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 release lands after this run’s window. 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).