2026-10-01 — Broadcom’s financing role in Anthropic’s compute buildout comes into view

• Broadcom has agreed to lend Anthropic up to $42 billion to finance infrastructure spending, according to Anthropic’s IPO prospectus as described by Reuters, and Anthropic is expected to become Broadcom’s largest compute customer in 2027 The convertible note could fund about a third of the $125.2 billion Anthropic has committed to a five-year lease of Google-designed TPU capacity, Anthropic warns that Broadcom’s dual role as supplier and financing partner creates “potential conflicts of interest,” and Broadcom projects AI semiconductor revenue of about $115 billion in fiscal 2027. (Reuters via Dunya News)
• A Broadcom-led bank syndicate is gathering $60 billion of AI chip financing to benefit Anthropic and other companies, Bloomberg reported, starting with a $42 billion senior-secured tranche Banks are poised to send syndication letters for the $42 billion Class A tranche and Blackstone is leading an $18 billion junior tranche, committing $9 billion from its funds; the deal has not been announced. (Bloomberg via Economic Times)

2026-09-23 — China surveys Broadcom switch use in state-backed data centres

• China’s state asset regulator has spent recent weeks surveying how widely Broadcom’s switches are used in state-controlled data centres, and the survey found they could account for as much as 90% of the equipment, the Financial Times reported on 2026-09-23. The review is part of Beijing’s “domestic chips for domestic use” push; Nvidia products are already barred from those facilities while Broadcom’s remain widely present. Reuters could not verify the report and neither Broadcom nor the regulator had commented. It concerns Broadcom’s networking business, not its custom-chip or VMware tolls. (Reuters via WHBL, Reuters)

2026-09-16 — A European cloud watchdog says Broadcom’s conduct toward VMware cloud partners has worsened and accuses it of obstructing the Commission’s probe

• The European Cloud Competition Observatory, set up by the cloud trade group CISPE, said in its annual report on 2026-09-16 that Broadcom’s conduct toward European VMware cloud providers had “deteriorated markedly” over the past year. It attributes this to Broadcom closing its Advantage Partner Program for VMware cloud service providers in Europe in favour of a more selective scheme, calls it a “serious and continuing attempt to restructure the market in its favour,” and accuses Broadcom of obstructing the European Commission’s investigation into its VMware licensing. Broadcom’s response was not in the coverage read. (The Stack)

2026-09-11 — EU competition officials escalate scrutiny of Broadcom’s VMware licensing

• European Commission competition officials have escalated their scrutiny of Broadcom’s VMware licensing changes, questioning European cloud providers on how replaceable VMware is and on any certification terms they consider unfair, Bloomberg reported on 2026-09-11. The questions went out in July; the Commission could still open a formal investigation or seek interim measures. It is a new regulatory front on Broadcom’s VMware licensing business, alongside the Tesco litigation. (SDxCentral, Tech Policy Law)

2026-09-08 — Amazon buys a second custom-silicon path, not from Broadcom

• Amazon and Qualcomm announced a multi-generational collaboration on custom AI data-centre silicon, and Qualcomm’s stock rose about 10%. Qualcomm issued Amazon a warrant for up to 25 million of its own shares as part of the deal. ⚠️ The “$4 billion” figure in most headlines is the value of that warrant, not cash consideration and not a disclosed contract size. The substance for this thread: Amazon already designs Trainium in-house, and this deal is Amazon buying a second custom-silicon path through Qualcomm rather than deepening the first — a customer-base data point for the ASIC-toll model that sits outside Broadcom’s own five/six-customer roster this thread otherwise tracks. (CNBC)

2026-09-02 — Broadcom’s Q3 date firms to today, after the close

• Broadcom confirmed through its own investor-relations release that it reports Q3 FY2026 results Wednesday, September 2, after market close, with the call at 5:00pm ET. That firms this thread’s “next earnings (~early Sept, unconfirmed)” watch line to a company-stated date. No numbers exist yet. The three things to check against this thread when it does: Jalapeño tape-out status, whether a sixth co-design customer joins the five already tracked (Google, Meta, OpenAI, Apple, ByteDance), and any updated backlog figure. Analyst consensus previews point to roughly $29.4bn revenue and ~$16bn of AI semiconductor revenue, but those are estimates and are recorded here as context, not as fact. (Broadcom investor relations)

• Broadcom reported record fiscal Q3 2026 results after Wednesday’s close — $29.6bn total revenue (+86% YoY) and $16.7bn AI semiconductor revenue (+221% YoY), both above the roughly $29.4bn/$16bn consensus this map had recorded — but Q4 guidance of $34.8bn total and $21.7bn AI revenue came in just below the ~$35bn Street estimate, and the stock fell about 5% after hours before closing down 3.00% at $356.22 on Thursday (down from Wednesday’s $367.47 close, off as much as 6.3% intraday). CEO Hock Tan raised multi-year AI targets to roughly $115bn in fiscal 2027 and $230bn in fiscal 2028, saying customer demand already exceeds those figures and the binding constraint is fab and packaging capacity, not orders. (PR Newswire, primary release, CNBC, Motley Fool)

• On the call Tan said Jalapeño — OpenAI’s custom inference chip — shipped in production during the quarter and the next-generation successor is now “approaching tape-out,” a real step forward from the June unveiling; he also claimed Broadcom now has “six XPU customers,” naming Google, Meta, OpenAI and, as a customer newly counted in its own right rather than folded into Google’s TPU relationship, Anthropic — which Tan said is “on track to become our largest XPU customer” by 2027-2028 (scaling from 1GW in 2026 to over 10GW by 2028) on the strength of roughly $21bn in TPU-rack orders placed this quarter. Apple and ByteDance, the map’s other two tracked co-design names, went unmentioned; the identity of a true sixth name beyond the five already tracked remains undisclosed. No updated aggregate AI-backlog figure was given. (SiliconANGLE, Benzinga)

2026-08-25 — Jalapeño’s first public benchmarks land at Hot Chips

• OpenAI and Broadcom published Jalapeño’s first performance benchmarks at Hot Chips 2026, claiming better performance-per-watt than Nvidia’s flagship. The inference ASIC the two announced in June was disclosed at 700W TDP, with a 128-chip pod at 1.7 exaflops in 4-bit and 27.5TB of HBM4, and claimed advantages up to 1.9x throughput-per-kilowatt and 3.6x lower latency on SemiAnalysis’s InferenceX benchmark; small-scale deployment inside OpenAI is targeted for end-2026, broader rollout 2027. The June item was a partnership announcement — this is the first evidence the silicon works. (OpenAI, TechCrunch)

2026-07-31 — MediaTek buys into the toll layer with a $5B financing budget

• MediaTek’s board approved a $5B discretionary financing budget for AI data-center ASICs, disclosed on its Q2 earnings call — CEO Rick Tsai sized the custom-AI-chip market at $80B by 2027 (raised from a prior $70-80B range) and set a 15-20% share target, against smartphone chip revenue down 20% in the quarter. A second serious bidder for the toll position this thread tracks, funding it explicitly rather than opportunistically. (The Star)

2026-07-25 — A new toll layer: Broadcom and Samsung sign a >$200B MOU through 2030

• Broadcom and Samsung signed a memorandum of understanding worth more than $200B through 2030 — announced 2026-07-25 at an event in San Francisco. Note the instrument: an MOU, not a definitive contract, which is why no 8-K accompanies it. Three parts: Samsung supplies Broadcom HBM4/HBM4E memory (the primary driver — HBM costs up ~20% in Q1), Samsung Foundry runs 2nm-and-below wafer manufacturing for Broadcom, and advanced packaging (2.3D/2.5D) is included rather than merely “expected to extend.” (Samsung Global Newsroom, primary)
• A genuinely new toll layer, distinct from the co-design customer book (Google/Meta/OpenAI/Apple/ByteDance): memory-supply lock-in plus a foundry hedge beyond TSMC-exclusivity. Cross-references Memory Squeeze (the HBM-cost driver) as much as this thread’s own ASIC-customer angle.

2026-07-06/13 — Apple joins

• Apple confirmed as the fifth co-design customer — $30B through 2031 (8+ outlets converge).

2026-06-24 — Jalapeño unveiled

• OpenAI’s Jalapeño went from rumor to unveiled product (OpenAI newsroom): LLM-inference chip, gigawatt-scale target from 2026 — no tape-out/production date confirmed.

2026-06-17/18 — The second toll litigated

• Tesco pulled 40,000 servers off VMware and sued for >£100M — the licensing backlash became litigation; “VMware refugees” coverage continues through late July.

2026-04-14 — Meta to 2029

• Meta extended the MTIA co-design through 2029 (1GW+); Hock Tan left Meta’s board the same week — resolving the optic.

2025-12 → 2026-04 — The Google book

• A $21B Google TPU order (Dec-25) plus Anthropic’s multi-gigawatt Google+Broadcom TPU expansion (04-07/08).