NuAxi groups public briefings that mention Meta: product launches, capacity and pricing signals, and supply-chain changes. The page updates continuously.
This is a background technology column published on October 29, 2026 (authored by Professor Younghyun Kim of Hanyang University ERICA), explaining — not reporting — industry developments; it contains no price, capacity or policy event. The article states that AI data center bottlenecks have shifted from GPU compute speed to data movement: copper interconnects suffer signal loss, heat and repeater power overhead at high speeds and longer distances, so networks are shifting to silicon-photonics-based optical interconnects; co-packaged optics (CPO) places optical engines next to the switch ASIC to shorten electrical paths, increase bandwidth density and save power. It cites Meta's evaluation of a CPO switch system presented at ECOC 2025, in which power consumption in the optical interconnect portion was approximately 65% lower than a comparable pluggable architecture, with connection stability confirmed under the specific test conditions; companies such as NVIDIA and Broadcom are accelerating commercialization by introducing network products incorporating CPO technology. The article also describes LPO (removable modules with the DSP eliminated) and NPO (optical engines close to but outside the switch package) as alternative approaches, explains that HBM and CPO reduce GPU data-wait time at the memory and network levels respectively, and notes remaining CPO challenges including laser light supply, fiber connection, heat from ASICs, and yield improvement.
co-packaged optics (CPO)silicon photonicsHBMNVIDIA, Broadcom and 3 totalSK Hynix Newsroom
On October 27, AMD said it will invest US$1 billion in Meta's equity offering, joining Nvidia in buying into the social media company's fundraising for AI infrastructure; critics describe this pattern as circular financing. Alongside this, AMD gave an update on its hardware work with Meta, with Helios racks entering testing. The investment is announced only, and the source does not state a completion timeline or the cause.
Helios racksAMD, Meta and 3 totalDigiTimes AsiaPartnership/M&A
Meta has become the first of the four major US hyperscale cloud providers to report its latest earnings, and supply chain players are closely monitoring changes in the company's capital expenditure plans and investment priorities. Per the title, Meta prioritizes high-margin AI inference, while Wiwynn and Quanta are eyeing CPU server demand. The source does not state the cause, and no specific figures or dates are provided.
AI inferenceCPU serversMeta, Wiwynn and 3 totalDigiTimes Asia
The 13th Criminal Division of the Seoul Southern District Court held the fourth hearing on Oct. 22 in the case against Fadu executives over alleged violations of the Capital Markets Act, with former CEO Lee Ji-hyo, current CEO Nam Yi-hyun and executive vice president Won Jong-taek attending as defendants. Prosecutors disclosed internal Microsoft Teams conversations from March 2023 — including 'If we can't go public, we'll have a funding hole' and 'So we'll just go under in July' — arguing Fadu pressed ahead with its IPO despite knowing revenue would plunge; Fadu went public in August 2023. Prosecutors also alleged Fadu used earlier order forecasts for its securities registration statement despite knowing major customers including SMART Modular, SpaceX and SK hynix had postponed or suspended orders for the second half of 2023, with Fadu having projected shipments of 3,200 SSDs for SpaceX satellite servers in Q4 2023. According to prosecutors, former CEO Lee Ji-hyo instructed staff on May 18, 2023 to 'assume no orders from SpaceX for now,' and the following day SpaceX informed Fadu there would be no additional POs before Q3 2024. EVP Won Jong-taek wrote in May 2023 that 'even if we lower the annual revenue target from 240 billion won to 150 billion won, actual results will likely be far lower.' Fadu's production plan from late May 2023 had removed H2 SpaceX orders, and contract manufacturer Quanta Storage demanded compensation for prolonged equipment downtime. The defense countered that the registration statement figures were conservative, worst-case estimates: Fadu included only about 50% of SMART Modular's anticipated order volume, excluded more than 30,000 additional units proposed by SMART Modular in late April 2023 and a 43,500-unit forecast from Hagiwara Solutions in May 2023; on the SpaceX suspension, the May 19 remark came from a SpaceX technical engineer rather than a purchasing official, and Lee Ji-hyo separately confirmed with SpaceX's sales organization the possibility of orders resuming within the year. Fadu resumed shipments toward the end of 2023, but supplied 2TB models for Falcon 9 rockets rather than the originally planned 8TB satellite models, with substantially smaller revenue contribution. Second witness Park Jong-hyun, head of Fadu's SoC business team, testified the sharp 2023 decline in SSD controller shipments for Meta stemmed from issues between SK hynix and Meta: SK hynix told Fadu in late 2022 it expected to supply about 900,000 controllers annually to Meta, but the forecast was abruptly cut to 350,000 units on Jan. 6 due to a dispute, and SK hynix discontinued the existing DRAM used in Meta SSDs and began redesigning. Fadu planned to cut the price of its third-generation SSD controllers from $43–$48 to around $30 to regain Meta's business, and included 434,000 controllers for Meta in its 2023 forecast — exceeding the 350,000-unit forecast — which appeared to contradict claims the estimates were conservative. Fadu ultimately failed to secure additional Meta orders through SK hynix; Meta switched its SSD supplier to SanDisk in 2024, and since 2024 Fadu has supplied SSD controllers to Meta through SanDisk.
SSD controllerssolid-state drives (SSDs)Fadu, SK hynix and 8 totalThe Elec
Meta Platforms is facing rising financial pressure from long-term AI infrastructure spending: free cash flow turned negative in Q3 2026, and future lease and non-cancelable contract commitments climbed to about US$825 billion. The company also raised the lower end of its 2026 capex forecast by US$5 billion to US$135 billion, signaling continued AI expansion as financing and investment returns become new challenges.
At PyTorch Conference 2026, held in San Jose, California, on Oct. 20 local time, Google and Amazon unveiled the open-source frameworks TorchTPU and TorchNeuron, respectively, designed to simplify migrating existing AI applications from Nvidia GPUs to their own chips and to challenge Nvidia's CUDA ecosystem dominance. PyTorch, developed in 2016 primarily by Meta's (formerly Facebook's) AI research lab FAIR and used by approximately 85% of AI application developers worldwide, was originally designed around Nvidia GPUs and the CUDA platform, making migration to other chips difficult. TorchTPU, developed by Google in collaboration with Meta, according to Google, allows more than 98% of existing PyTorch models to run on TPUs simply by changing the device string in the code to model.to("tpu"), with conversion taking from a few minutes to a few days, and Google said it enabled TPUs to operate as native devices on an equal footing with GPUs and CPUs at the core PyTorch level. AWS's TorchNeuron similarly allows PyTorch-based AI models to run directly on AWS's Trainium AI chips, requiring developers to change only a few lines of device settings, replacing Nvidia's cuda with AWS's neuron. Amazon has released TorchNeuron as a public beta through GitHub under the Apache 2.0 open-source license and is also contributing code directly to the main PyTorch open-source project. Previously, migrating AI applications optimized for Nvidia's ecosystem to Google's or Amazon's proprietary AI chips took at least several days and, in more complex cases, several months; the new tools effectively remove these time and cost barriers. As TorchNeuron, TorchTPU and similar tools gain adoption, dedicated AI chips are expected to compete primarily on underlying performance, power efficiency and pricing, though additional time will be needed to verify compatibility, performance and costs for individual models.
Nvidia CUDATorchTPUTorchNeuronGoogle, Amazon and 5 totalThe Elec
Related manufacturers
Google, Amazon, Meta, Nvidia, Amazon Web Services (AWS)
Meta Platforms reported third-quarter results: for the quarter ended September 30, 2026, revenue rose 26% year on year to US$64.6 billion, and its core advertising business is still growing fast enough to carry one of the largest infrastructure programs in the technology industry (its AI build-out); meanwhile the operating margin fell to 20% from 40% a year earlier, with legal liabilities becoming a material cost line. The source does not state the cause.
Meta revealed this week that more than 11 percent of the GPUs at its Prometheus supercluster in New Albany, Ohio, are either dead or sitting idle, waiting to step in when a GPU fails. Prometheus is among Meta's largest compute deployments to date, powered by Nvidia's Grace Blackwell and (soon) Vera Rubin rack systems, and Meta says the supercluster will eventually scale to a gigawatt of capacity spanning about a million accelerators. The cluster is currently powered by a combination of GB200 and GB300 systems, each with 72 GPUs per NVLink domain, but only 64 are ever in use at any given moment, with the other eight essentially hot spares; Meta did this because a single failed GPU would take out an entire rack and interrupt the training run. Meta says that when it began the first phase of the Prometheus deployment it was experiencing roughly 1,000 interruptions per day for every 8,000 GPUs, and in the year since it has cut interruptions by 99 percent, to about 11 per day per 8,000 GPUs. Meta also says it achieved more than 85 percent goodput for pretraining and more than 90 percent for reinforcement learning on its latest model Muse Wonder, and as it works toward a million GPUs it hopes to reduce the interruption rate to just two per day for every 8,000 GPUs, or about 100,000 GPU-hours mean time between interrupts. The source does not state any implication for GPU market supply or demand.
Sandisk announced in a press release issued from Milpitas, California, on October 14, 2026, that it will be the exclusive provider of flash storage for Meta VR Glasses, supplying Meta-certified microSD Express memory cards and USB-C flash drives when the glasses launch in spring 2027. The arrangement remains at the announcement stage and the product has not yet launched. The source does not state the cause.
Sandisk announced on Oct. 19 that it will be the exclusive supplier of flash memory for Meta's virtual reality (VR) glasses, which are scheduled for launch in spring 2027. The products include microSD Express memory cards and USB-C flash drives. Sandisk recently entered into a partnership with Meta to jointly design flash memory products for VR applications, and the products have undergone preliminary testing to meet Meta's quality standards. Sandisk will begin selling the related storage products following the launch of Meta's VR glasses. Sujin Park, vice president of consumer products at Sandisk, said high-performance microSD Express memory cards are compatible with Meta VR glasses, allowing users to instantly expand their available storage. The source does not state any impact on price or supply.
According to MacRumors, citing Bloomberg's Mark Gurman on his podcast, Apple is reportedly shifting its focus to the M7 chip lineup, skipping higher-end M6 variants as it prioritizes technologies designed for AI workloads; Apple has introduced the base M6 chip in the Mac mini, with M6-powered iMac and entry-level MacBook Pro models also planned, but the company reportedly intends to skip the M6 Pro, M6 Max, and M6 Ultra, moving directly from their M5 predecessors to the M7 lineup. Gurman noted that high-end touchscreen MacBook Pro models featuring M7 Pro and M7 Max chips are expected by the end of 2027, while an entry-level MacBook Pro powered by the standard M7 is also slated for 2027; the M7 Ultra is reportedly planned for the first half of 2028, making a Mac Studio refresh in 2027 unlikely. Per MacRumors, the upcoming OLED MacBook Pro models are expected to retain the M5 Pro and M5 Max chips used in current models, with no processor upgrade planned until 2027. Separately, according to CNBC, Apple and Meta have announced plans to enable developers to run Meta's Muse Spark Open model locally on Apple's high-end Mac Studio desktops; Meta describes Muse Spark Open as its most capable U.S. open-weight model, featuring 816 billion total parameters, 42 billion active parameters, and a 1 million-token context window, built on a mixture-of-experts architecture and designed to run locally on M5 Ultra-powered Mac Studios using Apple's MLX framework; each M5 Ultra supports up to 512GB of unified memory and 1.2TB/s of memory bandwidth, allowing two connected Mac Studios to accommodate the entire model. According to CNBC, a Mac Studio equipped with the required M5 Ultra chip and 256GB of memory starts at $10,799, while a 512GB configuration is expected later this month. Apple has also begun inviting select media representatives to a product launch experience in Los Angeles on Wednesday, October 28.
According to TrendForce analysis published on October 15, 2026, AR glasses are moving into the consumer market, with 2026 shipments forecast to double to 1.31 million units (a forecast, not an actual result), driven by Meta and Chinese brands. On the display side, OLEDoS is expected to shift toward monochrome green, while LEDoS advances to dual-color. The source does not state specific supply or price implications of these shifts.
British semiconductor IP company Arm said on Oct. 15 that it has developed a new CPU benchmarking framework for agentic AI, with support from Fujitsu, Meta and OpenAI, designed to evaluate real-world agentic AI workloads and performance across the entire computing process. Arm claimed that under the new framework its AGI CPU delivered approximately twice the task throughput per rack of AMD's fifth-generation EPYC Turin 9755 server CPU; the measurements were conducted by analyst firm Signal65 at Arm's request. The AGI CPU, unveiled in March, is Arm's first internally designed server CPU and was jointly developed with Meta. Arm said conventional CPU benchmarks and model inference metrics fail to capture CPU workload demands as AI agents grow. Meta is also adopting Arm's AGI CPU in its next-generation open server rack design, Yosemite YV5. The source does not state the cause; the comparison is Arm's claim, not independently verified.
CPUAGI CPUfifth-generation EPYC Turin 9755 server CPUArm, AMD and 6 totalThe Elec
At the Ayar Labs CPO Summit, Meta, Broadcom, AMD, and others weighed in on readiness for a 2028 co-packaged optics (CPO) demand surge and the OCI MSA's role. The item is an industry-discussion piece about production planning ahead of 2028; the source does not state any price or supply-demand changes or give specific quantities.
co-packaged opticsAyar Labs, Meta and 4 totalEE Times
Meta released the 30-billion-parameter open-source model Muse Glimmer on August 10, with CEO Mark Zuckerberg advocating for open-source AI to prevent monopolies. This signals Meta's strategic shift to differentiate itself through openness while catching up to competitors.
AI ChipsNPUEdge AI ChipsMeta PlatformsDigiTimes AsiaProduct Launch
TrendForce raised 2026 AI server shipment forecast to 31% YoY, citing $886.7B CapEx for top CSPs. North American hyperscalers drive demand for NVIDIA GB/VR platforms and in-house ASICs, while Chinese CSPs boost domestic AI infrastructure.
AI ServerGPU ClustersLiquid Cooling InfrastructureNVIDIA, Google and 10 totalEvertiqMarket Analysis
MediaTek confirms AI ASIC collaboration with Intel using EMIB-T advanced packaging, advancing 400G SerDes IP. Targeting next-gen opportunities including Google TPU v10 and Meta custom chips.
关注
暗示高端AI芯片封装与高速接口需求增长。
400G SerDes IPEMIB-T (Advanced Packaging)MediaTek, Intel and 4 totalTrendForce Official (Market Data)Partnership/M&A
TrendForce reports severe bottlenecks as KV Cache demand surges against constrained memory supply. Penguin Solutions, Marvell, and Meta deploy CXL solutions for capacity expansion, while NVIDIA and Google compress demand. Bullish: Supply constraints drive price pressure.
MemoryKV CachePenguin Solutions, Marvell and 5 totalTrendForce Official (Market Data)Shortage Alert
Google announced another increase in capital expenditure for 2027, with industry watchers expecting AWS and Microsoft to follow suit. This surge in spending aims to sustain AI infrastructure momentum amid growing demand. Bullish signal for AI chip demand and potential supply constraints.
AI AcceleratorsTPUGoogle, Meta and 4 totalDigiTimes AsiaCapacity Change