Meta to Deploy Custom AI Chips in Data Centers by 2027

GuruFocus··USTWCN·Read original
4▲3 ▼0Impact / 5
Summary · why it matters

Meta Platforms is moving deeper into custom AI silicon, with a new generation of internally designed chips set to enter its data centers in the first half of 2027. The company is currently testing its third-generation MTIA 450 processor, code-named Arke, while its successor, MTIA 500, or Astrid, is expected to complete design work in about a month and reach data centers by the end of 2027. Meta is working with Broadcom on chip design and Taiwan Semiconductor Manufacturing Co. on production, and has committed to deploying more than a gigawatt of the chips over a 12-month period. Twelve Arke chips delivered by TSMC on Sept. 1 performed within 2% to 3% of Meta's simulations and have already run Meta models alongside models from DeepSeek and Alibaba. Meta also canceled its planned Olympus processor, which was intended to handle both AI training and inference, partly because of cost concerns, and is instead prioritizing inference, the day-to-day running of AI models.

Impact on assets 5

Artificial Intelligence▲
Meta Platforms Inc.
META
▲ PositiveTechnologyrelevance

Meta is testing its third-gen MTIA 450/Arke and MTIA 500/Astrid custom AI chips for its data centers.

Broadcom Inc
AVGO
▲ PositiveDemandrelevance

Meta is working with Broadcom on the design of its custom MTIA AI chips, a concrete chip-design engagement.

Semiconductors▲

Theme Impact 8

Off-coverage companies 1

DeepSeeki
Private± Mixedrelevance

Related news

SpainEuropean UnionMacao SAR ChinaChina
▲impact 4

Alibaba Plans New AI Cloud Regions in Europe and Spain Data Centers

Alibaba Group Holding outlined a global AI cloud expansion for 2026 that includes new AI-focused cloud regions across Europe and early stage data center projects in Spain. The company also announced a partnership with Sands China to deploy embodied AI robots and spatial intelligence systems at a flagship resort. The buildout is part of a broader push to move from merchant-focused tools toward becoming a compute provider, with plans for 20 gigawatts of capacity plus custom chips. The article notes that profit margins have already moved from 14.8% to 7%, and that the heavy AI and cloud spending weighs on EBITDA as Alibaba trades short term earnings for a larger infrastructure role.
About megatrends
Artificial Intelligence › AI Data Center & Build-out ▲Supply
Artificial Intelligence › AI Compute Cloud & Neoclouds ▲Supply
Cloud & Digital Infrastructure › Hyperscale Cloud (IaaS / PaaS) Supply
Cloud & Digital Infrastructure › Mega-cap Hyperscalers Supply
Artificial Intelligence › Custom Silicon / ASIC ▲Supply
9988.HK · Capital · Neutral Heavy AI/cloud spending weighs on EBITDA and margins fell from 14.8% to 7%, offsetting the expansion's long-term infrastructure ambitions.
9988.HK · Technology · Positive Alibaba outlined new AI-focused cloud regions in Europe and Spain data centers plus custom chips, expanding its compute-provider capabilities.
1928.HK · Demand · Positive Alibaba partnership will deploy embodied AI robots and spatial intelligence systems at Sands China's flagship resort, a concrete adoption deal.
Read original ↗
Simply Wall St·4hRead more →
United States

Marvell Technology to Showcase AI Data Center Solutions at OCP Global Summit 2026

Marvell Technology plans to present a suite of next generation AI data center solutions at the OCP Global Summit 2026, highlighting new hardware for AI infrastructure, memory expansion, high speed connectivity, and integrated security. The offerings target the memory wall, faster networking requirements, and security for AI centric workloads. The showcase lines up with Marvell's push into AI data center infrastructure, spanning switching, optics, and memory expansion, with products including the Teralynx T100 102.4T switch, Structera CXL memory, and Ara 1.6T optics mapping to the high bandwidth AI clusters large cloud customers are building. The event supports the existing narrative rather than rewriting it, as the demos sit in the same AI connectivity and custom silicon areas analysts link to the US$12b FY27 and US$18b FY28 revenue guidance, with margins aiming for the 38% to 40% non GAAP range. The key test will be concrete follow through after the summit, such as new or expanded design wins with hyperscalers referencing Teralynx T100, Structera CXL, or 1.6T optics in coming quarters.
About megatrends
Artificial Intelligence › AI Networking & Interconnect ▲Technology
Artificial Intelligence › Switching & Networking Silicon/Systems ▲Technology
Semiconductors › Logic, Compute & Connectivity Processors Technology
Artificial Intelligence › AI Compute & Accelerator Silicon Technology
Artificial Intelligence › Custom Silicon / ASIC Technology
Artificial Intelligence › HBM & AI Memory Technology
MRVL · Technology · Positive Marvell will showcase next-gen AI data center solutions including Teralynx T100 switch, Structera CXL memory, and Ara 1.6T optics at OCP Global Summit 2026.
Read original ↗
Simply Wall St·4hRead more →
United States
▲impact 4

Google Launches Prototype TPU Satellite Aboard SpaceX Mission

Alphabet Inc. has launched a prototype Tensor Processing Unit satellite aboard SpaceX's Transporter-18 mission on October 1, 2026, in partnership with Planet, marking its first concrete step toward putting data centers in orbit. The test, part of Google's Project Suncatcher, will examine how its TPUs handle launch stress, radiation, and extreme temperatures in low Earth orbit, with ground testing already showing Trillium TPUs can withstand radiation levels higher than those expected across a five-year mission. Heat management remains a key challenge because conventional air cooling does not work in a vacuum, prompting Google to evaluate heat pipes and radiators, and a two-satellite test is planned for 2027. Google emphasized the current mission is intended to collect engineering data rather than demonstrate a commercial system. SpaceX could benefit from additional orbital AI experiments if hyperscalers continue choosing its launch services, even as its valuation of over $2 trillion at about 45 times forward sales rests on management's own targets, including Elon Musk's expectation of a $100 billion annual revenue run rate by December and a $1 trillion revenue target moved up to 2030.
About megatrends
Artificial Intelligence › Custom Silicon / ASIC ▲Technology
Space Economy › Launch Services & Propulsion ▲Technology
Cloud & Digital Infrastructure › Hyperscale Cloud (IaaS / PaaS) Technology
GOOG · Technology · Positive Alphabet launched a prototype TPU satellite with Planet, testing its TPUs in orbit as part of Project Suncatcher.
SPCX · Demand · Positive SpaceX's Transporter-18 mission launched Google's prototype TPU satellite, and it could gain more orbital AI launch business if hyperscalers keep choosing it.
PLANET.BK · Demand · Positive Planet partnered with Google on the prototype TPU satellite launch, gaining a concrete orbital AI collaboration.
Read original ↗
Insider Monkey·1dRead more →
United States

Sequoia-backed Nuvacore seeks $2.5bn valuation in funding round

Nuvacore, a six-month-old semiconductor startup backed by Sequoia Capital, is seeking to raise hundreds of millions of dollars at a valuation of approximately $2.5 billion, Reuters reported exclusively on Friday, citing two people familiar with the fundraising. The San Jose-based company is developing a central processing unit designed for data centers, targeting growing demand for computing infrastructure supporting artificial intelligence applications. Nuvacore has yet to release a product, and its latest funding round remains open, with the final valuation and amount raised subject to change. The Information previously reported that the startup was seeking at least $200 million, but the proposed $2.5 billion valuation had not been disclosed. Founded by semiconductor engineers Gerard Williams, John Bruno and Ram Srinivasan, Nuvacore plans to design its processor's core functionality before selecting an architecture, potentially avoiding limitations associated with Intel and Advanced Micro Devices' x86 technology or Arm-based designs. Williams previously worked at Apple and founded Nuvia, which Qualcomm acquired for $1.4 billion in 2021. Nuvacore declined to comment on the fundraising, Reuters reported.
About megatrends
Semiconductors › Logic, Compute & Connectivity Processors Capital
Artificial Intelligence › AI Compute & Accelerator Silicon ▲Capital
Artificial Intelligence › Custom Silicon / ASIC Capital
AMD · Competition · Neutral Nuvacore plans to design its CPU core to potentially avoid x86 limitations, a nascent competitive threat to AMD's x86 data-center CPUs.
INTC · Competition · Neutral Nuvacore aims to sidestep limitations of Intel's x86 technology, a potential long-term competitive challenge to Intel's data-center CPU business.
Read original ↗
Investing.com·1dRead more →
United States

Nvidia to invest in AI chip rival d-Matrix, report says

Nvidia is set to invest in d-Matrix, a maker of artificial intelligence server chips, The Information reported, citing three people with knowledge of the deal. The investment is part of Nvidia's initiative to make its technology and chips compatible with those of other designers, the outlet added. Nvidia has struck similar deals with Marvell Technology, Intel, and Amazon, among others. Nvidia did not immediately respond to a request for comment from Seeking Alpha.
About megatrends
Artificial Intelligence › AI Compute & Accelerator Silicon ▲Capital
Artificial Intelligence › GPU & Merchant Accelerators Capital
Artificial Intelligence › Custom Silicon / ASIC Capital
NVDA · Capital · Positive Nvidia is investing in AI chip rival d-Matrix as part of its compatibility initiative
d-Matrix · Capital · Positive d-Matrix is set to receive an investment from Nvidia
Read original ↗
Seeking Alpha·2dRead more →
United States
▲2impact 4

Synopsys Touts Amazon and OpenAI AI Deals, Raises Fiscal 2027 Outlook

Synopsys unveiled a major agreement with Amazon covering its intellectual property across custom silicon programs including Trainium, Graviton and Nitro, alongside a partnership with OpenAI to develop GPT-Synopsys, a specialized AI model for semiconductor design. The Amazon deal introduces a royalty-based component tied to chip production. Management also issued a stronger-than-expected outlook, projecting fiscal 2027 revenue of $11.10–$11.20 billion, above the roughly $10.81 billion analysts had expected, and non-GAAP earnings guidance of $19.04–$19.12 per share versus prior consensus near $17.81. The company now targets roughly 15% annual revenue growth through fiscal 2030, with operating margins approaching 50% and annual adjusted EPS growth in the mid-20% range, and intends to repurchase approximately $1 billion in shares over the coming months with a longer-term goal of returning up to 50% of free cash flow to shareholders. Synopsys currently holds a Zacks Rank #1 (Strong Buy), and over the last 60 days the consensus earnings estimate has risen 6.1% to $18.26 per share, with five upward revisions and one downward revision.
About megatrends
Semiconductors › EDA & Semiconductor IP ▲Technology
Artificial Intelligence › EDA & Semiconductor IP ▲Technology
Artificial Intelligence › AI Compute & Accelerator Silicon ▲Technology
Artificial Intelligence › Custom Silicon / ASIC ▲Technology
SNPS · Capital · Positive Synopsys raised fiscal 2027 revenue/EPS outlook above consensus and plans ~$1B buyback with up to 50% FCF returns.
SNPS · Demand · Positive Synopsys unveiled major Amazon IP agreement and OpenAI partnership to develop GPT-Synopsys for chip design.
AMZN · Demand · Positive Amazon agreement covers Synopsys IP across its custom silicon programs (Trainium, Graviton, Nitro), a concrete product deal.
OpenAI · Technology · Positive OpenAI partnership with Synopsys to develop GPT-Synopsys, a specialized AI model for semiconductor design.
Read original ↗
Zacks Investment Research·2dRead more →