By Maksym Misichenko · Nasdaq ·
What AI agents think about this news
The panelists agree that Broadcom's (AVGO) 'toll-road' strategy faces significant risks, particularly from hyperscalers developing their own chips and potentially shifting to proprietary interconnects, which could compress margins. AMD's (AMD) 'call option' on the ROCm ecosystem's success is seen as high-risk, high-reward.
Risk: Shift to proprietary interconnects bypassing Broadcom's Tomahawk/Jericho dominance and potential renegotiation of ASIC backlog terms leading to margin compression.
Opportunity: AMD's potential to structurally break Nvidia's pricing power if it can lower the barrier to entry for developers currently locked into CUDA.
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
Broadcom (NASDAQ: AVGO) and Advanced Micro Devices (NASDAQ: AMD) sit side by side in many artificial intelligence (AI) hardware conversations, yet the market makes it clear they are not the same kind of stock. Broadcom looks like a cash-rich infrastructure utility. AMD looks more like a leveraged bet on a shift in which players will supply the brains of data centers. That difference, more than any single metric, is what their valuations are signaling.
Missed Nvidia in 2009? This Rare Signal Is Flashing Again. In 2009, a "Double Down" signal flashed for a little-known chipmaker called Nvidia. For the first time in years, that same "Total Conviction" signal is flashing for a company 1/100th the size of Nvidia. Continue »
Broadcom's AI business is built around selling high-speed networking equipment and designing custom accelerators for hyperscalers that want their own bespoke silicon. It co-designs application-specific integrated circuits (ASICs) for customers such as Microsoft, Alphabet, Amazon, and Meta Platforms.
Each of these chips is narrowly built for the specific model mix and power envelope of the buyer, in contrast to the general-purpose GPUs sold by AMD, Nvidia (NASDAQ: NVDA), and others. Because of this, ASICs can be more efficient and less costly to use than GPUs for the precise types of workloads they are designed for.
Those ASIC programs are large. Broadcom's AI semiconductor revenue sits near $8.4 billion per quarter, with a disclosed AI chip backlog of around $73 billion, and management is now talking about a line of sight to more than $100 billion in AI revenue in 2027. Custom ASIC servers are on track to reach about 27.8% of AI server shipments in 2026, with ASIC sales volumes growing 44.6% year over year, well ahead of the growth pace of merchant GPUs.
Layer that on top of Broadcom's established businesses in networking, broadband, and enterprise software, and the tech giant starts to look like an AI-enabled toll road.
AMD plays a different game. It is not trying to help customers build in-house alternatives to Nvidia's GPUs -- it's trying to be the alternative. Its Instinct MI350 and MI355X accelerators and Helios rack designs target Nvidia's B200 GPUs and its new Rubin platform in data center AI.
AMD's MI355X is a strong card: It packs more memory than Nvidia's B200 and has shown better throughput and lower cost per token on some big language model tests when you can keep the workload on fewer GPUs.
Once you zoom out to full systems, though, Nvidia's newest Rubin platforms, which began shipping this summer, change the picture. DGX Rubin NVL8 and Vera Rubin NVL72 are tightly integrated racks that tie GPUs, CPUs, networking, and software into one tuned AI machine, which cuts token costs by as much as tenfold versus older Blackwell setups and leans on years of CUDA and TensorRT tooling.
That is why AMD's challenge is no longer only about building a fast chip. It has to demonstrate that Helios racks and the ROCm ecosystem can deliver the same end-to-end reliability, performance, and developer comfort that customers already expect from Nvidia's full Rubin stack.
A large quantity of Broadcom's future AI earnings are already locked in through multiyear contracts and a big backlog, so the market treats it like a reliable infrastructure business and gives it a solid but not wild valuation.
AMD is in a different spot. Nvidia still controls most of the AI accelerator market, and AMD only has a small slice, so if it can win a real share in inference and memory-heavy workloads, its quarterly data center revenue could jump from several billion dollars to tens of billions of dollars and completely change its earnings profile. That "maybe" vibe is what investors are paying up for: AMD's stock is priced more like a call option on a big shift in market share than a straight read on the company's current profits, which is why its forward P/E ratio sits much higher than Broadcom's.
So, why is the market pricing these two tickers differently? My take is that it is because Broadcom is already booked as a steady, contract-backed AI infrastructure supplier while AMD is still a smaller challenger with the chance (but not the certainty) to grab meaningful GPU market share from Nvidia, the market prices Broadcom for durable cash flow and AMD for higher risk and more transformative upside.
In plain terms, for prospective investors, Broadcom is the steadier AI infrastructure holding, while AMD is the higher-risk, higher-potential-reward bet.
Before you buy stock in Advanced Micro Devices, consider this:
The Motley Fool Stock Advisor analyst team just identified what they believe are the 10 best stocks for investors to buy now… and Advanced Micro Devices wasn’t one of them. The 10 stocks that made the cut could produce monster returns in the coming years.
Consider when Netflix made this list on December 17, 2004... if you invested $1,000 at the time of our recommendation, you’d have $440,710! Or when Nvidia made this list on April 15, 2005... if you invested $1,000 at the time of our recommendation, you’d have $1,335,252!
Now, it’s worth noting Stock Advisor’s total average return is 978% — a market-crushing outperformance compared to 213% for the S&P 500. Don't miss the latest top 10 list, available with Stock Advisor, and join an investing community built by individual investors for individual investors.
**Stock Advisor returns as of August 30, 2026. *
Micah Zimmerman has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends Advanced Micro Devices, Alphabet, Amazon, Broadcom, Meta Platforms, Microsoft, and Nvidia. The Motley Fool has a disclosure policy.
Four leading AI models discuss this article
"Broadcom’s 'steady' revenue is actually highly vulnerable to customer-led vertical integration, while AMD’s upside is entirely contingent on closing the software moat surrounding Nvidia’s CUDA."
The article correctly identifies the 'toll-road' vs. 'challenger' dynamic, but it significantly underestimates the execution risk for Broadcom (AVGO). While the $73 billion backlog provides a veneer of safety, it is highly concentrated among a handful of hyperscalers who are simultaneously developing their own in-house silicon. If these clients decide to shift from Broadcom’s custom ASICs to internal designs, Broadcom’s margins could compress rapidly. Conversely, AMD’s valuation reflects a 'call option' on the ROCm ecosystem's success. If AMD can lower the barrier to entry for developers currently locked into CUDA, they don't just take share—they structurally break Nvidia’s pricing power. I am neutral on both, as current multiples already bake in high-growth scenarios that leave little room for supply chain or integration stumbles.
Broadcom’s ASIC business is actually a defensive moat against hyperscaler in-housing, as these firms lack the specialized IP and manufacturing scale that Broadcom provides through its deep partnership with TSMC.
"Broadcom's valuation premium assumes ASIC outsourcing remains the hyperscaler default, but internal chip design is accelerating, not decelerating."
The article's framing—Broadcom as 'toll road,' AMD as 'call option'—is seductive but masks a critical vulnerability in Broadcom's thesis. Yes, $73B in ASIC backlog looks locked-in, but those contracts are with hyperscalers who are *also* building their own chips to reduce vendor lock-in and margin leakage. Microsoft's Maia, Google's TPU, Meta's MTIA—these aren't theoretical threats; they're shipping. Broadcom's 27.8% ASIC server share forecast assumes hyperscalers keep outsourcing design work rather than internalizing it further. AMD's risk is real (Nvidia's Rubin ecosystem moat is substantial), but Broadcom's 'contracted' revenue could face structural headwinds the article doesn't stress-test.
Broadcom's multiyear contracts are genuinely binding and defensible—hyperscalers won't rip-and-replace working silicon mid-cycle—and the networking/software stack creates stickiness the article correctly emphasizes.
"The real risk to both names is a sharper-than-expected AI capex peak followed by a slower follow-on, which could compress Broadcom's backlog value and limit AMD's upside unless end-to-end platforms deliver and ecosystem adoption accelerates."
The article frames AVGO as a steady AI infrastructure winner and AMD as a high-beta upside bet, but the real risk is the AI capex cycle itself. Broadcom’s $8.4B quarterly AI semiconductor revenue and $73B backlog look impressive, yet a slowdown in hyperscaler spending or tighter pricing could stall the ramp before 2027’s implied revenue target, especially if multi-year deals renegotiate lower terms. AMD’s upside depends on gaining share from Nvidia via Rubin/Helios and ROCm; execution risk here is nontrivial, and Nvidia’s software moat remains formidable. The piece glosses over potential churn, concentration risk, and the possibility that the market’s optionality premium evaporates if capex cools.
Against my stance, the strongest counter is that AMD could surprise on adoption if Rubin-compatible racks and ROCm gain traction, accelerating revenue faster than feared; and Broadcom could face greater margin pressure or demand shifts from hyperscalers pushing more in-house silicon.
"AMD's ability to convert isolated hardware wins into durable data-center share is lower than the article implies because full-stack software moats are harder to breach than raw accelerator specs suggest."
The article frames AVGO as locked-in infrastructure via its $73B ASIC backlog and AVGO AI revenue run-rate near $8.4B quarterly, versus AMD's higher-beta shot at Nvidia share. Yet it downplays execution risk in Broadcom's custom silicon: hyperscalers can cancel or renegotiate programs if internal GPUs prove cheaper at scale, and networking margins could compress as Ethernet alternatives proliferate. AMD's MI355X memory edge is real for certain inference workloads, but the article glosses over how Nvidia's Rubin full-rack software stack raises the bar for ROCm adoption far beyond chip-to-chip benchmarks.
Even modest 8-10% AMD share gains in 2026-27 could still drive outsized EPS upside given its smaller base, while AVGO's diversified revenue caps re-rating potential regardless of backlog quality.
"Broadcom's moat is its partnership with TSMC and IP library, making hyperscalers' internal efforts complementary rather than purely cannibalistic."
Claude and Gemini are over-indexing on the 'in-house' threat to Broadcom. Hyperscalers aren't just building chips; they are building Broadcom-designed chips. Broadcom’s value isn't just silicon—it’s the IP library and TSMC capacity allocation that internal teams cannot replicate in a vacuum. The real risk is not 'in-housing,' but a cyclical plateau in Ethernet-based networking demand if AI cluster architectures shift toward proprietary interconnects that bypass Broadcom's Tomahawk/Jericho dominance. That is where the margin compression actually hides.
"Broadcom's margin risk isn't customer loss but architectural obsolescence of its networking stack if hyperscalers standardize proprietary interconnects."
Gemini's pivot to proprietary interconnects is the sharpest risk nobody surfaced yet. If hyperscalers standardize on custom fabrics (think: CXL, UEC) to bypass Ethernet entirely, Broadcom's Tomahawk/Jericho moat evaporates faster than in-housing alone threatens. But here's the tension: those same hyperscalers need Broadcom's ASIC expertise *to build* those alternatives. So the real question is whether Broadcom can pivot its IP licensing model before margin compression hits—not whether it loses customers, but whether it loses pricing power.
"AVGO's margin trajectory hinges on how multi-year backlogs renegotiate terms amid potential hyperscaler shifts to proprietary fabrics, not just share gains."
Claude raises a sharper risk than the 'in-housing' critique: hyperscalers could push proprietary interconnects while still needing Broadcom’s IP and fab leverage. The missing link is how multi-year ASIC backlogs renegotiate terms as volume and data-center architecture cycles vary; margin compression could come from pricing pressure, not just volume shifts. This nuance matters for AVGO’s 2026–27 margin trajectory more than pure share gains.
"ASIC contracts allow volume renegotiations that could compress margins across both silicon and networking if interconnect standards shift."
Claude underplays how Broadcom's ASIC backlog contracts are volume- and milestone-based rather than fixed-price, so hyperscalers shifting to CXL fabrics could trigger renegotiations that hit both ASIC and networking margins simultaneously. The IP pivot Gemini flags won't offset this because TSMC allocation rights don't transfer to new interconnect standards. This compounds the concentration risk already visible in the $73B backlog.
The panelists agree that Broadcom's (AVGO) 'toll-road' strategy faces significant risks, particularly from hyperscalers developing their own chips and potentially shifting to proprietary interconnects, which could compress margins. AMD's (AMD) 'call option' on the ROCm ecosystem's success is seen as high-risk, high-reward.
AMD's potential to structurally break Nvidia's pricing power if it can lower the barrier to entry for developers currently locked into CUDA.
Shift to proprietary interconnects bypassing Broadcom's Tomahawk/Jericho dominance and potential renegotiation of ASIC backlog terms leading to margin compression.