When discussing the AI Monetization Supercycle, I would be remiss not to highlight Broadcom. The AI accelerator market will inevitably widen beyond Nvidia’s GPUs – the keyword is widen. More players will sell more AI systems as the market expands, and that growth supports both the clear leader (Nvidia) and those already in pole position, such as Broadcom.
Last week, amidst a flurry of noise in the AI market, my firm wrote an article on the AI Monetization Supercycle that is not being priced in. The analysis suggested the predominant risk is not an AI dot-com bubble or various headlines weighing on sentiment, but rather the risk investors face is missing out on what may be one of the strongest investing opportunities of our lifetime: what I’ve dubbed the AI Monetization Supercycle catalyzed by the inference phase.
While many refer to this as the “AI Supercycle,” I believe Monetization is a critical word missing from that description. The hallmark of the next phase will not be the architectural leap toward AI superintelligence (although important) – but rather, it will be defined by the ability to monetize this very expensive technology. As an investor, I am obligated to care more about the latter.
Which brings us back to Broadcom—a stock my firm highlighted in our free stock newsletter last June in an article entitled “This Stock is Set to Surge from Inference Demand.”
At the time, I wrote:
“Broadcom has already benefited from both increasing compute and networking needs – but the surge in inference demand will disproportionately (and positively) flow to Broadcom’s top line and bottom line. This is because custom silicon’s cost advantages and ability to drive lower inference serving costs at scale creates a strong value proposition for Big Tech. As more and larger clusters are deployed to serve exploding inference demand, there will be additional long-term tailwinds for the Ethernet networking giant.”
The inference phase – what I’m calling the Monetization Supercycle – is squarely in front of us. While many will understandably point toward companies like OpenAI as the biggest beneficiaries, it is one of the market’s greatest misconceptions that platform owners always outperform suppliers (hardware stocks). During the mobile era, Broadcom’s stock outperformed Apple precisely because it supplied RF and connectivity components to the iPhone giant.
Below, we look more closely to see if the “silent winner” Broadcom stock can repeat that outperformance again.
Source YCharts
Google TPU Ironwood v7: The Custom AI Chip Built for Inference
Last April, Google announced that its upcoming seventh-gen TPU Ironwood is its “most performant and scalable custom AI accelerator to date, and the first designed specifically for inference.” Individual Ironwood TPUs are interconnected into larger units called pods, coming in two sizes, a 256-chip pod and a 9,216-chip Superpod, with the larger size offering up to 42.5 exaflops of performance. Notably, the Superpod would deliver 24x the compute of El Capitan, the largest supercomputer in the world. The rack-scale architecture offers 64 TPUs compared to Nvidia’s racks with 72 GPUs, with a small cluster being four pods connected through an optical circuit switch network. While TPUs may excel at driving down costs on certain workloads, Nvidia’s GPUs still lead when it comes to processing performance.
Google adds that Ironwood offers 2x the performance per watt as last-year’s generation Trillium, with 6x more HBM and 4.5x the HBM bandwidth; versus TPU v5p, released in 2023, Ironwood brings a more than 10x improvement in peak performance per chip and per pod. The substantial increases in memory and bandwidth are critical for maintaining high performance when processing larger data sets while the improvements in power efficiency allows inference workloads to be run in a cost-effective manner.
It’s widely understood that Broadcom supplies Google with its custom TPUs. The incoming inference growth curve, that the I/O Fund detailed here, has led CEO Hock Tan to state Broadcom may witness an acceleration of XPU demand into the back half of 2026. He said, “In fact, what we’ve seen recently is that they are doubling down on inference in order to monetize their platforms. And reflecting this, we may actually see an acceleration of XPU demand into the back half of 2026 to meet urgent demand for inference on top of the demand we have indicated from training.”
Something similar was echoed in the FQ3 call, with Tan stating: “But also as for these guys, they got to be accountable to being able to create cash flows that can sustain their path. They [are] starting to also invest in inference in a massive way to monetize their models.” On that note, Google’s TPU business received a significant vote of confidence recently with Anthropic signing a deal for up to one million TPUs, including Ironwood, coming online in 2026. The deal is said to be worth tens of billions.
For Broadcom, the TPUs are expected to be the primary driver of AI revenue growth in fiscal 2026 – estimates from HSBC earlier this summer projected Google’s TPUs to represent ~58% of Broadcom’s ASICs shipments at 1.79 million, but account for ~78% of ASICs revenue at $22.1 billion. This is because Google’s TPUs were estimated to carry a significant price premium at $13,000 per chip versus Broadcom’s other projects at $5,000 per chip. However, this is still less than half the cost of Nvidia’s chips at $30,000 to $40,000 for a solo B200 ($60,000 to $70,000 for a GB200).
Looking beyond fiscal 2026, projections for TPU shipments are surging. Morgan Stanley now expects 5 million TPUs to be shipped in 2027, a 67% rise from its prior estimate for 3 million; for 2028, the firm estimates shipments as high as 7 million, a 120% increase from its prior estimate. This would project YoY growth of 40% from 2027 to 2028, a substantial increase from 6% previously, and will represent more than 2X growth in two years.
The I/O Fund first covered TPUs versus GPUs back in 2019 and revisited the topic in February 2024 in our analysis, Broadcom: Networking/ASICs Giant and the Second Largest by AI Revenue. Since then, we’ve provided quarterly coverage for two years.
If you want cutting-edge insights on AI stocks early in the cycle — including our take on Broadcom’s earnings this evening — sign up now.
Broadcom Stock’s AI Edge: Custom Silicon & Massive Hyperscaler Deals
Broadcom’s stock has been strong this year, outperforming the Nasdaq by nearly 50-points and SMH by 20-points. This strong performance is partly due to custom accelerators that are often multiples cheaper than Nvidia’s GPUs for inference tasks and also due to custom silicon becoming increasingly performant with each generation. By optimizing algorithms (software), Big Tech can drive higher performance from large language models — which helps to drive down costs while also increasing output for specific workloads.
For example, a rough idea as to how much it costs Nvidia to make merchant GPUs is estimated around $3,000 to $6,000 whereas the company charges $30,000 to $40,000 – hence the AI leader’s excellent margins. Reducing Nvidia’s high pricing power is what Big Tech is after and this can be accomplished both in the hardware costs but also through optimizing the workloads for specific use cases – for comparison, Ironwood is expected to cost around $13,000 per chip.
Big Tech is prominent in Broadcom’s custom silicon customer list, which includes Google and Meta. ByteDance reportedly emerged as the third customer last summer. The company announced its fourth customer in FQ3 with a $10 billion XPU order. Hock Tan said in the FQ3 earnings call, “Last quarter, one of these prospects released production orders to Broadcom, and we have accordingly characterized them as a qualified customer for XPUs and, in fact, have secured over $10 billion of orders of AI racks based on our XPUs.”
In late October, Anthropic signed a deal with Google worth tens of billions to access up to 1 million TPUs to bring online more than 1GW of capacity in 2026, although it has not explicitly confirmed if Anthropic is the mystery fourth customer.
Furthermore, OpenAI and Broadcom announced in October a strategic collaboration to deploy 10 gigawatts of OpenAI-designed AI accelerators. OpenAI and Broadcom will co-develop systems that include accelerators and Ethernet solutions from Broadcom for scale-up and scale-out. Broadcom plans to deploy racks of AI accelerators and network systems starting in the second half of 2026 and completed by the end of 2029.
The OpenAI deal represents a substantial three-year revenue ramp for Broadcom stock and further solidifies its position in the AI silicon market. Citi estimates the deal with OpenAI could bring in $100 billion in sales and $8.00 in earnings per share over the next few years; however, Mizuho highlighted that the deal to deploy 10GW of OpenAI’s custom ASIC, code named Titan, could be even larger at an estimated $150 billion to $200 billion deal over multiple years.
The enviable customer list is showing up in Broadcom’s results. This quarter, management guided Q4 AI revenue to $6.2 billion, which would represent ~19% sequential growth and eleven consecutive quarters of YoY growth.
Broadcom did not lay out a FY25 AI revenue target, yet FQ4 ending in October 2025 implies Broadcom is guiding for $19.9 billion in AI revenue for the year, up 63% YoY from $12.2 billion in FY24. Mizuho estimates that AI revenue will grow 103% YoY to $40.4 billion for the FY2026 and nearly double to $78 billion in FY2028. However, given the growing customer list, these estimates could prove to be too low.
Additionally, Hock Tan will be duly rewarded should AI revenue targets exceed current expectations. In September, Tan received a performance award of 610,251 shares of common stock as part of a recent contract extension. The award will fully vest if Broadcom reaches $90 billion in revenue from its AI products over any consecutive four-quarter period from FY2028 through FY2030. That award will double if Broadcom earns $105 billion in AI revenue and triple if revenue totals more than $120 billion. If Broadcom fails to hit $60 billion in AI revenue during the period, Tan will forfeit the entire award. This provides investors with a framework for upper targets for the bull case.
