The AI chip market is undergoing earthquake-changing. For years, Nvidia (NVDA) ruled the best, and its GPUs have driven everything from generating AI to high-performance computing. However, in 2025, top funds have reassigned capital to Broadcom (AVGO), Google (GOOGL) and Palantir (PLTR) and bet on a more fragmented, specialized AI ecosystem. Why and what does that mean for investors?
Broadcom: Custom Chip Challenger
Broadcom's AI revenues increased 63% year-on-year to $5.2 billion in Q3 2025, driven by custom silicon tailored to hyperscalers like Google and Meta. [1]. The company's XPU (Application-Specific Integrated Circuit (ASIC)) is designed for specific workloads and offers greater efficiency and cost-effectiveness compared to Nvidia's general-purpose GPUs. [2]. For example, Broadcom's Tomahahawk Ultra Ethernet Switch with 102.4 TBPS capacity enables low-latency connectivity for AI clusters and addresses critical bottlenecks in large-scale deployments [3].
Financially, Broadcom's AI business is on track to generate $20 billion a year, positioning it as the second-largest player after Nvidia's $6 billion run rate. [4]. The $10 billion custom chip deal with new cloud clients and a partnership with four Hyperschools highlights the momentum [1]. Meanwhile, Nvidia's advantage is facing headwinds. Its Blackwell GPU, albeit powerful, encountered thermal challenges in high density data centers, slowed deployment and raised questions about long-term stability [5].
Google: Efficiency for raw power
Google's tensor processing units (TPUs) continue to outperform with power and cost-effectiveness. The latest TPU V5P offers up to 460 petaflops per POD and provides nearly linear scaling on models like the GPT-3 175b [6]. This makes the TPU ideal for Google's cloud-based AI workloads. Energy consumption and total cost of ownership (TCO) are important for this workload. In contrast, NVIDIA's H100/H200 GPUs dominate with raw flops (141 Teraflops FP8 for the H200), but face scrutiny on power efficiency, especially in blackwell architectures. [7].
Google's strategic focus on custom silicon is paying off. By integrating TPUs into AI infrastructures and generative models like Gemini, the company is locking in demand for cloud services. For investors, this represents dual play: a synergistic effect of hardware and software that rivals Nvidia's standalone GPU models.
Palantir: A great country in AI software
Palantir is not a chip business, but the AI-driven analytics platform is restructuring fund allocations. The company's $795 million DOD contract extension and a potential $10 billion Army contract highlight its advantage in defense and government AI [8]. The Gotham and Foundry platforms are essential for institutions that integrate generated AI to automate decision-making and process huge data sets.
Palantir does not have a hardware benchmark, but the 40-score rule (83% adjusted operating margin + growth) outweighs many peers [9]. Top funds are betting on their ability to monetize AI in the niche sector, especially as governments prioritize safe ontology-driven solutions. This contrasts with Nvidia's broad but increasingly crowded market.
Nvidia's Challenge: Can it maintain the crown?
Nvidia's fourth quarter earnings reached $35.1 billion, but its shares have recently fallen below the 50-day moving average, signaling short-term anxiety [10]. Competitors like AMD and Broadcom are filling the gap. For example, AMD's MI300X matches Nvidia's H200 with some inference benchmarks [11]. Regulation hurdles in particular in China further complicate Nvidia's growth trajectory.
But maintains MLPERF training 4.0 with 2.2x performance leap from Nvidia's Cuda Ecosystem and Blackwell [12]. The question is whether that heat and whether the rise in competition will erode margins over time.
What is the investor's takeaway?
AI chip landscapes are becoming more diverse. Broadcom's custom ASICS, Google's efficient TPU, and Palantir's software-first approach are challenging Nvidia's hegemony. In the case of funds, this means spreading the bet. Broadcom for hardware innovation, Google for cloud integration efficiency, and Palantir for government AI control.
However, Nvidia's ecosystem and R&D pipeline remain horrible. Investors need to monitor technical benchmarks (such as the results of MLPERF) and regulatory developments, particularly in China. The key is to balance exposure between both “old security guards” and “new waves.”
sauce:
[1] Cloud Stock: Broadcom chips in Nvidia's market [https://www.sramanamitra.com/2025/09/05/cloud-stocks-broadcom-chipping-away-at-nvidias-market/]
[2] Nvidia vs. Broadcom: Provided by AI Semiconductor Stock [https://finance.yahoo.com/news/nvidia-vs-broadcom-ai-semiconductor-120800357.html]
[3] Broadcom (AVGO) thrives with custom AI explosions [https://www.barchart.com/story/news/34545182/broadcom-avgo-thrives-in-custom-ai-explosion]
[4] Technology advances as Broadcom benefits at Nvidia's expense [https://www.morningstar.com/news/dow-jones/202509058369/tech-advances-as-broadcom-gains-at-nvidias-expense-tech-roundup]
[5] Nvidia's Blackwell GPU struggles with overheating in data centers [https://winbuzzer.com/2024/11/18/nvidias-blackwell-gpus-struggle-with-overheating-in-data-centers-xcxwbn/]
[6] GPU and TPU Comparative Analysis Report | Bytebridge [https://bytebridge.medium.com/gpu-and-tpu-comparative-analysis-report-a5268e4f0d2a]
[7] GPU and TPU Comparative Analysis Report | Bytebridge [https://bytebridge.medium.com/gpu-and-tpu-comparative-analysis-report-a5268e4f0d2a]
[8] 2Artificial Intelligence (AI) Inventory The US Government is actively supporting it in 2025 [https://www.aol.com/2-artificial-intelligence-ai-stocks-220000990.html]
[9] AI Stock Monitoring: nvidia Stock, Broadcom Stock, pltr… [https://www.markets.com/analysis/ai-stocks-to-watch-nvidia-stock-broadcom-stock-pltr-stock-amd-stock-1]
[10] nvidia stock plummets below key level as AI rivals hit altogether… [https://www.investors.com/research/nvidia-stock-buy-or-sell-now-after-earnings-report-china-chip/]
[11] The first AI benchmark pits AMD against Nvidia [https://www.nextplatform.com/2024/09/03/the-first-ai-benchmarks-pitting-amd-against-nvidia/]
[12] MLPERF Training 4.0 – Nvidia Still King. Power and LLM [https://www.hpcwire.com/2024/06/12/mlperf-training-4-0-nvidia-still-king-power-and-llm-fine-tuning-added/]
