Broadcom reported fiscal third quarter revenue of $29.6 billion on 2 September 2026, up 86 percent year over year, with AI semiconductor revenue of $16.7 billion that grew 221 percent year over year and 54 percent quarter over quarter.1 Broadcom Inc. (PR Newswire) 2026-09-02 Q3 FY2026 revenue $29.6 billion up 86 percent; semiconductor solutions $20.8 billion up 127 percent; infrastructure software $8.8 billion up 29 percent; AI semiconductor revenue $16.7 billion up 221 percent year over year and 54 percent sequentially; non-GAAP operating income $20.1 billion up 92 percent; GAAP net income $13.1 billion; free cash flow $13.7 billion at 46 percent of revenue; Q4 guidance $34.8 billion with AI semiconductor revenue of $21.7 billion up 236 percent; dividend $0.65 per share. Open source On the call the company went further than a quarter, guiding AI revenue to $58 billion in fiscal 2026, roughly $115 billion in fiscal 2027 and roughly $230 billion in fiscal 2028, a doubling in each of the next two years.2 Yahoo Finance 2026-09-02 AI at 56 percent of total revenue; XPU shipments up 3.5 times and 73 percent of AI revenue; AI revenue targets of $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028; Google Ironwood TPU v7 and TPU v8i at multi tens of billions annually; Anthropic 5 gigawatts of TPU v8i in 2027 and 10 gigawatts in 2028; OpenAI Jalapeno at 1.3 gigawatts in 2027; Meta MTIA production shipments in Q4; constraints named as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory. Open source The stake is that custom silicon has stopped being a hedge against Nvidia and has become the volume path for the largest model developers, and Broadcom is now the single supplier through which four of them route. We assess with moderate confidence that the ladder to $230 billion is more likely to be limited by power, wafers and memory than by cancelled orders, and that the risk in the story has moved from demand to delivery.2 Yahoo Finance 2026-09-02 AI at 56 percent of total revenue; XPU shipments up 3.5 times and 73 percent of AI revenue; AI revenue targets of $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028; Google Ironwood TPU v7 and TPU v8i at multi tens of billions annually; Anthropic 5 gigawatts of TPU v8i in 2027 and 10 gigawatts in 2028; OpenAI Jalapeno at 1.3 gigawatts in 2027; Meta MTIA production shipments in Q4; constraints named as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory. Open source

What the quarter actually shows

The mix is the news, not the headline number. Semiconductor solutions produced $20.8 billion of the $29.6 billion, up 127 percent, while infrastructure software produced $8.8 billion, up 29 percent.1 Broadcom Inc. (PR Newswire) 2026-09-02 Q3 FY2026 revenue $29.6 billion up 86 percent; semiconductor solutions $20.8 billion up 127 percent; infrastructure software $8.8 billion up 29 percent; AI semiconductor revenue $16.7 billion up 221 percent year over year and 54 percent sequentially; non-GAAP operating income $20.1 billion up 92 percent; GAAP net income $13.1 billion; free cash flow $13.7 billion at 46 percent of revenue; Q4 guidance $34.8 billion with AI semiconductor revenue of $21.7 billion up 236 percent; dividend $0.65 per share. Open source Inside the semiconductor line, AI was $16.7 billion, or 56 percent of the whole company.2 Yahoo Finance 2026-09-02 AI at 56 percent of total revenue; XPU shipments up 3.5 times and 73 percent of AI revenue; AI revenue targets of $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028; Google Ironwood TPU v7 and TPU v8i at multi tens of billions annually; Anthropic 5 gigawatts of TPU v8i in 2027 and 10 gigawatts in 2028; OpenAI Jalapeno at 1.3 gigawatts in 2027; Meta MTIA production shipments in Q4; constraints named as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory. Open source The composition inside AI matters more still: XPU custom accelerators, the parts Broadcom co designs for Google, Anthropic, OpenAI and Meta, grew shipments more than 3.5 times year over year and accounted for 73 percent of AI revenue.3 TIKR 2026-09-03 Revenue $29.6 billion up 85.5 percent, beating consensus by about half a percent; adjusted EPS $3.32 versus $3.24 expected; XPU accelerators for Google, Anthropic, OpenAI and Meta up more than 3.5 times and 73 percent of AI revenue; Ironwood TPU v7 shipped in volume to Google and Anthropic; OpenAI Jalapeno runs inference at less than half the cost of a GPU; Anthropic tied to 15 gigawatts of TPU additions across 2027 and 2028; operating margin 67.9 percent, up 240 basis points; Hock Tan on shipping more if supply can be secured. Open source Networking, the business that carried the AI story in earlier quarters, is now the minority of it.

The economics moved with the mix. Non-GAAP operating income reached $20.1 billion, up 92 percent, ahead of the 86 percent revenue growth, and GAAP net income was $13.1 billion.1 Broadcom Inc. (PR Newswire) 2026-09-02 Q3 FY2026 revenue $29.6 billion up 86 percent; semiconductor solutions $20.8 billion up 127 percent; infrastructure software $8.8 billion up 29 percent; AI semiconductor revenue $16.7 billion up 221 percent year over year and 54 percent sequentially; non-GAAP operating income $20.1 billion up 92 percent; GAAP net income $13.1 billion; free cash flow $13.7 billion at 46 percent of revenue; Q4 guidance $34.8 billion with AI semiconductor revenue of $21.7 billion up 236 percent; dividend $0.65 per share. Open source TIKR puts operating margin at 67.9 percent, an expansion of 240 basis points year over year.3 TIKR 2026-09-03 Revenue $29.6 billion up 85.5 percent, beating consensus by about half a percent; adjusted EPS $3.32 versus $3.24 expected; XPU accelerators for Google, Anthropic, OpenAI and Meta up more than 3.5 times and 73 percent of AI revenue; Ironwood TPU v7 shipped in volume to Google and Anthropic; OpenAI Jalapeno runs inference at less than half the cost of a GPU; Anthropic tied to 15 gigawatts of TPU additions across 2027 and 2028; operating margin 67.9 percent, up 240 basis points; Hock Tan on shipping more if supply can be secured. Open source Free cash flow of $13.7 billion equalled 46 percent of revenue.1 Broadcom Inc. (PR Newswire) 2026-09-02 Q3 FY2026 revenue $29.6 billion up 86 percent; semiconductor solutions $20.8 billion up 127 percent; infrastructure software $8.8 billion up 29 percent; AI semiconductor revenue $16.7 billion up 221 percent year over year and 54 percent sequentially; non-GAAP operating income $20.1 billion up 92 percent; GAAP net income $13.1 billion; free cash flow $13.7 billion at 46 percent of revenue; Q4 guidance $34.8 billion with AI semiconductor revenue of $21.7 billion up 236 percent; dividend $0.65 per share. Open source 5 24/7 Wall St. 2026-09-02 Revenue $29.59 billion versus a $29.44 billion estimate; adjusted EPS $3.32 beating $3.24 by 2.5 percent; free cash flow $13.66 billion at 46 percent of revenue; Q4 AI semiconductor guidance $21.7 billion up 236 percent; $0.65 quarterly dividend; VMware based private AI cloud platforms announced for enterprise infrastructure; ninth consecutive earnings beat. Open source That combination, a 3.5 times shipment increase in custom parts alongside margin expansion, is the point analysts had been waiting on. The standard argument against custom accelerators is that they are engineering services dressed as silicon and should compress margins as volume grows. This quarter it did the opposite.

The ladder, and who is on it

Management named the customers and, in several cases, the deployments behind the forward numbers. Google took Ironwood TPU v7 and TPU v8i shipments described on the call as multi tens of billions annually.2 Yahoo Finance 2026-09-02 AI at 56 percent of total revenue; XPU shipments up 3.5 times and 73 percent of AI revenue; AI revenue targets of $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028; Google Ironwood TPU v7 and TPU v8i at multi tens of billions annually; Anthropic 5 gigawatts of TPU v8i in 2027 and 10 gigawatts in 2028; OpenAI Jalapeno at 1.3 gigawatts in 2027; Meta MTIA production shipments in Q4; constraints named as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory. Open source Anthropic is tied to 5 gigawatts of TPU v8i deployment in 2027 and a further 10 gigawatts in 2028, 15 gigawatts across the two years.2 Yahoo Finance 2026-09-02 AI at 56 percent of total revenue; XPU shipments up 3.5 times and 73 percent of AI revenue; AI revenue targets of $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028; Google Ironwood TPU v7 and TPU v8i at multi tens of billions annually; Anthropic 5 gigawatts of TPU v8i in 2027 and 10 gigawatts in 2028; OpenAI Jalapeno at 1.3 gigawatts in 2027; Meta MTIA production shipments in Q4; constraints named as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory. Open source 3 TIKR 2026-09-03 Revenue $29.6 billion up 85.5 percent, beating consensus by about half a percent; adjusted EPS $3.32 versus $3.24 expected; XPU accelerators for Google, Anthropic, OpenAI and Meta up more than 3.5 times and 73 percent of AI revenue; Ironwood TPU v7 shipped in volume to Google and Anthropic; OpenAI Jalapeno runs inference at less than half the cost of a GPU; Anthropic tied to 15 gigawatts of TPU additions across 2027 and 2028; operating margin 67.9 percent, up 240 basis points; Hock Tan on shipping more if supply can be secured. Open source OpenAI's first custom part, Jalapeno, is expected at 1.3 gigawatts of deployment in 2027, and Broadcom described it as running inference at less than half the cost of a GPU.2 Yahoo Finance 2026-09-02 AI at 56 percent of total revenue; XPU shipments up 3.5 times and 73 percent of AI revenue; AI revenue targets of $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028; Google Ironwood TPU v7 and TPU v8i at multi tens of billions annually; Anthropic 5 gigawatts of TPU v8i in 2027 and 10 gigawatts in 2028; OpenAI Jalapeno at 1.3 gigawatts in 2027; Meta MTIA production shipments in Q4; constraints named as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory. Open source 3 TIKR 2026-09-03 Revenue $29.6 billion up 85.5 percent, beating consensus by about half a percent; adjusted EPS $3.32 versus $3.24 expected; XPU accelerators for Google, Anthropic, OpenAI and Meta up more than 3.5 times and 73 percent of AI revenue; Ironwood TPU v7 shipped in volume to Google and Anthropic; OpenAI Jalapeno runs inference at less than half the cost of a GPU; Anthropic tied to 15 gigawatts of TPU additions across 2027 and 2028; operating margin 67.9 percent, up 240 basis points; Hock Tan on shipping more if supply can be secured. Open source Meta's MTIA accelerator enters production shipments in the fourth quarter.2 Yahoo Finance 2026-09-02 AI at 56 percent of total revenue; XPU shipments up 3.5 times and 73 percent of AI revenue; AI revenue targets of $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028; Google Ironwood TPU v7 and TPU v8i at multi tens of billions annually; Anthropic 5 gigawatts of TPU v8i in 2027 and 10 gigawatts in 2028; OpenAI Jalapeno at 1.3 gigawatts in 2027; Meta MTIA production shipments in Q4; constraints named as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory. Open source These are the company's own characterisations of customer plans rather than disclosed contracts, and the customers have not separately confirmed the gigawatt figures, so they should be read as supplier guidance rather than audited commitments.

Read the cost claim carefully, because it is the load bearing one. If a custom inference part genuinely lands at under half the cost per unit of work of a merchant GPU, then for any lab whose inference volume is large and whose model architecture is stable, the decision to tape out its own silicon stops being strategic and becomes arithmetic.3 TIKR 2026-09-03 Revenue $29.6 billion up 85.5 percent, beating consensus by about half a percent; adjusted EPS $3.32 versus $3.24 expected; XPU accelerators for Google, Anthropic, OpenAI and Meta up more than 3.5 times and 73 percent of AI revenue; Ironwood TPU v7 shipped in volume to Google and Anthropic; OpenAI Jalapeno runs inference at less than half the cost of a GPU; Anthropic tied to 15 gigawatts of TPU additions across 2027 and 2028; operating margin 67.9 percent, up 240 basis points; Hock Tan on shipping more if supply can be secured. Open source That is the mechanism underneath the ladder. It also explains why the ladder is expressed in gigawatts rather than units: at this scale the customers are buying deployable power envelopes, and the chip is the thing that decides how much useful work fits inside one.

The constraint moved, and so did the ledger

The most consequential sentence of the call was not a revenue figure. Management named the limits on deployment as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory.2 Yahoo Finance 2026-09-02 AI at 56 percent of total revenue; XPU shipments up 3.5 times and 73 percent of AI revenue; AI revenue targets of $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028; Google Ironwood TPU v7 and TPU v8i at multi tens of billions annually; Anthropic 5 gigawatts of TPU v8i in 2027 and 10 gigawatts in 2028; OpenAI Jalapeno at 1.3 gigawatts in 2027; Meta MTIA production shipments in Q4; constraints named as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory. Open source Hock Tan said the company could ship significantly more if it could secure more supply.3 TIKR 2026-09-03 Revenue $29.6 billion up 85.5 percent, beating consensus by about half a percent; adjusted EPS $3.32 versus $3.24 expected; XPU accelerators for Google, Anthropic, OpenAI and Meta up more than 3.5 times and 73 percent of AI revenue; Ironwood TPU v7 shipped in volume to Google and Anthropic; OpenAI Jalapeno runs inference at less than half the cost of a GPU; Anthropic tied to 15 gigawatts of TPU additions across 2027 and 2028; operating margin 67.9 percent, up 240 basis points; Hock Tan on shipping more if supply can be secured. Open source Every item on that list sits outside Broadcom's control, at TSMC, at substrate makers, at the three high bandwidth memory suppliers, and at utilities. We assess with moderate confidence that from here the fiscal 2027 and 2028 numbers are better forecast by tracking those four inputs than by tracking model lab announcements.

Who gains. Anthropic and OpenAI gain the most durable advantage, because a part that halves inference cost compounds against every token they serve, and both have now secured multi year positions in Broadcom's capacity queue rather than in a merchant allocation list.2 Yahoo Finance 2026-09-02 AI at 56 percent of total revenue; XPU shipments up 3.5 times and 73 percent of AI revenue; AI revenue targets of $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028; Google Ironwood TPU v7 and TPU v8i at multi tens of billions annually; Anthropic 5 gigawatts of TPU v8i in 2027 and 10 gigawatts in 2028; OpenAI Jalapeno at 1.3 gigawatts in 2027; Meta MTIA production shipments in Q4; constraints named as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory. Open source 3 TIKR 2026-09-03 Revenue $29.6 billion up 85.5 percent, beating consensus by about half a percent; adjusted EPS $3.32 versus $3.24 expected; XPU accelerators for Google, Anthropic, OpenAI and Meta up more than 3.5 times and 73 percent of AI revenue; Ironwood TPU v7 shipped in volume to Google and Anthropic; OpenAI Jalapeno runs inference at less than half the cost of a GPU; Anthropic tied to 15 gigawatts of TPU additions across 2027 and 2028; operating margin 67.9 percent, up 240 basis points; Hock Tan on shipping more if supply can be secured. Open source TSMC and the high bandwidth memory suppliers gain pricing power, since Broadcom has publicly stated it is supply limited, which is the weakest possible position from which to negotiate. Broadcom shareholders gained a visible three year revenue frame that few semiconductor companies can offer.2 Yahoo Finance 2026-09-02 AI at 56 percent of total revenue; XPU shipments up 3.5 times and 73 percent of AI revenue; AI revenue targets of $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028; Google Ironwood TPU v7 and TPU v8i at multi tens of billions annually; Anthropic 5 gigawatts of TPU v8i in 2027 and 10 gigawatts in 2028; OpenAI Jalapeno at 1.3 gigawatts in 2027; Meta MTIA production shipments in Q4; constraints named as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory. Open source

Who loses. Nvidia loses the inference half of the market's growth first, which is the half with stable workloads and the least need for general purpose flexibility. Smaller model developers lose relative position, because custom silicon at these volumes requires committing to a design two years before the workload exists, and only firms with Anthropic or OpenAI scale can carry that risk. Broadcom's own enterprise customers may lose too: the company used the same week to announce VMware based private AI cloud platforms, and an engineering organisation absorbed by four hyperscale roadmaps is not obviously the one to build out a fifth business.5 24/7 Wall St. 2026-09-02 Revenue $29.59 billion versus a $29.44 billion estimate; adjusted EPS $3.32 beating $3.24 by 2.5 percent; free cash flow $13.66 billion at 46 percent of revenue; Q4 AI semiconductor guidance $21.7 billion up 236 percent; $0.65 quarterly dividend; VMware based private AI cloud platforms announced for enterprise infrastructure; ninth consecutive earnings beat. Open source

The counter-case

The market did not fully accept the story. Fourth quarter revenue guidance of $34.8 billion landed slightly below an analyst consensus near $35 billion, and the shares did not respond to a quarter with 221 percent AI growth the way the number alone would suggest.4 The Motley Fool 2026-09-03 Fiscal Q3 ended 2 August 2026; revenue growth accelerated from 48 percent in Q2 to 86 percent; adjusted EPS $3.32 up 96 percent and GAAP diluted EPS $2.68 up 215 percent; fourteenth consecutive quarter of AI centric growth; Q4 guidance of $34.8 billion below analyst consensus near $35 billion; Hock Tan calls demand for custom AI accelerators and networking very strong; CFO Amie Thuener cites record revenue, operating profit and free cash flow; shares at $357.90. Open source The honest reading of that is not scepticism about this quarter but about concentration: four customers, one of which is unprofitable at the model layer, underwrite a $230 billion figure that is two fiscal years out.2 Yahoo Finance 2026-09-02 AI at 56 percent of total revenue; XPU shipments up 3.5 times and 73 percent of AI revenue; AI revenue targets of $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028; Google Ironwood TPU v7 and TPU v8i at multi tens of billions annually; Anthropic 5 gigawatts of TPU v8i in 2027 and 10 gigawatts in 2028; OpenAI Jalapeno at 1.3 gigawatts in 2027; Meta MTIA production shipments in Q4; constraints named as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory. Open source A single customer deferring one gigawatt of 2027 deployment removes more revenue than most semiconductor companies earn in a year.

For our assessment to be wrong, one of two things has to be true. Either the demand is softer than described, in which case a supply constrained company would quietly stop being supply constrained and the language would change before the numbers do. Or the constraint binds so hard that the ladder simply stretches: 15 gigawatts of Anthropic capacity is a utility scale interconnection problem, and grid queues do not accelerate because a supplier guided to them.2 Yahoo Finance 2026-09-02 AI at 56 percent of total revenue; XPU shipments up 3.5 times and 73 percent of AI revenue; AI revenue targets of $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028; Google Ironwood TPU v7 and TPU v8i at multi tens of billions annually; Anthropic 5 gigawatts of TPU v8i in 2027 and 10 gigawatts in 2028; OpenAI Jalapeno at 1.3 gigawatts in 2027; Meta MTIA production shipments in Q4; constraints named as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory. Open source Note also that the fiscal 2027 and 2028 figures are management targets stated on a call, not booked backlog disclosed in a filing.2 Yahoo Finance 2026-09-02 AI at 56 percent of total revenue; XPU shipments up 3.5 times and 73 percent of AI revenue; AI revenue targets of $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028; Google Ironwood TPU v7 and TPU v8i at multi tens of billions annually; Anthropic 5 gigawatts of TPU v8i in 2027 and 10 gigawatts in 2028; OpenAI Jalapeno at 1.3 gigawatts in 2027; Meta MTIA production shipments in Q4; constraints named as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory. Open source

What to watch

  • Whether the fourth quarter AI number lands at $21.7 billion. Guidance is for 236 percent year over year growth in the quarter ending in November.1 Broadcom Inc. (PR Newswire) 2026-09-02 Q3 FY2026 revenue $29.6 billion up 86 percent; semiconductor solutions $20.8 billion up 127 percent; infrastructure software $8.8 billion up 29 percent; AI semiconductor revenue $16.7 billion up 221 percent year over year and 54 percent sequentially; non-GAAP operating income $20.1 billion up 92 percent; GAAP net income $13.1 billion; free cash flow $13.7 billion at 46 percent of revenue; Q4 guidance $34.8 billion with AI semiconductor revenue of $21.7 billion up 236 percent; dividend $0.65 per share. Open source A miss driven by supply rather than demand is the first real test of whether the constraint story is accurate or convenient.
  • Whether XPU share of AI revenue holds above 70 percent. It reached 73 percent this quarter.3 TIKR 2026-09-03 Revenue $29.6 billion up 85.5 percent, beating consensus by about half a percent; adjusted EPS $3.32 versus $3.24 expected; XPU accelerators for Google, Anthropic, OpenAI and Meta up more than 3.5 times and 73 percent of AI revenue; Ironwood TPU v7 shipped in volume to Google and Anthropic; OpenAI Jalapeno runs inference at less than half the cost of a GPU; Anthropic tied to 15 gigawatts of TPU additions across 2027 and 2028; operating margin 67.9 percent, up 240 basis points; Hock Tan on shipping more if supply can be secured. Open source If that share falls back through 2027 while total AI revenue still grows, networking is carrying the ladder and the custom silicon thesis is weaker than presented.
  • Whether operating margin stays near 68 percent as XPU volume doubles. Margin expanded 240 basis points this quarter.3 TIKR 2026-09-03 Revenue $29.6 billion up 85.5 percent, beating consensus by about half a percent; adjusted EPS $3.32 versus $3.24 expected; XPU accelerators for Google, Anthropic, OpenAI and Meta up more than 3.5 times and 73 percent of AI revenue; Ironwood TPU v7 shipped in volume to Google and Anthropic; OpenAI Jalapeno runs inference at less than half the cost of a GPU; Anthropic tied to 15 gigawatts of TPU additions across 2027 and 2028; operating margin 67.9 percent, up 240 basis points; Hock Tan on shipping more if supply can be secured. Open source Compression through fiscal 2027 would show that custom design economics degrade with scale after all.
  • Whether any customer confirms its own gigawatt figure. The 5 and 10 gigawatt Anthropic numbers and the 1.3 gigawatt OpenAI number came from the supplier.2 Yahoo Finance 2026-09-02 AI at 56 percent of total revenue; XPU shipments up 3.5 times and 73 percent of AI revenue; AI revenue targets of $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028; Google Ironwood TPU v7 and TPU v8i at multi tens of billions annually; Anthropic 5 gigawatts of TPU v8i in 2027 and 10 gigawatts in 2028; OpenAI Jalapeno at 1.3 gigawatts in 2027; Meta MTIA production shipments in Q4; constraints named as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory. Open source Independent confirmation in a customer disclosure or a signed power agreement before mid 2027 would convert supplier guidance into something checkable.
  • Whether high bandwidth memory or substrate supply becomes the named reason for a shortfall. Management listed both.2 Yahoo Finance 2026-09-02 AI at 56 percent of total revenue; XPU shipments up 3.5 times and 73 percent of AI revenue; AI revenue targets of $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028; Google Ironwood TPU v7 and TPU v8i at multi tens of billions annually; Anthropic 5 gigawatts of TPU v8i in 2027 and 10 gigawatts in 2028; OpenAI Jalapeno at 1.3 gigawatts in 2027; Meta MTIA production shipments in Q4; constraints named as data center readiness, power availability, leading edge wafers, substrates and high bandwidth memory. Open source The first quarter in which one of them is cited as the cause of a miss is the quarter the AI buildout stops being a semiconductor design story.

The interesting fact in this quarter is not that a chip company grew 86 percent. It is that its forward numbers are now denominated in gigawatts, and that the ceiling it described sits in substations, wafer starts and memory stacks rather than in anything it designs.