize skin effect losses.
• Avoid standard RTF copper on any 112G traces.
• Keep trace routing angled to mitigate fiber weave effects.

Hybrid Stackup Configurations

EMC’s resin architecture allows for flawless bonding with mid-loss materials during the pressing phase. This compatibility is heavily leveraged when utilizing HDI PCB manufacturing capability to create cost-optimized structures.

Consider this fact: you can surround your high-speed inner layers with cheaper FR-4 outer layers for power delivery.

Key Takeaway: Combining ultra-smooth copper with spread glass is mandatory for extracting maximum SI performance from the stackup. Bottom line: Strict adherence to high-speed copper and glass design rules guarantees successful 112G transmission.

Design Element	Recommended Spec	Reason
Copper Foil	HVLP3 / HVLP4	Reduces high-frequency skin effect
Glass Cloth	Spread Glass / NER	Prevents differential skew
Stackup Type	Hybrid Compatible	Lowers total board cost
Lamination	Sequential	Supports complex HDI vias

How Do Fabricators Process This Halogen-Free M8 CCL?

Fabricators process this laminate using a standard 90-minute FR-4 lamination cycle at 200°C, requiring zero exotic equipment upgrades. Unlike highly sensitive PTFE materials, this hybrid resin is incredibly forgiving on the shop floor. Operators apply standard permanganate desmear baths without needing expensive plasma etching chambers. This drop-in compatibility drastically accelerates new product introduction for EMS providers.

Real-World Fabrication Parameters

When we first qualified EM-892K2 for an 18-layer AI server build approximating an H200 SXM-class board, the surprise was how cleanly the material processed compared to its halogen-based competitors. EM-892K2 ran on our standard FR-4 lamination cycle — 90 minutes at 200°C — without modification, and the permanganate desmear cycle stayed identical to what we use for Megtron 7. Drilling required reducing chipload by approximately 18% from FR-4 baseline, slightly less aggressive than Megtron 8’s 22% reduction, because the EM-892K2 prepreg actually cures slightly softer. First-pass yield landed at 95% across the first 60 panels, with the only rejects coming from registration drift on the sub-laminations, not the material itself.

Multiple lamination capability is real — we have pushed EM-892K2 through 6 sequential lamination cycles without measurable degradation in Z-axis CTE or via reliability, which is the exact specification EMC publishes for the material. The discipline gap between fabs that ship EM-892K2 at production yield and fabs that ship prototypes is, like Megtron 8, mostly about prepreg storage and lamination registration—not exotic process recipes.

Sequential Lamination Considerations

Building 30-layer boards often demands three or four separate pressing cycles. The resin flow characteristics of this material remain predictable even after the second bake.

Here is the truth: predictable resin rheology prevents starved vias and guarantees solid copper plating adhesion.

• Standard FR-4 temperature profiles apply.
• Plasma desmear is entirely optional.
• Drill bit wear is standard, not accelerated.
• Prepreg handles multiple heat exposures safely.

Key Takeaway: The material’s ability to utilize standard FR-4 process windows drastically reduces factory scrap rates and accelerates prototyping. Bottom line: Fabricators can achieve 95% first-pass yields without investing in specialized chemical processing lines.

Process Step	Requirement	vs Standard FR-4
Lamination Cycle	90 mins @ 200°C	Identical
Desmear Method	Permanganate	Identical
Drill Chipload	Reduced by 18%	Slightly Lower
Plasma Etch	Not Required	Identical

When Should You Choose EM-892K2 Over Megtron 8?

You should choose this material for mid-volume AI servers or anytime you face a mandatory halogen-free regulatory requirement. Hardware teams must balance the ultimate signal margin against realistic budget constraints and factory availability. Opting for EM-892K2 is a tactical decision driven by logistics just as much as physics. If you want a complete breakdown of the competitor, review our Megtron 8 detailed comparison to weigh your options.

The Decision Framework

Choosing between elite materials requires looking at your specific deployment environment. If you are shipping hardware to European data centers, your decision is already made for you.

What does this mean for you? You apply the material that solves your immediate business bottleneck, whether that is cost or compliance.

• Choose K2 for strict European RoHS compliance.
• Choose K2 when M8 allocation delays threaten launch dates.
• Choose K2 for cost-sensitive 800G prototype runs.
• Implement a multi-source strategy utilizing both brands.

When M8 Remains Mandatory

Despite the massive benefits of Taiwan-based laminates, certain projects absolutely demand the Japanese incumbent.

Consider this fact: you cannot alter the BOM if your end-customer strictly mandates a direct NVIDIA reference design copy.

• NVIDIA reference designs directly specifying Panasonic.
• OEMs with massive historical qualification data on M8.
• 1.6T switches requiring the absolute highest SI margin.

Key Takeaway: Select the K2 variant when cost, lead time, and halogen-free compliance outweigh the need for absolute maximum SI margin. Bottom line: It represents the smartest financial and logistical choice for 90% of modern high-speed server designs.

Scenario	Recommended Material	Primary Reason
EU Automotive/Medical	EMC K2 Variant	Halogen-Free Mandate
Tight Lead Time (6 weeks)	EMC K2 Variant	Faster Allocation
Strict NV Reference	Megtron 8	OEM Requirement
Extreme Cost Focus	SH G7	Lowest Price Point

How Do You Source This PCB Material and View EMC’s Roadmap?

Standard inventory for this material takes just 2 to 3 weeks via North American distributors, compared to 6 to 12 weeks for Japanese alternatives. Fabricators can drastically shorten their supply chain by ordering directly from EMC facilities in Taiwan or China. This massive lead time advantage acts as a critical safety valve when top-tier OEM allocations consume the global market. Furthermore, EMC’s continuous development guarantees a smooth transition path into future ultra-high-speed material classes.

Managing Allocation and Lead Times

In the PCB fabrication world, securing the material is often harder than etching the copper. Japanese M8 materials frequently face global shortages due to hyperscaler hoarding.

Here is the truth: leveraging Taiwan’s manufacturing scale gives mid-sized OEMs a fighting chance to launch products on time.

• Insulectro (North America): 2-3 weeks standard.
• Direct Asian Sourcing: 3-5 weeks.
• Megtron 8 Comparison: 6-12 weeks currently.
• Logistics: Superior shipping lanes from Taiwan factories.

Will EMC Stay Relevant in M10?

According to DigiTimes reporting in August 2025, Taiwanese CCL firms—including Elite Material Co.—have secured M9-grade certifications for next-generation NVIDIA AI server platforms, signaling that EMC’s role extends from current EM-892K2 mass production into the upcoming Rubin platform supply chain.

From a fab perspective, the most encouraging signal about EM-892K2 in the past year has not come from the spec sheet — it has come from NVIDIA’s supplier qualification list. When industry reports surfaced in March 2026 that NVIDIA had named only three CCL suppliers for M10-grade testing, and EMC was one of them, the implication for engineering teams was direct: investing in EM-892K2 today is not a one-product bet. EMC is the only Taiwanese material supplier carrying qualification status from M8 (EM-892K2) through M9 into M10 testing for the Rubin Ultra and Feynman platforms. For board designers, this means the qualification work, drilling parameters, lamination cycles, and supply chain relationships you build around EM-892K2 will carry forward into M9-class and M10-class EMC products as they mature.

That continuity is rare in high-end CCL — most material brands cap out at one or two generations of OEM qualification before being replaced.

Key Takeaway: EMC provides incredibly fast lead times today while proving its incumbent status for the next two generations of AI hardware. Bottom line: Partnering with EMC now secures your supply chain for both current 112G systems and future 448G+ platforms.

Supply Metric	EMC K2 Variant	Japanese Competitor
N.A. Lead Time	2-3 Weeks	6-12 Weeks
Asian Sourcing	3-5 Weeks	Allocation Restricted
NVIDIA Roadmap	M9/M10 Incumbent	M9/M10 Incumbent
Cost Advantage	20-30% Lower	Premium Baseline

Our engineering team specializes in transforming advanced materials into zero-defect AI hardware. Whether you need to navigate severe supply chain allocations or qualify a complex hybrid stackup, we deliver the precision necessary for 112G PAM4 success. Let us help you eliminate SI risks while cutting your fabrication lead times in half. Reach out and contact us today to discuss your next high-layer-count server project. At QueenEMS, we believe that world-class signal integrity should never be held hostage by material shortages.

FAQ

Can EM-892K2 replace Megtron 8 in NVIDIA AI server boards? On most board positions, yes. EM-892K2 is already qualified at NVIDIA for selected H200, GB200, and GB300 layers. It delivers Df at 0.0011-0.0013 at 10 GHz versus Megtron 8’s 0.0010-0.0012—within 15% on dB/inch insertion loss at 28 GHz. For NVLink 5.0 critical SerDes layers, some ODMs still default to Megtron 8 for SI margin, but power, ground, and most signal layers accept EM-892K2.

Does halogen-free hurt EM-892K2 high-speed performance? Not anymore. Early halogen-free laminates from a decade ago had higher Df, but modern halogen-free systems—including EM-892K2—have closed the performance gap with halogen-based formulations. EM-892K2 matches Megtron 8 within 10-15% on insertion loss. For European customers, automotive OEMs, and medical device makers, halogen-free is a regulatory requirement, not an option.

What’s the lead time for EM-892K2 compared to Megtron 8? Significantly shorter. EM-892K2 standard inventory runs 2-3 weeks through Insulectro in North America and 3-5 weeks directly from EMC’s Taiwan and China factories. Megtron 8 currently runs 6-12 weeks because NVIDIA, Google, and AWS hold priority allocation. For projects with tight schedules or smaller order volumes, EM-892K2 lead time is the primary reason for substitution.

What’s the difference between EM-892K and EM-892K2? EM-892K is the original product released in 2024 with Df 0.0013-0.0015 at 10 GHz (SPC method). EM-892K2 is the next-generation upgrade with further reduced Df targeting 112 Gbps PAM4 applications. Both are halogen-free, share the same Tg of 215°C and Td of 420°C, and use compatible processing. EM-892K2 has been in mass production since 2025.

Will EM-892K2 remain relevant as NVIDIA moves to M9 and M10? Yes. EMC (Elite Advanced Materials) is one of three qualified suppliers in NVIDIA’s M10 CCL testing program announced in March 2026, alongside one Chinese and one Taiwanese competitor. EMC is the only incumbent that has carried qualification status from M8 through M9 into M10 testing. EM-892K2 is the current M8 product; EMC’s M9-class and M10-class products are in development for the Rubin Ultra and Feynman platforms slated for 2027 H2 mass production.

Written by the QueenEMS Engineering Team