Broadcom’s Battle to Defend Its ASIC Share: Google TPU, MediaTek, and the Mass-Production Hurdles Behind an 80% Share
目录
Disaggregating the Share Debate: A Second Supplier Does Not Mean a 50/50 Revenue Split
Broadcom’s Three Mass-Production Hurdles: Memory, Packaging, and System Validation
From 10GW to 15GW: Broadcom Can Offset Share Losses Through Market Expansion
The $502 Price Target Reflects Delivery Certainty and Customer Diversification
Signals That Would Invalidate the 80% Share Assumption
Five Numbers to Watch
本内容基于公开资料和研报数据整理，不构成任何投资建议，不代表任何个人观点，仅供学习参考，请理性阅读
Google has allocated part of its tensor processing unit (TPU) supply chain to MediaTek, prompting the market to price in a halving of Broadcom’s share. Morgan Stanley’s rebuttal centers on three mass-production requirements: high-bandwidth memory, advanced packaging, and large-scale system validation. Whether Broadcom can retain approximately 80% of revenue share will determine whether it can deliver on its $120 billion AI revenue forecast for FY2027.
Disaggregating the Share Debate: A Second Supplier Does Not Mean a 50/50 Revenue Split
The market has extrapolated directly from “MediaTek’s entry” to “Broadcom’s replacement.” Google’s introduction of a second application-specific integrated circuit (ASIC) partner is unsurprising. Google is highly cost-sensitive and wants to reduce its dependence on a single supplier. It also wants to procure certain commodity components separately, reducing the pass-through revenue paid to Broadcom. MediaTek has secured a credible opportunity for the 3nm tensor processing unit (TPU), and the supply chain’s view that it has entered the program is grounded in fact.
The disagreement concerns the share trajectory. The bearish view is that once MediaTek passes initial validation, Broadcom’s TPU share will quickly fall to 50% and could ultimately be eliminated altogether. Morgan Stanley’s base case is that Broadcom maintains approximately 80% of TPU revenue share over the long term, with MediaTek closer to 15%–20%.
These figures are also more consistent with the companies’ publicly stated long-term targets. Broadcom believes it can retain more than 80% of its serviceable market, while MediaTek has set a long-term ASIC share target of approximately 20%.
Unit shipments will overstate the economic value captured by MediaTek. MediaTek’s solution is priced materially lower, so unit share and revenue share are not directly comparable. Even under the relatively bullish shipment assumptions of Morgan Stanley’s MediaTek team, MediaTek’s revenue share next year would remain closer to 20%. Investors focusing solely on tape-out volumes, wafer allocations, or initial orders could easily overestimate the impact on Broadcom’s revenue.
This debate resembles the earlier discussion around Amazon’s Trainium supply chain. New suppliers can indeed secure meaningful programs, while incumbents may lose some share. However, the market often prices “the customer adding a backup solution” as “the incumbent being reduced to zero.” Morgan Stanley acknowledges MediaTek’s opportunity but wants to see evidence of delivery at scale before cutting its Broadcom estimates.
Broadcom, MediaTek, and the HBM Re-Rating: Google TPU Shipments Raised to 35 Million Units
Broadcom’s Three Mass-Production Hurdles: Memory, Packaging, and System Validation
The first hurdle is high-bandwidth memory (HBM). One of the cost advantages of MediaTek’s solution is that Google can procure certain components separately instead of purchasing them as part of a Broadcom package. This calculation assumes that critical components remain readily available at reasonable prices. HBM supply is currently tight, and spot prices have risen materially, meaning ad hoc procurement may not be cheaper. Broadcom previously secured part of its supply under contract. What once appeared to be a routine procurement arrangement is becoming an advantage in both cost and delivery.
HBM affects more than chip bill-of-materials costs. During shortages, only solutions that can obtain memory on schedule can proceed to system assembly and customer deployment. Even if MediaTek completes its chip design, insufficient or unpredictably priced HBM would make it difficult to deliver the overall cost reduction Google seeks. Broadcom’s advantage now extends from design capabilities to supply assurance.
The second hurdle is advanced packaging. Morgan Stanley expects MediaTek to secure some chip-on-wafer-on-substrate (CoWoS) capacity for minimum-volume production of the 2nm TPU, but uncertainty remains over whether Intel’s embedded multi-die interconnect bridge (EMIB) can support the complexity required by Google. Substrate capacity could also become a bottleneck. The report views EMIB as a potential route to greater scale and lower costs around 2029 if execution proceeds smoothly, while CoWoS remains the more reliable fallback.
This means revenue from the 2nm program cannot be inferred solely from the process node. Passing sample validation, entering limited risk production, and reaching volume production are separated by packaging yields, substrates, memory, system validation, and delivery schedules. Any delay would increase Google’s emphasis on the stability of existing platforms, and Broadcom has already cleared these hurdles.
The third hurdle is large-scale system validation. TPUs must operate continuously in large clusters, and chip performance is only the starting point. Suppliers must also complete system-level certification, ensure that multiple TPU racks operate reliably together, and demonstrate that the new solution will not disrupt Google’s AI infrastructure roadmap. A successful tape-out can generate initial revenue, but it does not prove that thousands of chips, memory modules, networking components, and software will all perform reliably at scale.
Broadcom’s offering therefore includes certainty of mass production. Once customers incorporate training and inference capacity into their capital expenditure schedules, the cost of delays may exceed the procurement savings on individual chips. Google is willing to support a second supplier, but it will also avoid an aggressive transition before the new solution has demonstrated its ability to scale.
From 10GW to 15GW: Broadcom Can Offset Share Losses Through Market Expansion
Broadcom’s growth does not require it to retain 100% of Google TPU share indefinitely. Morgan Stanley forecasts approximately $120 billion of AI revenue for Broadcom in FY2027, corresponding to roughly 10GW of AI deployments.
Google’s internal deployments and external TPU opportunities together account for approximately 75% of deployment volume. At roughly $10 billion–$12 billion per gigawatt, this implies approximately $80 billion of TPU-related revenue.
These figures provide a more practical analytical framework. Even if Broadcom concedes approximately 20% of TPU revenue share, its absolute revenue can still grow as long as Google’s overall deployments continue to expand. Investors need to assess both the total market and market share rather than cutting Broadcom’s entire AI revenue forecast simply because MediaTek wins orders.
The key change in FY2028 will be customer mix. Morgan Stanley believes Broadcom will be capable of supporting at least 15GW of AI deployments by then, with TPU declining to approximately 60% of AI revenue. This lower percentage would reflect the ramp-up of other ASIC customers rather than weakness in Google’s business. Multiple new programs are expected to begin scaling in 2H27, gradually shifting Broadcom’s AI revenue from dependence on a single major customer toward growth across multiple customers.
This shift would reduce the valuation impact of the debate over Google share. Before 2027, Google TPU will remain the primary pillar of Broadcom’s AI revenue. Entering FY2028, the timely mass production of new customer programs will increasingly determine the revenue ceiling. If Broadcom can retain its position as Google’s primary supplier while bringing new customers into volume production, MediaTek capturing approximately 20% share may remain a manageable loss within an expanding market.
Broadcom Earnings Deep Dive: AI ASIC Ramp-Up, Expanding AI Networking Demand, and the High-Margin Picks-and-Shovels Provider Behind the Cloud Giants
The $502 Price Target Reflects Delivery Certainty and Customer Diversification
Morgan Stanley continues to rank Broadcom as its core AI compute exposure behind only Nvidia. The report maintains an Overweight rating and a $502 price target, based on 28 times calendar-year 2027 modeled EPS of $17.92. Relative to the pre-publication closing price of $384.05, the target implies approximately 31% upside.
The bull-, base-, and bear-case price targets are $637, $502, and $308, respectively. The bull case requires new ASIC customers, networking, and software synergies all to exceed expectations. The base case relies on custom silicon, networking, and software driving strong revenue growth in 2026–2027. The bear case assumes that new customers fail to reach volume production, revenue growth falls short of expectations, and valuation multiples contract.
Broadcom’s premium valuation is supported by more than a single TPU order. The company combines data-center switching chips, custom compute, and software cash flow, while new customers can further diversify its AI revenue. The market is willing to pay a premium for this portfolio, provided Broadcom continues to demonstrate supply-chain and mass-production execution. Stable market share is only the first level of validation; converting new programs into revenue on schedule is the second.
Institutional research should also be read in light of potential conflicts of interest. Morgan Stanley disclosed that it had provided advisory services to a Broadcom-related AI financing platform and maintains investment-banking relationships with several companies under coverage. The $502 price target can serve as a valuation scenario, but it cannot replace subsequent shipment, revenue, and supply-chain data.
Signals That Would Invalidate the 80% Share Assumption
Morgan Stanley identifies specific disconfirming evidence that investors can track individually. The first signal would be MediaTek’s chip performance materially exceeding current expectations and establishing a sustained advantage in power consumption, cost, or system efficiency. The second would be Google and MediaTek securing HBM at competitive prices and volumes, eliminating Broadcom’s advantage from contracted supply.
The third signal would be EMIB demonstrating that it can reliably support large-scale production of complex 2nm TPUs, with substrate supply no longer constrained. The fourth would be a transition from risk production to formal mass production that is materially faster than in previous AI accelerator programs. If all four conditions are met in succession, MediaTek would move from a backup supplier to a primary supplier capable of rapid expansion, requiring a reduction in Broadcom’s assumed approximately 80% revenue share.
Conversely, if MediaTek can secure only limited CoWoS capacity, lacks an HBM cost advantage, and faces continued EMIB delays, Google is more likely to maintain a dual-supplier structure while allocating most volume orders to Broadcom. In this scenario, MediaTek’s entry would strengthen Google’s negotiating position without undermining Broadcom’s overall ASIC growth.
Five Numbers to Watch
Broadcom’s TPU revenue share. Approximately 80% is the central assumption underpinning the report. It must be cross-checked against orders, wafer allocations, and revenue across Google’s TPU generations rather than inferred solely from unit shipments.
HBM supply and cost. Whether MediaTek can secure sufficient HBM at competitive prices will determine whether its lower-cost solution can be implemented in practice.
2nm packaging progress. CoWoS allocations, EMIB reliability, and substrate capacity will jointly determine the ceiling for volume production. Focusing only on tape-out timing would be premature.
10GW and $120 billion of AI revenue in FY2027. Deployment volume, value per gigawatt, and approximately $80 billion of TPU revenue must all materialize.
New-customer ramp in 2H27. Whether these programs can lift FY2028 deployments to at least 15GW and reduce TPU to approximately 60% of AI revenue will determine whether Broadcom can shed its single-customer valuation discount.
Broadcom’s ASIC moat has expanded from design capabilities to supply, packaging, and system delivery. MediaTek’s entry into Google’s TPU supply chain will result in real share losses and reduce Broadcom’s pricing power, but the available evidence remains insufficient to support a rapid decline to 50% share or complete replacement. Market expansion and new-customer production ramps can absorb limited share losses. The most important validation over the next two years will be whether Broadcom can continue converting mass-production certainty into revenue.
Source: Morgan Stanley’s July 14, 2026 research report on Broadcom. All market shares, revenue figures, deployment volumes, ratings, and price targets cited herein are institutional forecasts.
Related Reading
TSMC CoWoS Deep Dive: 200kwpm Packaging, US$75bn CapEx, and Redistribution Across CPU, GPU, ASIC, and Optical InterconnectBroadcom’s Battle to Defend Its ASIC Share: Google TPU, MediaTek, and the Mass-Production Hurdles Behind an 80% Share
目录
Disaggregating the Share Debate: A Second Supplier Does Not Mean a 50/50 Revenue Split
Broadcom’s Three Mass-Production Hurdles: Memory, Packaging, and System Validation
From 10GW to 15GW: Broadcom Can Offset Share Losses Through Market Expansion
The $502 Price Target Reflects Delivery Certainty and Customer Diversification
Signals That Would Invalidate the 80% Share Assumption
Five Numbers to Watch
本内容基于公开资料和研报数据整理，不构成任何投资建议，不代表任何个人观点，仅供学习参考，请理性阅读
Google has allocated part of its tensor processing unit (TPU) supply chain to MediaTek, prompting the market to price in a halving of Broadcom’s share. Morgan Stanley’s rebuttal centers on three mass-production requirements: high-bandwidth memory, advanced packaging, and large-scale system validation. Whether Broadcom can retain approximately 80% of revenue share will determine whether it can deliver on its $120 billion AI revenue forecast for FY2027.
Disaggregating the Share Debate: A Second Supplier Does Not Mean a 50/50 Revenue Split
The market has extrapolated directly from “MediaTek’s entry” to “Broadcom’s replacement.” Google’s introduction of a second application-specific integrated circuit (ASIC) partner is unsurprising. Google is highly cost-sensitive and wants to reduce its dependence on a single supplier. It also wants to procure certain commodity components separately, reducing the pass-through revenue paid to Broadcom. MediaTek has secured a credible opportunity for the 3nm tensor processing unit (TPU), and the supply chain’s view that it has entered the program is grounded in fact.
The disagreement concerns the share trajectory. The bearish view is that once MediaTek passes initial validation, Broadcom’s TPU share will quickly fall to 50% and could ultimately be eliminated altogether. Morgan Stanley’s base case is that Broadcom maintains approximately 80% of TPU revenue share over the long term, with MediaTek closer to 15%–20%.
These figures are also more consistent with the companies’ publicly stated long-term targets. Broadcom believes it can retain more than 80% of its serviceable market, while MediaTek has set a long-term ASIC share target of approximately 20%.
Unit shipments will overstate the economic value captured by MediaTek. MediaTek’s solution is priced materially lower, so unit share and revenue share are not directly comparable. Even under the relatively bullish shipment assumptions of Morgan Stanley’s MediaTek team, MediaTek’s revenue share next year would remain closer to 20%. Investors focusing solely on tape-out volumes, wafer allocations, or initial orders could easily overestimate the impact on Broadcom’s revenue.
This debate resembles the earlier discussion around Amazon’s Trainium supply chain. New suppliers can indeed secure meaningful programs, while incumbents may lose some share. However, the market often prices “the customer adding a backup solution” as “the incumbent being reduced to zero.” Morgan Stanley acknowledges MediaTek’s opportunity but wants to see evidence of delivery at scale before cutting its Broadcom estimates.
Broadcom, MediaTek, and the HBM Re-Rating: Google TPU Shipments Raised to 35 Million Units
Broadcom’s Three Mass-Production Hurdles: Memory, Packaging, and System Validation
The first hurdle is high-bandwidth memory (HBM). One of the cost advantages of MediaTek’s solution is that Google can procure certain components separately instead of purchasing them as part of a Broadcom package. This calculation assumes that critical components remain readily available at reasonable prices. HBM supply is currently tight, and spot prices have risen materially, meaning ad hoc procurement may not be cheaper. Broadcom previously secured part of its supply under contract. What once appeared to be a routine procurement arrangement is becoming an advantage in both cost and delivery.
HBM affects more than chip bill-of-materials costs. During shortages, only solutions that can obtain memory on schedule can proceed to system assembly and customer deployment. Even if MediaTek completes its chip design, insufficient or unpredictably priced HBM would make it difficult to deliver the overall cost reduction Google seeks. Broadcom’s advantage now extends from design capabilities to supply assurance.
The second hurdle is advanced packaging. Morgan Stanley expects MediaTek to secure some chip-on-wafer-on-substrate (CoWoS) capacity for minimum-volume production of the 2nm TPU, but uncertainty remains over whether Intel’s embedded multi-die interconnect bridge (EMIB) can support the complexity required by Google. Substrate capacity could also become a bottleneck. The report views EMIB as a potential route to greater scale and lower costs around 2029 if execution proceeds smoothly, while CoWoS remains the more reliable fallback.
This means revenue from the 2nm program cannot be inferred solely from the process node. Passing sample validation, entering limited risk production, and reaching volume production are separated by packaging yields, substrates, memory, system validation, and delivery schedules. Any delay would increase Google’s emphasis on the stability of existing platforms, and Broadcom has already cleared these hurdles.
The third hurdle is large-scale system validation. TPUs must operate continuously in large clusters, and chip performance is only the starting point. Suppliers must also complete system-level certification, ensure that multiple TPU racks operate reliably together, and demonstrate that the new solution will not disrupt Google’s AI infrastructure roadmap. A successful tape-out can generate initial revenue, but it does not prove that thousands of chips, memory modules, networking components, and software will all perform reliably at scale.
Broadcom’s offering therefore includes certainty of mass production. Once customers incorporate training and inference capacity into their capital expenditure schedules, the cost of delays may exceed the procurement savings on individual chips. Google is willing to support a second supplier, but it will also avoid an aggressive transition before the new solution has demonstrated its ability to scale.
From 10GW to 15GW: Broadcom Can Offset Share Losses Through Market Expansion
Broadcom’s growth does not require it to retain 100% of Google TPU share indefinitely. Morgan Stanley forecasts approximately $120 billion of AI revenue for Broadcom in FY2027, corresponding to roughly 10GW of AI deployments.
Google’s internal deployments and external TPU opportunities together account for approximately 75% of deployment volume. At roughly $10 billion–$12 billion per gigawatt, this implies approximately $80 billion of TPU-related revenue.
These figures provide a more practical analytical framework. Even if Broadcom concedes approximately 20% of TPU revenue share, its absolute revenue can still grow as long as Google’s overall deployments continue to expand. Investors need to assess both the total market and market share rather than cutting Broadcom’s entire AI revenue forecast simply because MediaTek wins orders.
The key change in FY2028 will be customer mix. Morgan Stanley believes Broadcom will be capable of supporting at least 15GW of AI deployments by then, with TPU declining to approximately 60% of AI revenue. This lower percentage would reflect the ramp-up of other ASIC customers rather than weakness in Google’s business. Multiple new programs are expected to begin scaling in 2H27, gradually shifting Broadcom’s AI revenue from dependence on a single major customer toward growth across multiple customers.
This shift would reduce the valuation impact of the debate over Google share. Before 2027, Google TPU will remain the primary pillar of Broadcom’s AI revenue. Entering FY2028, the timely mass production of new customer programs will increasingly determine the revenue ceiling. If Broadcom can retain its position as Google’s primary supplier while bringing new customers into volume production, MediaTek capturing approximately 20% share may remain a manageable loss within an expanding market.
Broadcom Earnings Deep Dive: AI ASIC Ramp-Up, Expanding AI Networking Demand, and the High-Margin Picks-and-Shovels Provider Behind the Cloud Giants
The $502 Price Target Reflects Delivery Certainty and Customer Diversification
Morgan Stanley continues to rank Broadcom as its core AI compute exposure behind only Nvidia. The report maintains an Overweight rating and a $502 price target, based on 28 times calendar-year 2027 modeled EPS of $17.92. Relative to the pre-publication closing price of $384.05, the target implies approximately 31% upside.
The bull-, base-, and bear-case price targets are $637, $502, and $308, respectively. The bull case requires new ASIC customers, networking, and software synergies all to exceed expectations. The base case relies on custom silicon, networking, and software driving strong revenue growth in 2026–2027. The bear case assumes that new customers fail to reach volume production, revenue growth falls short of expectations, and valuation multiples contract.
Broadcom’s premium valuation is supported by more than a single TPU order. The company combines data-center switching chips, custom compute, and software cash flow, while new customers can further diversify its AI revenue. The market is willing to pay a premium for this portfolio, provided Broadcom continues to demonstrate supply-chain and mass-production execution. Stable market share is only the first level of validation; converting new programs into revenue on schedule is the second.
Institutional research should also be read in light of potential conflicts of interest. Morgan Stanley disclosed that it had provided advisory services to a Broadcom-related AI financing platform and maintains investment-banking relationships with several companies under coverage. The $502 price target can serve as a valuation scenario, but it cannot replace subsequent shipment, revenue, and supply-chain data.
Signals That Would Invalidate the 80% Share Assumption
Morgan Stanley identifies specific disconfirming evidence that investors can track individually. The first signal would be MediaTek’s chip performance materially exceeding current expectations and establishing a sustained advantage in power consumption, cost, or system efficiency. The second would be Google and MediaTek securing HBM at competitive prices and volumes, eliminating Broadcom’s advantage from contracted supply.
The third signal would be EMIB demonstrating that it can reliably support large-scale production of complex 2nm TPUs, with substrate supply no longer constrained. The fourth would be a transition from risk production to formal mass production that is materially faster than in previous AI accelerator programs. If all four conditions are met in succession, MediaTek would move from a backup supplier to a primary supplier capable of rapid expansion, requiring a reduction in Broadcom’s assumed approximately 80% revenue share.
Conversely, if MediaTek can secure only limited CoWoS capacity, lacks an HBM cost advantage, and faces continued EMIB delays, Google is more likely to maintain a dual-supplier structure while allocating most volume orders to Broadcom. In this scenario, MediaTek’s entry would strengthen Google’s negotiating position without undermining Broadcom’s overall ASIC growth.
Five Numbers to Watch
Broadcom’s TPU revenue share. Approximately 80% is the central assumption underpinning the report. It must be cross-checked against orders, wafer allocations, and revenue across Google’s TPU generations rather than inferred solely from unit shipments.
HBM supply and cost. Whether MediaTek can secure sufficient HBM at competitive prices will determine whether its lower-cost solution can be implemented in practice.
2nm packaging progress. CoWoS allocations, EMIB reliability, and substrate capacity will jointly determine the ceiling for volume production. Focusing only on tape-out timing would be premature.
10GW and $120 billion of AI revenue in FY2027. Deployment volume, value per gigawatt, and approximately $80 billion of TPU revenue must all materialize.
New-customer ramp in 2H27. Whether these programs can lift FY2028 deployments to at least 15GW and reduce TPU to approximately 60% of AI revenue will determine whether Broadcom can shed its single-customer valuation discount.
Broadcom’s ASIC moat has expanded from design capabilities to supply, packaging, and system delivery. MediaTek’s entry into Google’s TPU supply chain will result in real share losses and reduce Broadcom’s pricing power, but the available evidence remains insufficient to support a rapid decline to 50% share or complete replacement. Market expansion and new-customer production ramps can absorb limited share losses. The most important validation over the next two years will be whether Broadcom can continue converting mass-production certainty into revenue.
Source: Morgan Stanley’s July 14, 2026 research report on Broadcom. All market shares, revenue figures, deployment volumes, ratings, and price targets cited herein are institutional forecasts.