The Shifting AI Chip Landscape: Why Top Funds Are Abandoning Nvidia for Broadcom, Google, and Palantir

The AI chip market is undergoing a seismic shift. For years, Nvidia (NVDA) reigned supreme, its GPUs powering everything from generative AI to high-performance computing. But in 2025, top funds are reallocating capital toward Broadcom (AVGO), Google (GOOGL), and Palantir (PLTR), betting on a more fragmented and specialized AI ecosystem. Here’s why—and what it means for investors.

Broadcom: The Custom Chip Challenger

Broadcom’s AI revenue surged 63% year-over-year to $5.2 billion in Q3 2025, driven by custom silicon tailored for hyperscalers like Google and Meta [1]. The company’s XPUs—Application-Specific Integrated Circuits (ASICs)—are designed for specific workloads, offering higher efficiency and cost-effectiveness compared to Nvidia’s general-purpose GPUs [2]. For instance, Broadcom’s Tomahawk Ultra Ethernet switch, with 102.4 Tbps capacity, enables low-latency connectivity for AI clusters, addressing a critical bottleneck in large-scale deployments [3].

Financially, Broadcom’s AI business is on track to generate $20 billion annually, positioning it as the second-largest player behind Nvidia’s $60+ billion run rate [4]. A $10 billion custom chip deal with a new cloud client and partnerships with four hyperscalers underscore its momentum [1]. Meanwhile, Nvidia’s dominance faces headwinds: its Blackwell GPUs, while powerful, have encountered thermal challenges in high-density data centers, delaying deployments and raising questions about long-term stability [5].

Google: Efficiency Over Raw Power

Google’s Tensor Processing Units (TPUs) continue to outperform in power and cost efficiency. The latest TPU v5p delivers up to 460 petaFLOPS per pod, with near-linear scaling on models like GPT-3 175b [6]. This makes TPUs ideal for Google’s cloud-based AI workloads, where energy consumption and total cost of ownership (TCO) are critical. In contrast, Nvidia’s H100/H200 GPUs, while dominant in raw FLOPS (141 teraflops FP8 for the H200), face scrutiny over power efficiency, particularly in the Blackwell architecture [7].

Google’s strategic focus on custom silicon is paying off. By integrating TPUs into its AI infrastructure and generative models like Gemini, the company is locking in demand for its cloud services. For investors, this represents a dual play: hardware and software synergy that rivals Nvidia’s standalone GPU model.

Palantir: The AI Software Powerhouse

Palantir isn’t in the chip business, but its AI-driven analytics platforms are reshaping fund allocations. The company’s $795 million DoD contract extension and a potential $10 billion Army deal highlight its dominance in defense and government AI [8]. Its Gotham and Foundry platforms integrate generative AI to automate decision-making, making it indispensable for agencies handling vast data sets.

While Palantir lacks hardware benchmarks, its Rule of 40 score (83% adjusted operating margin + growth) outpaces many peers [9]. Top funds are betting on its ability to monetize AI in niche sectors, particularly as governments prioritize secure, ontology-driven solutions. This contrasts with Nvidia’s broad but increasingly crowded market.

Nvidia’s Challenges: Can It Maintain Its Crown?

Nvidia’s Q4 FY2024 revenue hit $35.1 billion, but its stock recently fell below the 50-day moving average, signaling short-term jitters [10]. Competitors like AMD and Broadcom are closing the gap: AMD’s MI300X, for example, matches Nvidia’s H200 in some inference benchmarks [11]. Regulatory hurdles, particularly in China, further complicate Nvidia’s growth trajectory.

Yet, Nvidia’s CUDA ecosystem and Blackwell’s 2.2x performance leap in MLPerf Training 4.0 keep it ahead [12]. The question is whether its thermal issues and rising competition will erode margins over time.

What’s the Takeaway for Investors?

The AI chip landscape is diversifying. Broadcom’s custom ASICs, Google’s efficient TPUs, and Palantir’s software-first approach are challenging Nvidia’s hegemony. For funds, this means spreading bets: Broadcom for hardware innovation, Google for cloud-integrated efficiency, and Palantir for government AI dominance.

However, Nvidia’s ecosystem and R&D pipeline remain formidable. Investors should monitor technical benchmarks (e.g., MLPerf results) and regulatory developments, particularly in China. The key is balancing exposure to both the “old guard” and the “new wave.”

Source:
[1] Cloud Stocks: Broadcom Chipping Away at NVIDIA's Market [https://www.sramanamitra.com/2025/09/05/cloud-stocks-broadcom-chipping-away-at-nvidias-market/]
[2] NVIDIA vs. Broadcom: Which AI Semiconductor Stock Offers [https://finance.yahoo.com/news/nvidia-vs-broadcom-ai-semiconductor-120800357.html]
[3] Broadcom (AVGO) Thrives In Custom AI Explosion [https://www.barchart.com/story/news/34545182/broadcom-avgo-thrives-in-custom-ai-explosion]
[4] Tech Advances as Broadcom Gains 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 GPUs Struggle 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 | by ByteBridge [https://bytebridge.medium.com/gpu-and-tpu-comparative-analysis-report-a5268e4f0d2a]
[7] GPU and TPU Comparative Analysis Report | by ByteBridge [https://bytebridge.medium.com/gpu-and-tpu-comparative-analysis-report-a5268e4f0d2a]
[8] 2 Artificial Intelligence (AI) Stocks the U.S. Government Is Actively Backing in 2025 [https://www.aol.com/2-artificial-intelligence-ai-stocks-220000990.html]
[9] AI stocks to watch: 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 Plunges Below Key Level As AI Rival Hits All- ... [https://www.investors.com/research/nvidia-stock-buy-or-sell-now-after-earnings-report-china-chip/]
[11] The First AI Benchmarks Pitting 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/]