OpenAI and Anthropic Build Custom AI Chips: Top Stocks to Watch Now

The artificial intelligence arms race is entering a new, hardware-centric phase. For years, the AI industry has been almost entirely dependent on Nvidia’s graphics processing units (GPUs) to power large language models (LLMs) like ChatGPT and Claude. However, a seismic shift is underway. According to a recent report by The Motley Fool, both OpenAI and Anthropic—two of the most valuable private AI companies in the world—are now developing their own custom AI chips.

This move is not just a technical experiment. It is a strategic necessity. Training and running cutting-edge AI models is astronomically expensive. By designing custom silicon tailored specifically to their architectures, OpenAI and Anthropic hope to slash costs, reduce dependence on a single supplier, and accelerate their path toward Artificial General Intelligence (AGI). But what does this mean for public market investors?

If you are looking for the stocks that could benefit the most from this shift, you’ve come to the right place. While the headline news might initially seem like bad news for Nvidia (since it implies lost business), the reality is far more nuanced. Custom chip development doesn’t happen in a vacuum. It relies on a complex ecosystem of foundries, intellectual property (IP) licensors, chip design tools, and interconnect specialists. Below, we break down the top stocks that are uniquely positioned to win as OpenAI and Anthropic build their own AI hardware.

Why Are OpenAI and Anthropic Making Their Own Chips?

Before diving into specific tickers, it’s critical to understand the “why” behind this hardware pivot.

• Cost Control: Training a single model like GPT-4 or Claude 3 costs hundreds of millions of dollars. Custom chips can optimize the cost-per-inference and cost-per-training significantly.
• Supply Chain Independence: Nvidia’s H100 and B200 GPUs are in notoriously short supply. Building custom chips allows OpenAI and Anthropic to design around their own bottlenecks and timelines.
• Architectural Efficiency: General-purpose GPUs are great for many AI tasks, but they are not perfect. Custom chips (often called ASICs – Application Specific Integrated Circuits) can integrate memory, networking, and compute logic in ways that slashes latency and power consumption.
• Competitive Moat: Just as Apple uses custom silicon to differentiate its products, OpenAI and Anthropic see custom chips as a way to build a unique technical advantage that competitors cannot easily copy.

Now, let’s explore the public companies that are poised to supply the tools, manufacturing, and components for this new custom chip wave.

The Top Stocks to Watch Now

1. Taiwan Semiconductor Manufacturing Company (TSMC)

Ticker: TSM

If there is one “picks and shovels” play in this story, it is TSMC. Neither OpenAI nor Anthropic owns a chip fabrication plant. They are “fabless” chip designers. They will design their own silicon, but they will need a world-class foundry to actually build it. TSMC is the only game in town for cutting-edge 3nm and 2nm process nodes.

• Why it benefits: Every new custom chip from OpenAI, Anthropic, or any other AI hyperscaler will likely be fabbed at TSMC. The demand for advanced packaging (like CoWoS, which stacks chips together) is also soaring.
• Key Insight: Even if Nvidia loses some volume to custom designs, TSMC wins regardless—because it manufactures the chips for everyone, including Nvidia, AMD, and now OpenAI.
• Risk: Geopolitical tensions regarding Taiwan remain a persistent overhang.

2. Broadcom (AVGO)

Ticker: AVGO

Broadcom is the silent giant of the custom chip world. The company has a massive business unit called Broadcom’s ASIC division, which designs custom silicon for hyperscale clients. It is widely rumored that Google’s TPU (Tensor Processing Unit) and Meta’s training chips were co-developed with Broadcom’s help.

• Why it benefits: OpenAI and Anthropic are software companies first. They need expert hardware architects. Broadcom offers a turnkey solution: they can design the chip, integrate vital networking (Tomahawk, Jericho switches), and secure the manufacturing at TSMC.
• Key Insight: If OpenAI decides to license pre-verified chip building blocks (PHYs, SerDes, high-speed interconnects) rather than building everything from scratch, Broadcom is the natural partner.
• Catalyst: The company has explicitly stated that it is pursuing multiple “AI custom compute” opportunities with unnamed hyperscalers. This fits perfectly.

3. Marvell Technology (MRVL)

Ticker: MRVL

Marvell is the second major player in the custom ASIC market, often competing head-to-head with Broadcom. Marvell has already announced a partnership with Amazon (AWS) for custom AI chips (Trainium and Inferentia).

• Why it benefits: Marvell has deep expertise in data infrastructure and optimized silicon. They provide the “glue” logic that makes custom chips work inside massive server racks.
• Key Insight: Unlike Broadcom, which tends to work on the “big brain” chip, Marvell focuses on the I/O, storage controllers, and networking chips that are required to move data between the custom GPUs.
• Potential Partnership: As Anthropic and OpenAI expand their cloud partnerships (with Google and Microsoft/Azure respectively), Marvell could be contracted to build specialized chips for inference or data movement within those clouds.

4. Synopsys (SNPS)

Ticker: SNPS

To design a custom chip, you need design software. This is called Electronic Design Automation (EDA). Synopsys is the 800-pound gorilla in this space.

• Why it benefits: Every new custom chip project—whether from OpenAI, Anthropic, or any other firm—requires millions of dollars worth of EDA licenses. Synopsys sells the tools for logic synthesis, verification, and simulation.
• Key Insight: The more custom chips are built, the more “seats” (licenses) Synopsys sells. This is a royalty-based revenue stream that scales with industry complexity.
• Moats: Switching away from Synopsys’ tools is incredibly difficult. Their IP libraries are the industry standard for advanced nodes.

5. Cadence Design Systems (CDNS)

Ticker: CDNS

Where Synopsys excels in logic design, Cadence excels in physical design, timing analysis, and liquid cooling simulation—a critical factor for AI chips that run extremely hot.

• Why it benefits: Custom chip design for AI requires advanced thermal and electromagnetic simulation. Cadence’s tools are essential for ensuring the chips don’t melt under load.
• Key Insight: As AI chips become larger and more complex (moving toward “wafer-scale” designs), the complexity of placement and routing increases exponentially. This drives higher spending on Cadence tools.
• Duopoly Power: Together, SNPS and CDNS control over 80% of the EDA market. They are essentially a toll booth on the custom chip highway.

6. Nvidia (NVDA) – The Paradox Play

Ticker: NVDA

It seems counterintuitive, but Nvidia might still benefit from this trend. Here’s why:

• Hybrid Architectures: Even if OpenAI and Anthropic custom chips become reality, they will likely be used for inference (running the model) or specific training tasks, but not for all workloads. Nvidia’s GPUs will remain essential for the general-purpose training of future frontier models.
• Networking & Software: Nvidia’s CUDA software and NVLink interconnect are deeply entrenched. Custom chips will need to interface with Nvidia systems.
• Long Lead Time: It takes 3–5 years to design a competitive custom chip from scratch. In the meantime, Nvidia’s next-gen chips (Rubin, Blackwell Ultra) will keep dominating.
• Risk: If OpenAI and Anthropic succeed, it represents a long-term threat to Nvidia’s 90%+ market share. However, for the next 12–18 months, Nvidia remains the most powerful AI compute engine.

7. Advanced Micro Devices (AMD)

Ticker: AMD

AMD provides the alternative to Nvidia. As OpenAI and Anthropic seek to reduce dependency on Nvidia, they are likely to diversify their training clusters. AMD’s MI300X and upcoming MI400 series are powerful GPUs that offer strong competition.

• Why it benefits: The custom chip narrative actually helps AMD. It signals that the market is moving away from a single-vendor solution. AMD is the only other public company that can supply high-performance GPUs at scale.
• Key Insight: If OpenAI uses a mix of custom chips (for inference) and AMD GPUs (for training), AMD gains a major customer that was previously locked into Nvidia.

The Dark Horse: Applied Materials (AMAT) & ASML (ASML)

Let’s not forget the guys who build the machines that build the chips. Applied Materials (thin-film deposition) and ASML (extreme ultraviolet lithography) supply TSMC. As custom chip volumes increase, TSMC will need more of their tools.

• AMAT: For every new chip design, there are new material recipes. AMAT dominates this.
• ASML: The monopoly on EUV lithography machines. You cannot make high-performance AI chips without them.

What This Means for Investors

The development of custom AI chips by OpenAI and Anthropic is a major validation of the AI hardware thesis. It confirms that AI is not a passing trend but a fundamental shift in computing that requires dedicated, specialized infrastructure.

Short-Term (0–12 months): Nvidia and AMD remain the primary beneficiaries. The custom chips are years away.

Medium-Term (1–3 years): TSMC, Broadcom, and Marvell become the primary beneficiaries as design contracts are awarded. Synopsys and Cadence see a surge in licensing revenue.

Long-Term (3–5 years): If the custom chips deliver on their promise, we could see a fragmentation of the GPU market. This would benefit hyperscalers and AI companies but could compress margins for GPU suppliers.

The Bottom Line

The news from The Motley Fool that OpenAI and Anthropic are building custom chips is a wake-up call for investors. It signals that the cost of AI is too high for even the richest startups to rely entirely on merchant silicon. The stocks that will benefit the most are not the ones building the models, but the enablers: the foundries (TSMC), the ASIC designers (Broadcom, Marvell), and the tool makers (Synopsys, Cadence).

As with any technology transition, there will be winners and losers. By focusing on the infrastructure layer rather than the application layer, investors can position their portfolios to capture the most value from this historic hardware revolution. Keep your eyes on these names as the custom chip era dawns.