Broadcom paints huge AI future

[Marees]

Broadcom predicts that three of its clients, among hyperscalers, will deploy AI supercomputers with one million XPUs in fiscal 2027.

said Hock Tan, President and CEO of Broadcom, at the company's Q4 2024 earnings call.


Broadcom develops chips for AI, general-purpose data processing, or custom data center hardware — for multiple big-name companies, including Google and Meta. Broadcom and its customers identify the workload demands, such as AI training, inference, or data processing. Then, the company and its partners define the specifications of their chips and develop key aspects of their main differentiators, such as the architecture of processing units, leveraging Broadcom's expertise in silicon design. Broadcom then implements this architecture in silicon and equips it with platform-specific IP, caches, inter-chip interconnects, and interfaces. Broadcom-designed high-performance XPUs are then manufactured by TSMC.

Broadcom may sell its XPUs or custom ASICs directly to customers with long-term supply agreements depending on the contract with a particular customer. In addition, Broadcom may assist in developing products, charging for collaborative engineering and/or research and development efforts.



Broadcom's XPU business is a key ingredient of its strategy to capitalize on the growing demand for AI and cloud infrastructure, making it a critical player in the AI hardware ecosystem. The company believes that in 2027, the serviceable addressable market (SAM) for AI XPU and networking will be between $60 and $90 billion, and the firm is positioned to command a leading share of this market. It is unclear how the company counts SAM, though, as this year, Nvidia alone will earn about $100 billion selling its GPUs, DPUs, and networking hardware to the AI market.

https://www.tomshardware.com/tech-i...says-three-ai-supercomputers-are-in-the-works

 

[Marees]

Compare & contrast Broadcom's position with that of Intel

TSMC founder says Intel has neither a strategy nor a CEO​

News
By Anton Shilov
published December 11, 2024
Intel could develop more money on proper AI chips than on foundry services

Morris Chang, the founder of TSMC, claims that Intel should have focused on the development of AI processors rather than on making chips using leading-edge nodes.


"I do not know why Pat resigned," said Chang at a press conference dedicated to the publication of his biography, Reuters reports. "I do not know if his strategy was bad or if he did not execute it well." [...] "Compared with AI, he seemed to focus more on becoming a foundry. Of course, now it seems that (Gelsinger) should have focused on AI."

Chang's criticism is understandable. Nvidia earns tens of billions of dollars per year selling AI processors, considerably more than TSMC earns on fabbing those chips. If Intel had competitive AI processors, it could make more money than it would on making chips for others, so it's easy to argue that an AI strategy should have been prioritized years ago.


https://www.tomshardware.com/tech-i...focus-on-ai-not-advanced-process-technologies

 

[philb2]

Compare & contrast Broadcom's position with that of Intel

TSMC founder says Intel has neither a strategy nor a CEO​

News
By Anton Shilov
published December 11, 2024
Intel could develop more money on proper AI chips than on foundry services

Morris Chang, the founder of TSMC, claims that Intel should have focused on the development of AI processors rather than on making chips using leading-edge nodes.


"I do not know why Pat resigned," said Chang at a press conference dedicated to the publication of his biography, Reuters reports. "I do not know if his strategy was bad or if he did not execute it well." [...] "Compared with AI, he seemed to focus more on becoming a foundry. Of course, now it seems that (Gelsinger) should have focused on AI."

Chang's criticism is understandable. Nvidia earns tens of billions of dollars per year selling AI processors, considerably more than TSMC earns on fabbing those chips. If Intel had competitive AI processors, it could make more money than it would on making chips for others, so it's easy to argue that an AI strategy should have been prioritized years ago.


https://www.tomshardware.com/tech-i...focus-on-ai-not-advanced-process-technologies

Yeah, but corporate ego. If Intel was a commercial foundry it would be a member of a very select club. Otherwise, it's just Joe Average company

 

[Marees]

OpenAI to launch its first AI chip in 2026 with Broadcom, FT reports​

By Reuters

OpenAI is set to produce its first artificial intelligence chip next year in partnership with U.S. semiconductor giant Broadcom (AVGO.O), opens new tab, the Financial Times reported on Thursday, citing people familiar with the matter.
OpenAI plans to put the chip to use internally rather than make it available to external customers, the FT report said, citing one person close to the project.

Tan earlier this year had hinted at four new potential customers who were "deeply engaged" with the company to create their own custom chips, in addition to its three existing large clients.

https://www.reuters.com/business/op...hip-2026-with-broadcom-ft-reports-2025-09-05/

 

[Marees]

OpenAI Designs its First AI Chip in Collaboration with Broadcom and TSMC​

by AleksandarK Oct 31st, 2024 12:38

OpenAI has assembled a dedicated chip team of approximately 20 people, led by former Google TPU engineers Thomas Norrie and Richard Ho. The company has secured manufacturing capacity with TSMC, targeting a 2026 timeline for its first custom-designed chip. While Broadcom's involvement leverages its expertise in helping companies optimize chip designs for manufacturing and manage data movement between chips—crucial for AI systems running thousands of processors in parallel—OpenAI is simultaneously diversifying its compute strategy. This includes adding AMD's Instinct MI300X chips to its infrastructure alongside its existing NVIDIA deployments.

 

[Marees]

• In-House Chip Development with Broadcom: OpenAI has chosen Broadcom as its primary design partner for its AI chip, scheduled for completion in 2026. Broadcom’s expertise in chip design is expected to be pivotal in developing a solution optimized for inference tasks, crucial as OpenAI expands AI deployment beyond initial training phases.

• Manufacturing Partnership with TSMC: To produce its custom chip, OpenAI has secured a manufacturing agreement with TSMC. As the world’s leading chip manufacturer, TSMC’s capabilities ensure OpenAI’s chip development will proceed within the projected timeline, although adjustments may arise based on production or design challenges.

• Diversifying with AMD’s MI300X Chips: OpenAI has started incorporating AMD chips via Microsoft’s Azure, complementing its current reliance on Nvidia GPUs. AMD’s MI300X chips, set to launch late this year, aim to provide a competitive alternative to Nvidia’s GPUs, which currently dominate the AI market.

https://nationalcioreview.com/artic...first-ai-chip-venture-with-broadcom-and-tsmc/

 

[littlegreenmenfromhome]

Broadcom trying as many others to gain a foot hold in the A.I. market due to Nvidia having +90% A.I. hardware marketsshare.
Their problem is that they are Broadcom and custormers are weary due to their history.

 

[LukeTbk]

Broadcom trying as many others to gain a foot hold in the A.I. market due to Nvidia having +90% A.I. hardware marketsshare.
Their problem is that they are Broadcom and custormers are weary due to their history.

They are already very much in that scene, they are one of the largest player in that supply chain.

Google TPU, Meta custom chips, Microsoft custom chips, the high speed networking...
Within the chips business, our model shows AI chips revenues – which includes switch ASICs, XPU ASIC design, packaging, and shepherding for hyperscalers and cloud builders, as well as other chippery – at Broadcom rose by 46.7 percent to $4.42 billion, while other chip sales slipped by 4.8 percent to $3.99 billion. Broadcom has three custom compute engine customers – Google was the first, but now Meta Platforms is working on chips with the company and so is OpenAI – with four other prospects – which we think includes Apple, ByteDance, and two others – who are looking to use Broadcom as XPU shepherds.

They have been the quietest trillion dollar company (for a while) in that space: https://www.reuters.com/technology/broadcom-rallies-forecast-booming-ai-chip-demand-2024-12-13/, maybe because they kept their old name.

That 90% AI hardware marketshare number is on the open market, the majority of Nvidia competition are internal (google TPU, AWS, openAI/meta/microsoft/tesla custom solution, everything in China) and do not show up in it, Nvidia actual marketshare is probably more 60% than 90%.

 

[littlegreenmenfromhome]

They are already very much in that scene, they are one of the largest player in that supply chain.

Google TPU, Meta custom chips, Microsoft custom chips, the high speed networking...
Within the chips business, our model shows AI chips revenues – which includes switch ASICs, XPU ASIC design, packaging, and shepherding for hyperscalers and cloud builders, as well as other chippery – at Broadcom rose by 46.7 percent to $4.42 billion, while other chip sales slipped by 4.8 percent to $3.99 billion. Broadcom has three custom compute engine customers – Google was the first, but now Meta Platforms is working on chips with the company and so is OpenAI – with four other prospects – which we think includes Apple, ByteDance, and two others – who are looking to use Broadcom as XPU shepherds.

View attachment 752137

They have been the quietest trillion dollar company (for a while) in that space: https://www.reuters.com/technology/broadcom-rallies-forecast-booming-ai-chip-demand-2024-12-13/, maybe because they kept their old name.

That 90% AI hardware marketshare number is on the open market, the majority of Nvidia competition are internal (google TPU, AWS, openAI/meta/microsoft/tesla custom solution, everything in China) and do not show up in it, Nvidia actual marketshare is probably more 60% than 90%.

Microsoft has a way broader goal, they are going for a "triangle" and are 5-10 years ahead of the competion as A.I is just one cog in the wheel.
Google and AWS are 5-10 years behind Microsoft there.
Tesla, OpenAI, Meta and Broadcom have not even entered that race.

IBM is the closest thing to a competitor Microsoft has in the final cog, but they lack the other parts to complete the "triangle".

The "long game" is what people need to pay attention to.
10 years ago we had 50-60 qubits.
Now we are ~1200 qubits.
~4000 qubits and SHA-2048 is broken.
If people think A.I is the major revolution of the decade/century they are in for some surprises because plans are already being made and bugdets are approved at lof of places for PQC due to "harvest now, decrypt later".
And Broadcom is not one of the players so they can wave their hands all they want, to me they are becoming irrelevant.

 

[LukeTbk]

The idea that anyone is ahead of Google on any part AI related seem a dangerous bet to make, specially by many years ? They are probably number 1 at least in demonstrated capacity quantumn computing wise

https://www.nytimes.com/2024/12/09/technology/google-quantum-computing.html
https://blog.google/technology/research/google-willow-quantum-chip/

Google has arguably the best hardware, the best data, the best models and the best portal/access to customer for it, all at the same time, from the start with deepmind AlphaGO, always ahead of the pack and could stay ahead.

 

[littlegreenmenfromhome]

The idea that anyone is ahead of Google on any part AI related seem a dangerous bet to make, specially by many years ? They are probably number 1 at least in demonstrated capacity quantumn computing wise

https://www.nytimes.com/2024/12/09/technology/google-quantum-computing.html
https://blog.google/technology/research/google-willow-quantum-chip/

Google has arguably the best hardware, the best data, the best models and the best portal/access to customer for it, all at the same time, from the start with deepmind AlphaGO, always ahead of the pack and could stay ahead.

Google is at 105 qubits.
Try looking for news from 2025 Google is not in the front or having the best hardware IBM broke 1.000 qubits back in 2023.
The pace is blazing fast right now.

 

[NamelessPFG]

They have been the quietest trillion dollar company (for a while) in that space: https://www.reuters.com/technology/broadcom-rallies-forecast-booming-ai-chip-demand-2024-12-13/, maybe because they kept their old name.

Kept their old name? Isn't present-day, enshittification-speedrunning Broadcom really Avago wearing Broadcom's name as a hat after an acquisition? Note the AVGO stock ticker.

We've seen what they've done with PLX Technology (PCIe switch chips), LSI (SAS controller chips), and now VMware. I'm betting that last one is what they're hinging on to make profit off of all these AI hyperscalers through providing the ESXi hypervisor for all these new datacenters, never mind that businesses are moving off of VMware ESXi left and right due to the extortionate price increases.

I also know they sell Ethernet NICs, but all the Broadcom chipset stuff I've encountered is mere 1Gb Ethernet, not the 100Gb Ethernet and such that the AI datacenters seek - and at those speeds, I've only seen Mellanox/NVIDIA hardware being used.

 

[LukeTbk]

Kept their old name? Isn't present-day, enshittification-speedrunning Broadcom really Avago wearing Broadcom's name as a hat after an acquisition? Note the AVGO stock ticker.

Yes using that old Broadcom name was part of the acquisition deal, apparently (at least that what one of the All-in podcast guy heard from the previous Broadcom CEO).

If it was a new startup name doing what they are doing, Hock Tan would be more of an household name, they are killing-it. Or maybe if you do not have any regular consumer facing product you will always stay a bit unknown... 5G infracstructure and datacenter networking is not super sexy for the news.

I also know they sell Ethernet NICs, but all the Broadcom chipset stuff I've encountered is mere 1Gb Ethernet, not the 100Gb Ethernet and such that the AI datacenters seek - and at those speeds, I've only seen Mellanox/NVIDIA hardware being used.

https://www.broadcom.com/products/ethernet-connectivity/switching/strataxgs/bcm56980-series

• Up to 32x400GbE, 64x200GbE or 128x100GbE ports

they are the most under the radar tech ticker that reached 1 trillion...

 

[littlegreenmenfromhome]

Yes using that old Broadcom name was part of the acquisition deal, apparently (at least that what one of the All-in podcast guy heard from the previous Broadcom founder).

If it was a new startup name doing what they are doing, Hock Tan would be more of an household name, they are killing-it. Or maybe if you do not have any regular consumer facing product you will always stay a bit unknown... 5G infracstructure and datacenter networking is not super sexy for the news.


https://www.broadcom.com/products/ethernet-connectivity/switching/strataxgs/bcm56980-series

• Up to 32x400GbE, 64x200GbE or 128x100GbE ports

they are the most under the radar tech ticker that reached 1 trillion...

12.8 TB/s aggregate is not market leading.
It is less than 1/4 of what is common commerical available.

 

[LukeTbk]

12.8 TB/s aggregate is not market leading.
It is less than 1/4 of what is common commerical available.

you can double it with Tomahawk4 if you want... Broadcom made revenues a 64 billions a year last quarter, giant amount of their growth is from AI networking product, I am not sure what is being suggested here ?

 

[littlegreenmenfromhome]

you can double it with Tomahawk4 if you want... Broadcom made revenues a 64 billions a year last quarter, giant amount of their growth is from AI networking product, I am not sure what is being suggested here ?

A.I. is not networking you cannot blanket drop a number without seperating it into divisions.

 

[LukeTbk]

they are the most under the radar tech ticker that reached 1 trillion...

Not anymore:
https://openai.com/index/openai-and-broadcom-announce-strategic-collaboration/

OpenAI and Broadcom announce strategic collaboration to deploy 10 gigawatts of OpenAI-designed AI accelerators​

View: https://www.youtube.com/watch?v=h0Q0Ez7T4zM

https://www.cnbc.com/2025/10/13/ope...com-custom-ai-chips-alongside-nvidia-amd.html

• OpenAI and Broadcom have been collaborating for 18 months on a new line of co-designed chips optimized for inference and networked through Broadcom’s Ethernet stack.

While the companies have been working together for 18 months, they’re now going public with plans to develop and deploy racks of OpenAI-designed chips starting late next year.
Broadcom has been one of the biggest beneficiaries of the generative AI boom, as hyperscalers have been snapping up its custom AI chips, which the company calls XPUs. Broadcom doesn’t name its large web-scale customers, but analysts have said dating back to last year that its first three clients were Google, Meta and TikTok parent ByteDance.

That for an OpenAI custom chips.

 

[OFaceSIG]

For the rest of my life, professional, or personal, I will purposefully try to avoid doing any monetary business with Broadcom. There are souless monsters like Jensen in this world, but at least his company actually develops and engineers things. Hock Tan has publicly stated he buys companies and extracts value. He does nothing but digest companies to make money. He is the Unicron of the business world. He can eat the thickest of wieners.

 

[philb2]

For the rest of my life, professional, or personal, I will purposefully try to avoid doing any monetary business with Broadcom. There are souless monsters like Jensen in this world, but at least his company actually develops and engineers things. Hock Tan has publicly stated he buys companies and extracts value. He does nothing but digest companies to make money. He is the Unicron of the business world. He can eat the thickest of wieners.

Remember Computer Associates? Where good products "go to die."