8. Broadcom Inc. (NASDAQ:AVGO)
Billionaire Ken Fisher’s Stake: $4.79 billion
Broadcom Inc. (NASDAQ:AVGO) ranks 8th in our list of the best AI stocks to buy according to billionaire Ken Fisher.
Companies are racing to sign multi-year AI chip contracts with Broadcom Inc. (NASDAQ:AVGO). Recent deals include hyperscaler commitments from Google, Meta, and TikTok/ByteDance. What’s special about Broadcom Inc. (NASDAQ:AVGO) is its ability to deliver custom, workload-specific silicon that general-purpose GPUs cannot match.
Broadcom Inc’s (NASDAQ:AVGO) moat is strong because there are only a few companies capable of delivering high‑performance custom chips tailored for large‑scale AI workloads, and that specialization delivers real economic advantages. While Nvidia’s GPUs dominate general‑purpose AI compute, Broadcom designs application‑specific ASICs and XPUs that hyperscalers prefer for massive inference tasks because they can reduce total cost of ownership by roughly 40–60% compared with GPU clusters, especially in steady, predictable production environments.
According to a Deloitte report, the custom AI chip market is expected to exceed $50 billion in 2026, driven by a shift in workloads from training to inference, which will account for roughly two-thirds of all compute that year. While lighter inference tasks can be handled on PCs and smartphones with onboard accelerators, enterprise and hyperscale workloads still require high-performance, custom-designed chips to efficiently manage post-training and test-time scaling.
Analysts and industry reports also suggest that deploying Broadcom Inc’s (NASDAQ:AVGO) custom silicon at scale can save companies billions in upfront capital expenditures and cut electricity costs by about half, due to the chips’ much lower power draw and higher performance per watt.
Emerald Wealth Partners Focused Equity Strategy stated the following regarding Broadcom Inc. (NASDAQ:AVGO) in its fourth quarter 2025 investor letter:
“We took advantage of a consolidation in Broadcom Inc.’s (NASDAQ:AVGO) stock price to initiate a position. Its semiconductor design unit has a dominant position in custom chips (ASICs). Broadcom caters for cloud hyperscalers and companies building LLMs that seek to design their own chips. The idea is to create chips designed to efficiently process the specific workload they need and cut their reliance on general purpose units (GPUs) that are very flexible but expensive.
GPUs remain indispensable in training new AI models. But as AI use expands, the scale of the infrastructure required to process user prompts will grow exponentially. Running these models using GPU chips that can cost US$60,000 per unit is not economically sustainable. Broadcom has demonstrated its capabilities in the field as it designed Google’s Tensor Processing Units (TPUs). Alphabet has used TPUs to run search queries since 2015 and now to run inference of the Gemini LLM. Given the cost advantage it gives Alphabet, every large GPU customer is incentivized to design its own ASICs. Down the line it may well be a matter of survival. Since such endeavors are a long journey and require considerable resources, it seems likely most will favor the semiconductor design partner with the most proven experience in the field (i.e. Broadcom) to ensure faster results and minimize the risk of failure.
Broadcom’s other businesses are developing well too. Demand for its networking equipment is strong thanks to the dynamic data center build-up. Fragmentation of the IT hardware environment provides a good fundamental backdrop for VMWare, the virtualization software business Broadcom acquired in 2024.”
