Broadcom

Recent news

• Jun 11, 2026

  Will Co-Packaged Optics Transform Data Centers?

  This article explains co-packaged optics and assesses practicality and adoption timelines.

  • Explains CPO and current status: The piece describes Co-packaged optics (CPO) as the integration of optical transceivers with processors in very close proximity, and cites claimed benefits of up to 350% power-efficiency gains and up to 1,000% bandwidth increases. It notes that Broadcom has made CPO-capable switches available, hyperscalers like Meta have run experiments, and Nvidia is working on processor-integrated optics (technology still in development).
  • Limitations and implementation details: The article highlights concrete constraints: limited CPO hardware availability, a shortage of lasers affecting production, thermal management challenges from higher power density, maintenance limitations due to integrated transceivers, and proprietary transceiver ecosystems that hinder third-party interchangeability. No firm industry-wide deployment timeline is announced; examples cited are vendor products and hyperscaler experiments.
• Jun 2, 2026

  Alphabet’s $80B Fundraising Spotlights AI’s Soaring Capital Needs

  Alphabet has announced a planned equity raise of approximately $80 billion to fund AI infrastructure and global compute capacity.

  • Main announcement: Alphabet will raise ~$80 billion in equity to fund capital expenditures to scale AI infrastructure and global compute, including a $10 billion private placement with Berkshire Hathaway (Berkshire to buy $5 billion each of Class A and Class C shares).
  • Background and details: The article cites analysts and industry observers noting dramatically increased data center capex (Google expected to double data center capex this year) and emphasizes that funds are intended for data centers, power delivery, networking, cooling, and AI accelerators; portions of proceeds may also cover employee equity and corporate finance rather than direct buildout.
• May 1, 2026

  Google-Anthropic Deal: AI Capacity Now Pre-Sold at Gigawatt Scale

  Google is understood to be in talks to invest up to $40 billion in Anthropic.

  • Main terms: Reported initial $10 billion investment, with up to $30 billion contingent on performance targets (total potential $40 billion), plus up to 5 GW of computing power over five years supplied by Google; companies have not publicly confirmed these terms.
  • Background and implementation details: Builds on prior Google investments and TPU commitments to Anthropic and a separate partnership with Broadcom; delivering 5 GW approaches utility scale and will require multi-year coordination on land acquisition, permitting, grid interconnection, and multiple hyperscale campuses.
• May 1, 2026

  AI Capacity Is Being Pre-Sold at Gigawatt Scale

  Google is reportedly negotiating to invest up to $40 billion in Anthropic and to supply reserved compute capacity tied to future performance, though the companies have not publicly confirmed the terms.

  • Main announcement: Reported deal would provide an initial $10 billion investment with up to $30 billion contingent on performance, and Google would supply as much as 5 GW of computing power over five years to Anthropic; the companies have not publicly confirmed these terms.
  • Background and structure: The arrangement pairs capital with pre-negotiated capacity (including TPU commitments) and builds on existing ties with Google and a separate Broadcom partnership for custom silicon; delivering ~5 GW would require multiple hyperscale campuses, years of land acquisition, permitting, and grid interconnection, and coordinated planning and financing.
• Apr 9, 2026

  Hyperscalers will own two-thirds of data center capacity by 2031

  Synergy Research Group reported that hyperscalers will account for 67% of all data center capacity by 2031.

  • Main announcement: Synergy Research Group says hyperscalers (Google, Microsoft, AWS) will reach 67% of global data center capacity by 2031, with enterprise on-prem data centers dropping from 56% in 2018 to 19% by 2031; the report also notes almost 60% of hyperscale capacity is in own-built facilities and non-hyperscale colocation accounts for ~20%.
  • Background & details: The article cites planned > $500 billion in capex by Google/Microsoft/AWS for AI infrastructure in fiscal year 2026, cites hyperscalers operating ~1,297 large data centers in Q3 2025 (1,360 by end-2025), references commitments such as the Ratepayer Protection Pledge (Google, Oracle, xAI, Meta, Microsoft, OpenAI, Amazon) and highlights electricity demand concerns (EIA: price hikes up to 79% in areas like Texas by 2027); it references expanded compute partnerships (Anthropic–Google/Broadcom; OpenAI–AMD) with multi-gigawatt capacity starting 2027.
• Apr 8, 2026

  Aria Networks raises $125M and debuts its approach for AI-optimized networks

  Aria Networks has launched the Deep Networking platform and disclosed $125 million in total funding.

  • Main announcement: Aria Networks (founded January 2025 by Mansour Karam) has made Deep Networking generally available this week: a combined offering of purpose-built switching hardware, a hardened SONiC implementation, microsecond ASIC-level telemetry, and intelligent agents across the stack, plus an exposed MCP server and embedded FDE (forward deployed engineers) model; the company disclosed $125 million in total funding from Sutter Hill Ventures, Atreides Management, Valor Equity Partners and Eclipse Ventures.
  • Background and specifics: The platform targets AI operational metrics Model FLOPS Utilization (MFU) and token efficiency (MFU typically 33–45% for training, <30% for inference); Aria models a 3% MFU improvement across a 10,000-XPU cluster as ~$49.8 million annual revenue gain (≈7.9% revenue improvement). Hardware portfolio details include Aria Switch 800G (64 x 800G, Broadcom Tomahawk 5), Aria Switch 1.6T High Radix (128 x 800G, TH6), and Aria Switch 1.6T (64 x 1.6T, air and full liquid cooling). Software update cadence targeted at weekly releases, and FDE feedback is engineered back into products.
• Apr 8, 2026

  Anthropic Secures Multi-Gigawatt TPU Deal With Google, Broadcom

  Anthropic has signed a deal with Google and Broadcom to secure multiple gigawatts of next-generation TPU capacity beginning in 2027.

  • Deal details and timeline: The agreement will deliver multiple gigawatts of TPU-based compute capacity (Broadcom disclosed it will support roughly 3.5 gigawatts) beginning in 2027, with most capacity in the US; the move expands on Anthropic’s previously announced $50 billion investment in domestic computing infrastructure and is described as the company’s largest compute commitment to date.
  • Background and context: Anthropic reported a reported revenue run rate surpassing $30 billion in 2026 (up from ~$9 billion at end-2025) and said more than 1,000 customers now spend over $1 million annually; the deal reflects a strategic shift toward long-term, utility-scale TPU commitments (vertical hardware/cloud alignment) while AWS remains their primary partner.
• Apr 7, 2026

  Anthropic signs AI compute deal with Google &amp; Broadcom as it surpasses OpenAI’s revenue

  Anthropic has signed a new agreement with Google and Broadcom to secure multiple gigawatts of next‑generation AI compute capacity, expected to come online from 2027.

  • Main announcement: Anthropic secured multiple gigawatts of next‑generation AI compute capacity from Google and Broadcom, with most capacity located in the United States and expected to be online from 2027; this expands Anthropic’s infrastructure aligned with its earlier $50 billion US investment commitment for AI infrastructure.
  • Background and additional details: The company reported annualised revenue run rate (ARR) crossed $30 billion in 2026 (up from ~$9 billion at end‑2025) and said >1,000 enterprise customers now spend > $1 million annually (doubling in under two months); Anthropic uses a multi‑platform hardware strategy (AWS, Google TPUs, NVIDIA GPUs), with Amazon remaining its primary cloud and training partner (ongoing Project Rainier).
• Apr 6, 2026

  Hyperscaler backlogs show growing demand for AI infrastructure

  Dell’Oro Group reported that global data center capital expenditures grew 57% in 2025 to $726 billion and are forecast to grow more than 50% in 2026, pushing total capex to cross the $1 trillion mark.

  • Main announcement: Dell’Oro Group: data center capex up 57% to $726 billion in 2025, with an estimated >50% growth in 2026 that will make data center capex surpass $1 trillion (earlier forecast for $1T was 2029). The report attributes the surge primarily to AI-driven demand from hyperscalers.
  • Background and details:Hyperscalers drove the increase — Amazon, Google, Meta, Microsoft raised data center capex by 76% collectively; reported company figures include Amazon $131 billion capex in 2025 (mostly data centers) and Amazon expects ~$200 billion in capex in 2026, Google plans ~$180 billion in 2026; hyperscaler backlogs cited: Amazon backlog $244 billion, Google backlog $240 billion. Surveys (BCG) show companies plan to double AI spending (from 0.8% to 1.7% of revenues) in the year referenced.
• Apr 3, 2026

  ‘Inference Is Bigger Than Any One Chip’ – d-Matrix CEO on GigaIO Deal

  d-Matrix has announced the acquisition of GigaIO’s data center business to internalize interconnect technology and accelerate delivery of rack-scale AI inference infrastructure.

  • Business action: d-Matrix completed a business unit acquisition of GigaIO’s data center assets (ownership of the unit’s related assets transfers to d-Matrix); financial terms were not disclosed; the deal builds on a collaboration that began in 2025 and integrates GigaIO’s SuperNode and FabreX PCIe fabric into d-Matrix’s inference stack (which also includes Corsair inference accelerators, JetStream networking, Aviator software, and the SquadRack reference architecture developed with Broadcom and Arista).
  • Background & implementation details: GigaIO will continue operating independently and refocus on edge computing; a team of systems engineers based in Carlsbad, California joins d-Matrix, establishing a new Southern California engineering presence; d-Matrix now operates six innovation hubs across North America, Europe, and Asia; target customers include hyperscalers, frontier AI labs, and enterprise deployments.
• Apr 1, 2026

  OpenAI Raises $122B to Expand AI Infrastructure, Broadens Cloud and Chip Strategy

  OpenAI has announced it raised $122 billion in a funding round valuing the company at $852 billion to expand compute capacity, cloud partnerships, and data center infrastructure.

  • Funding and purpose:$122 billion raised at an $852 billion valuation to support expanded compute capacity, cloud partnerships, and data center expansion; OpenAI also expanded its credit facility to $4.7 billion to provide additional flexibility for capacity investments.
  • Infrastructure and partners: OpenAI is working across Microsoft, Oracle, AWS, CoreWeave, and Google Cloud and using a mix of silicon platforms including Nvidia GPUs, AMD chips, AWS Trainium, Cerebras systems, and custom Broadcom silicon; the platform now processes more than 15 billion tokens per minute, and analysts named include Holger Mueller, Daniel Newman, and Matt Kimball.
• Apr 1, 2026

  With new Marvell deal, Nvidia is chasing the AI control layer

  Nvidia has announced a partnership with Marvell Technology and a $2 billion strategic investment in Marvell.

  • Main announcement: Nvidia and Marvell will integrate Marvell XPUs and scale-up networking with Nvidia NVLink Fusion, enabling customers to build “semi-custom” AI infrastructure that mixes non-Nvidia accelerators with Nvidia GPUs, LPUs, DPUs and Spectrum-X switches; Nvidia is investing $2 billion in Marvell as part of the deal. No specific implementation timeline is provided in the article.
  • Background and additional details: The partnership includes collaboration on 5G/6G AI-RAN (Aerial AI-RAN), advanced optical interconnects and silicon photonics; Nvidia has also announced other ecosystem investments (a combined $4 billion for photonics vendors Coherent and Lumentum and a $5 billion purchase of Intel stock) to expand NVLink-enabled architectures and broader ecosystem alignment.
• Mar 31, 2026

  Cisco extends its Enterprise Agreement to include Nutanix Cloud Platform

  Cisco has extended its Enterprise Agreement to include the Nutanix Cloud Platform, adding Nutanix HCI to Cisco’s EA licensing and services program.

  • Main announcement: Cisco has officially extended its Enterprise Agreement (EA) to include Nutanix Cloud Platform (HCI), providing customers with predictable pricing, price protection for the EA term, and flexible, true-forward consumption (ability to increase Nutanix usage during the year and pay at the annual anniversary) without renegotiating contracts.
  • Background and implementation details: Cisco and Nutanix have had a multi-year partnership (Cisco ended development of HyperFlex in 2023 and handed HCI to Nutanix); the vendors deliver products such as Cisco Compute Hyperconverged with Nutanix (combining Cisco hardware and Nutanix Cloud Platform), tightened Intersight–Nutanix integrations, support for Nutanix GPT-in-a-Box, and e-bonded global support and remote cluster deployment capabilities (per World Wide Technology commentary).
• Mar 31, 2026

  Nvidia Deepens AI Push With $2B Marvell Deal

  Nvidia announced a partnership and a $2 billion investment in Marvell Technology to integrate Marvell’s custom silicon and interconnects into Nvidia’s NVLink Fusion ecosystem.

  • Partnership details: Nvidia is investing $2 billion and has unveiled a collaboration that pairs Marvell’s custom XPUs, NVLink-compatible scale-up networking, optical DSP and silicon photonics with Nvidia’s NVLink Fusion rack-scale architecture; Nvidia will supply Vera CPU, ConnectX NICs, BlueField DPUs, NVLink interconnect, and Spectrum-X switching platform. The announcement was unveiled today and positions Marvell as a semi-custom AI infrastructure enabler within Nvidia’s ecosystem.
  • Background and scope: The deal emphasizes heterogeneous AI architectures and expands NVLink beyond Nvidia-native silicon (Marvell will add NVLink support to its XPUs); the companies will also collaborate on silicon photonics, optical interconnects, and AI-RAN telecom deployments using Nvidia’s Aerial platform. No explicit multi-year timeline or implementation dates were provided in the article.