Broadcom (NASDAQ: AVGO) has reached an interesting point in the AI boom. Broadcom AI Chips are now at the center of that debate. Nvidia (NASDAQ: NVDA) just told Wall Street that demand is still accelerating. Nvidia expects revenue to grow about 70% in its next fiscal year, even while supply remains constrained. Now Broadcom has to show what that expanding AI market means for custom silicon.
Broadcom expects roughly $16 billion of AI semiconductor revenue in fiscal Q3, up more than 200% from a year earlier. That follows $10.8 billion in Q2, when AI semiconductor sales rose 143%.
Yet the bigger question is no longer whether demand exists. It is whether Broadcom can remain essential as its largest customers gain more control over their own chips.
Google is adding Marvell. Anthropic is exploring deeper chip development. OpenAI has designed its own processor with Broadcom. These developments could fragment Broadcom’s opportunity. They could also dramatically expand it.
Wednesday should give investors another clue about which direction this custom-silicon market is taking.
Nvidia Just Changed What Broadcom AI Chips Visibility Looks Like
Nvidia has raised the standard that Broadcom now faces. The company reported $96 billion of quarterly revenue and $89 billion from data centers. More importantly, management expects roughly 70% revenue growth in fiscal 2028. Demand could be even higher, but Nvidia says supply remains the limiting factor.
That is an unusually long window into AI demand.
Nvidia says the top five hyperscalers could spend around $1.3 trillion in 2027. Its revenue opportunity per gigawatt has also expanded sharply. Management estimates roughly $18 billion per gigawatt with Hopper, $25 billion with Blackwell, and $40 billion with Vera Rubin.
Broadcom now needs to demonstrate comparable durability inside custom AI silicon. Broadcom AI Chips will be judged not only on current growth, but also on how far that growth can extend.
Management already expects fiscal 2026 AI semiconductor revenue of roughly $56 billion. It has also reiterated that fiscal 2027 AI semiconductor revenue should exceed $100 billion. Broadcom had more than $30 billion of AI semiconductor bookings during Q2 alone. Management said visibility now stretches into 2028.
So $16 billion this quarter is only part of the story. The more important issue is whether those bookings, gigawatt commitments, and customer programs continue expanding far beyond 2026.
Google Diversifies & Broadcom Keeps Its Seat
Google’s expanding relationship with Marvell gets directly to the heart of Broadcom’s strategic problem.
Google recently signed a large custom-chip agreement with Marvell. The arrangement could eventually generate up to $120 billion for Marvell through fiscal 2033 if performance targets are reached. Google also received warrants that could give it a sizable equity position in the chipmaker.
That certainly introduces another supplier into Broadcom’s territory.
But Google adding Marvell does not mean Google is abandoning Broadcom.
In April, Broadcom disclosed a long-term agreement to develop and supply custom TPUs for future Google generations. Another agreement covers networking and other components for Google’s next-generation AI racks through as late as 2031.
Hock Tan has already acknowledged that Google will probably diversify suppliers. He also described Broadcom’s existing commitment as financially substantial.
Marvell management, meanwhile, expects major revenue from its Google agreement to emerge around fiscal 2029. Analysts cited by Reuters have generally viewed the arrangement as supplier expansion rather than immediate Broadcom displacement.
That distinction matters for Broadcom AI Chips.
If Google’s AI requirements grow faster than its supplier count, Broadcom can keep growing even while losing exclusivity. The risk emerges if diversification eventually becomes meaningful share loss.
OpenAI Could Validate Broadcom’s Custom Silicon Model
OpenAI may offer the clearest example of why proprietary chips do not automatically threaten Broadcom.
OpenAI is designing its own AI processors. But Broadcom is helping turn those designs into deployable silicon and infrastructure. The companies announced a 10-gigawatt custom accelerator partnership in 2025, with deployments expected through 2029.
That relationship has now moved beyond planning.
In June, OpenAI and Broadcom unveiled Jalapeño, OpenAI’s first custom inference processor. OpenAI designed the architecture around its own workloads. Broadcom contributed silicon implementation, networking, connectivity, and production expertise.
Then came an important update in August. OpenAI released early benchmark results showing higher throughput per unit of power and lower latency than the commercial systems included in its comparison. These are OpenAI’s own measurements, so they should not be treated as independent validation. Still, they show the chip has progressed into real testing.
Broadcom expects OpenAI production to begin in late 2026. Management has also disclosed a 1.3-gigawatt contractual commitment for 2027.
This gets to the key point for Broadcom AI Chips. OpenAI owning the architecture does not remove Broadcom from the equation. It changes what Broadcom gets paid to do.
Anthropic Shows Why Customer Independence Cuts Both Ways
Anthropic makes the custom-chip story even more complicated.
Broadcom, Google, and Anthropic expanded their relationship in April. Anthropic agreed to access multiple gigawatts of next-generation TPU capacity beginning in 2027. Broadcom’s filing specifically identified approximately 3.5 gigawatts tied to that expanded arrangement.
Broadcom is also helping finance the infrastructure behind this expansion.
Its AI XPV platform with Apollo and Blackstone is designed to enable more than 20 gigawatts of compute through 2028. An initial $35 billion transaction supports more than one gigawatt of Anthropic infrastructure.
Yet Anthropic is not standing still.
Reuters reported that Anthropic explored acquiring chip startup MatX for roughly $7 billion. Discussions later shifted toward a possible partnership. MatX was founded by former Google TPU engineers, and Anthropic has been expanding its own silicon expertise.
Anthropic is also continuing to buy outside compute. Reuters reported this week that it signed a $35 billion cloud agreement with Nvidia-backed Lambda.
This tells investors something important about Broadcom AI Chips. The largest AI labs may not choose one semiconductor ecosystem. They can use Nvidia GPUs, Google TPUs, Broadcom-enabled custom silicon, and their own designs simultaneously.
Broadcom therefore needs more than one chip contract. Its networking, packaging, connectivity, and implementation expertise must remain valuable as customer architectures evolve.
Final Thoughts
The question surrounding Broadcom is becoming clearer. It is not simply whether AI demand remains strong. Nvidia has already provided plenty of evidence on that front.
The question is how much economic value Broadcom can capture as AI customers increasingly design their own hardware and diversify suppliers.
There is evidence on both sides. Google is bringing Marvell deeper into its silicon strategy. Anthropic is building internal chip expertise. At the same time, Broadcom retains long-term Google commitments, has working silicon with OpenAI, and expects more than $100 billion of AI semiconductor revenue in fiscal 2027.
Valuation makes that debate more important.
As of August 31, Broadcom trades at roughly 23.95x LTM enterprise value-to-revenue, 42.94x LTM EV/EBITDA, and 61.63x LTM diluted earnings. Nvidia stands at about 17.52x, 26.37x, and 27.92x, respectively.
The comparison is not perfectly apples-to-apples. Broadcom has VMware, different accounting effects, and a different business mix. Still, Broadcom currently carries the richer trailing multiples on all three measures.
That valuation leaves investors paying for substantial future earnings growth from custom AI silicon. Wednesday can provide more evidence about whether the Google, OpenAI, Anthropic, networking, and broader XPU opportunities are developing fast enough to support those expectations.
Disclaimer: We do not hold any positions in the above stock(s). Read our full disclaimer here.
