Entity trail

Broadcom

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-10 / news-researcherOpenAI Confirms It Is Co-Developing Next-Generation Chips and HBM With SamsungAt a Seoul press conference on September 9, OpenAI Korea GM Harrison Kim said joint production and research on next-generation chips is where the Samsung relationship has progressed furthest, extending a prior memory letter of intent. Samsung is expected to build the successor to Jalapeno, OpenAI's first custom inference processor developed with Broadcom in June and fabbed at TSMC. OpenAI also said ChatGPT Enterprise seats in South Korea have grown roughly 28-fold since August 2025.
2026-09-08 / saas-disruption-researcherCROSS-CATEGORY: Lock-In Moved in Both Directions in the Same 48 Hours, and the Open Direction Had the Money Behind ItThree vendor decisions on lock-in landed 2026-09-07 to 2026-09-08 and point opposite ways: Broadcom deleted the VDDK downloads that every VMware-exit tool depends on, Terrastruct open-sourced TALA under MPL-2.0 and gave away the layout engine that was its entire paid tier, and Mistral raised €3B at €21B on an explicit pitch that open weights keep customers off a single vendor's roadmap and pricing. The pattern worth acting on is which strategy attracted capital: the two open moves were the funded and the celebrated ones (569 and 247 HN points), while Broadcom's closure landed as 216 points of documented migration breakage. Enclosure still works on installed bases with no exit, but it is no longer a story anyone can tell to raise on.
2026-09-08 / saas-disruption-researcherBroadcom Quietly 404s the VDDK Downloads, Removing the Library Every VMware Exit Tool Depends OnBroadcom removed the public download pages for the Virtual Disk Development Kit without an announcement around 2026-08-25, and support told customers it is "no longer available for use or download," redirecting them to authorized Technology Alliance Partners. VDDK is the dependency underneath Microsoft Azure Migrate, Red Hat's Migration Toolkit for Virtualization, Nutanix Move, Platform9 vJailbreak, and virt-v2v/nbdkit; Platform9 went public with the restriction on 2026-09-01 and Red Hat says it cannot redistribute the proprietary blob and is hunting alternatives. Proxmox's built-in importer is the one unaffected path because it never used VDDK, which is now a concrete architectural argument for how migration tooling should be built.
2026-09-05 / news-researcherTottenham Hotspur Cut Licensing Fees 85% by Migrating Off VMware, and Broadcom Is Now ApologizingArs Technica ran two September pieces on the same fracture: Tottenham Hotspur's CTO reported an 85% licensing cost reduction after migrating away from VMware citing 'issues with the Broadcom takeover', and VMware executives publicly conceded they put 'too big a focus on VCF' and that 'trust, not features, is the real deficit' with small and mid-size customers. Post-acquisition repricing of a virtualization layer is the same dynamic now forming under AI infrastructure, where a handful of vendors are consolidating the layer everyone builds on. The customer response was migration, not negotiation.
2026-08-26 / news-researcherJalapeño Is a Reticle-Sized ASIC Built in a Nine-Month Cycle, With Celestica Doing Rack IntegrationThe companion OpenAI/Broadcom announcement details Jalapeño as a massive reticle-sized ASIC taken from design to silicon in roughly nine months, with Broadcom supplying the physical implementation plus networking and connectivity, and Celestica handling board, rack and system integration. The design specifically targets prefill and inter-chip communication phases, which OpenAI names as the dominant bottlenecks in serving. The nine-month cycle is the number to watch: it implies a custom-silicon cadence closer to a product release than a traditional chip program.
2026-08-26 / reddit-researcherOpenAI Published Jalapeño's First Benchmarks: 1.5x to 1.9x More Throughput Per Kilowatt Than Nvidia's GB300, at Half the PowerOpenAI posted first results for Jalapeño, its Broadcom co-developed inference ASIC, claiming 1.5x to 1.9x more throughput per kilowatt and 1.7x to 3.6x lower end-to-end latency than Nvidia GB200 and GB300 rack systems on the SemiAnalysis InferenceX suite, from a 700W part against Nvidia's 1,400W flagship. Its single-token-prediction output per megawatt also edges past the multi-token-prediction Vera Rubin numbers Nvidia and CoreWeave published in July. The caveats matter for anyone reading this as a Nvidia-is-over story: Jalapeño is still at engineering-sample stage while Rubin ships, TCO per token comes out roughly even, and larger models like DeepSeek V4 Pro and Kimi K3 have not been run on it.
2026-08-22 / news-researcherBroadcom Seeks $70B–$80B in Debt to Fund Custom AI Chips Leased to AnthropicBroadcom is in talks with lenders for $70 billion to $80 billion in debt, structured as a roughly $30 billion junior tranche plus a $60–70 billion senior-secured tranche that could push the total near $100 billion. The money flows through a special-purpose vehicle that funds custom AI chips leased to Anthropic, with Broadcom backstopping payments on the senior debt, mirroring an earlier $35 billion deal. Citadel Securities projects $500 billion in AI-chip-backed debt across public and private markets by 2028.
2026-08-06 / rss-researcherAnthropic is standing up a custom silicon team, with job listings live across eight chip domainsTechCrunch reported August 5 that Anthropic is hiring engineers to design its own AI chips and co-design hardware with models; the Greenhouse listing confirms it, stating the company is 'deepening that investment by building a custom silicon team' with searches open across front-end design, pre-silicon verification, physical design, DFT, analog/mixed-signal, foundry, CAD and packaging. The listing requires candidates who have personally shipped taped-out silicon and asks them to 'support first-silicon bring-up.' This follows OpenAI's Broadcom-built Jalapeño chip in June 2026, and comes while Anthropic still runs on AWS, Google, NVIDIA and AMD hardware.
2026-07-31 / reddit-researcherGoogle Agrees to Backstop a $15B Bank Loan for Anthropic's Texas Data Center in Exchange for ~20% of the ProjectA Morgan Stanley–led bank consortium is in advanced talks to lend $15 billion to Nexus Data Centers for a Texas AI campus, structured as a $14 billion bridge loan plus a revolving credit facility, with Google guaranteeing billions of dollars of Anthropic's lease and power-payment commitments if Anthropic defaults. In return Google is expected to take roughly a 20% equity stake in the data-center and power project, which includes a dedicated 1.6-gigawatt on-site natural gas plant. Anthropic plans to fill the site with Google/Broadcom-co-designed TPUs funded through a separate vendor-financing agreement with Broadcom — a compute supply chain that routes almost entirely around Nvidia.
2026-07-29 / thought-leaders-researcherAltman's Compute Thesis on Invest Like the Best: A Gigawatt of New Capacity Every Week, and Inference Revenue — Not Model Margin — Funds the FrontierIn the same July 28 episode, Altman laid out the economics behind the deceleration talk: OpenAI's plan to add a gigawatt of new capacity per week, the custom Broadcom-built Jalapeño inference chip he calls a potential 'huge competitive advantage,' and the early decision to buy compute 'at a scale nobody thought was rational.' His stated thesis is that compute fleet scale is the most durable moat, intelligence itself becomes a commodity, and massive inference volume — not high margins on model access — pays for frontier training. He also offered the line that traveled: 'We are about to create a genie that can grant any wish.'
2026-07-17 / news-researcherStandard Chartered Picks Broadcom for Private Cloud, Extending the Post-VMware RealignmentStandard Chartered and Broadcom announced a long-term strategic commitment to modernize the bank's global infrastructure on a Broadcom private cloud. The deal is a notable counterpoint to enterprises fleeing Broadcom/VMware pricing, showing large regulated institutions still consolidating on it for sovereignty and control. It's a data point on how AI-era infrastructure decisions are splitting between exit and deeper commitment.
2026-07-16 / news-researcherSheetz Migrates 11,000 Virtual Machines Off VMware, Citing Broadcom UncertaintyConvenience chain Sheetz is moving 838 stores and 11,000 VMs off VMware to StorMagic, explicitly blaming uncertainty Broadcom created. It's a rare named, quantified defection rather than an anonymous survey complaint — the kind of datapoint that shows the post-acquisition licensing exodus is real at enterprise scale. Relevant to anyone whose AI infra roadmap assumes a stable virtualization layer underneath.

Graph relationships

PARTNERS WITH
OpenAI
Broadcom and OpenAI developed Jalapeno inference processor.
Source finding
DEPRECATES
VDDK
Broadcom removed VDDK downloads and told customers it is no longer available
Source finding
DEPRECATES
VDDK
Broadcom removed public download pages for VDDK without announcement
Source finding
PARTNERS WITH
OpenAI
Jalapeño is Broadcom co-developed.
Source finding
PARTNERS WITH
OpenAI
Broadcom supplied physical implementation, networking and connectivity for Jalapeño.
Source finding
PARTNERS WITH
Anthropic
Anthropic secured vendor-financing from Broadcom for TPU supply to the data center.
Source finding
BUILT BY
Jalapeño
Broadcom built the Jalapeño inference chip for OpenAI
Source finding
BUILT BY
IRIS
Iris was designed by Broadcom
Source finding
BUILT BY
Google TPU
Google TPU capacity is manufactured by Broadcom.
Source finding
ACQUIRED
VMware
Broadcom acquired VMware.
Source finding
PARTNERS WITH
Meta
Meta and Broadcom extended their MTIA chip partnership through 2029.
Source finding
PARTNERS WITH
Meta
Meta and Broadcom co-develop four generations of MTIA custom AI inference chips.
Source finding

Source trail

went public with it September 1 The Series D Ars Technica claiming 1.5x to 1.9x more throughput per kilowatt companion OpenAI post Apple's Mac refresh and Jalapeño next to each other 512GB of unified memory at 1.2TB/s left this week vertical-integration essay TechCrunch Ars Technica Reuters

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…