To The Moon

The AI chip market may be entering a new phase.

For years, the AI infrastructure boom has been synonymous with GPUs, with $NVIDIA(NVDA)$ at the center of the buildout. But as hyperscalers increasingly develop chips optimized for their own AI workloads, custom accelerators are becoming a larger part of the semiconductor equation.

According to $JPMorgan Chase(JPM)$'s Fall 2026 U.S. Semiconductor & Semiconductor Equipment Industry Update, ASICs and XPUs are expected to account for 54% of AI accelerator unit shipments in 2027, up from about 41% in 2026, before reaching 55% in 2028. $JPMorgan Chase(JPM)$ also estimates the custom AI ASIC market will reach $60 billion–$70 billion in 2026, with a future CAGR of more than 40%–50%.

📌 Key Insight: The headline is about unit shipments — not revenue, computing power or market value. ASICs can account for a majority of accelerator units while GPUs continue to represent a much larger share of AI-chip revenue.

Still, the forecast points to an important structural shift:

AI computing is becoming more specialized.

1.From “More GPUs” to “The Right Chip for the Job”

The AI boom has created enormous demand for general-purpose accelerators.

NVIDIA's latest results show just how strong that demand remains. In Q2 FY2027, NVIDIA generated $96.2 billion in revenue, up 106% year over year, while Data Center revenue reached $89.0 billion, up 117%. NVIDIA is guiding for approximately $108 billion of revenue in Q3.

So why would companies want something other than GPUs?

The answer is increasingly about efficiency.

A GPU is designed to handle a broad range of workloads. A custom accelerator can instead be designed around a much narrower set of tasks.

That can potentially improve:

• Performance per watt

• Latency

• Cost per token

• Hardware utilization

• Power efficiency

• Supply-chain control

📌 Key Insight: The AI chip race is shifting from simply buying more compute toward optimizing how that compute is delivered.

2.What Exactly Is an ASIC?

An ASIC, or Application-Specific Integrated Circuit, is a chip designed for a particular application rather than a broad range of computing tasks.

Think of the difference this way:

Google's TPU, Amazon's Trainium and Inferentia, and Meta's MTIA are examples of hyperscaler efforts to develop specialized AI silicon.

The trade-off is important.

A custom chip requires significant engineering investment and is less flexible than a general-purpose GPU. But when a company knows exactly what workloads it needs to run — and plans to run them at enormous scale — that specialization can become economically attractive.

📌 Key Insight: ASICs are not universally “better” than GPUs. They are potentially more efficient when the workload is predictable enough and the deployment is large enough to justify the engineering cost.

3.JPMorgan Sees ASICs Passing GPUs on Units

This is where $JPMorgan Chase(JPM)$'s forecast becomes particularly interesting.

AI Accelerator Unit Shipment Mix

$JPMorgan Chase(JPM)$ expects custom accelerators to move from a minority of accelerator shipments in 2026 to more than half by 2027.

But there is an important caveat.

One accelerator chip is simply one physical unit. A Google TPU, Amazon Trainium processor and NVIDIA GPU can differ substantially in computing capability, memory configuration, architecture and selling price.

Therefore:

54% of units ≠ 54% of AI-chip revenue.

📌 Key Insight: JPMorgan's forecast should be interpreted as a shift in deployment volume, not a prediction that GPUs will suddenly lose their economic dominance.

🤔 The key question:

If custom accelerators are taking a larger share of deployments, who captures the value?

That brings Broadcom into focus.

4.Why Are Hyperscalers Building Custom Silicon?

The economics of AI are changing.

As AI models become increasingly used for inference, companies are processing enormous numbers of tokens. At that scale, even relatively small improvements in efficiency can translate into significant infrastructure savings.

For hyperscalers, the calculation increasingly looks like:

More AI workloads → More compute demand → Higher infrastructure costs → Greater incentive to optimize the hardware

A company operating millions of AI workloads has a much stronger reason to build a specialized chip than a smaller customer running a handful of models.

Custom silicon can also give hyperscalers greater control over their infrastructure instead of relying entirely on commercially available accelerators.

📌 Key Insight: The bigger and more predictable the AI workload becomes, the easier it is to justify the upfront cost of designing a custom accelerator.

5.Broadcom Is at the Center of the Custom-Chip Story

JPMorgan estimates that $Broadcom(AVGO)$ and $Marvell Technology(MRVL)$ together account for roughly 90% of the custom AI ASIC market, with Broadcom alone estimated at around 80%–85%.

The market therefore appears highly concentrated.

And $Broadcom(AVGO)$'s recent numbers show why investors are paying attention.

In Q3 FY2026, $Broadcom(AVGO)$ reported:

• $29.6B total revenue, up 86% YoY

• $16.7B AI semiconductor revenue, up 221% YoY

• AI semiconductor revenue up 54% QoQ

• Q4 AI semiconductor revenue guidance of $21.7B, up 236% YoY

$Broadcom(AVGO)$ itself said demand for its custom AI accelerators and networking remained very strong.

📌 Key Insight: Broadcom's AI opportunity isn't limited to the accelerator itself. The company also benefits from the networking infrastructure required to connect increasingly large AI clusters.

6.Google, Meta and OpenAI Are Turning Custom Silicon Into Reality

The custom-chip story is no longer theoretical.

Hyperscalers and AI labs are increasingly developing silicon around their own workloads.

The broader model looks something like this:

Hyperscaler / AI Lab
↓
Custom Architecture
↓
ASIC Design & Connectivity Partners
↓
Foundry
↓
AI Data Center Deployment

$Broadcom(AVGO)$ is deeply involved in this ecosystem.

The company has a long-term relationship with Google around future TPU development and supply, while also providing networking and other components for next-generation AI infrastructure.

OpenAI is taking a similar direction.

In June 2026, OpenAI and $Broadcom(AVGO)$ unveiled Jalapeño, an AI accelerator designed around OpenAI's requirements for large language model inference. Broadcom said the processor was developed from design to production in nine months and is intended for deployment at gigawatt scale across multiple generations.

Meta is also developing its own MTIA accelerator family.

📌 Key Insight: Hyperscalers do not necessarily need to manufacture everything themselves. They can control the architecture and workload optimization while relying on semiconductor specialists for chip design, networking and manufacturing relationships.

7.Marvell Is Smaller — But Its Custom Business Is Accelerating

$Broadcom(AVGO)$ dominates the custom-ASIC discussion, but Marvell remains an important player.

Marvell reported $2.739 billion in Q2 FY2027 revenue, up 37% year over year. Data Center revenue increased 46%, reaching $2.17 billion.

More importantly, Marvell said:

AI-related bookings remain exceptionally robust.

The company also expects a significant acceleration in its Custom business beginning in the second half of fiscal 2027.

Marvell's exposure also extends beyond custom accelerators into connectivity and optical infrastructure.

💬 Your Turn: Join the Discussion

Do you think GPUs and custom ASICs will coexist, or will one side win out by 2028? Which stock would you rather hold: $NVIDIA(NVDA)$, $Broadcom(AVGO)$ or $Marvell Technology(MRVL)$?

Share your view below. Thoughtful comments may receive Tiger Coins! 🪙

Markets are always moving - and sometimes, the best move is knowing what works for you.

With Treasury yields, oil prices and rate expectations keeping markets on edge this week, investors are once again thinking carefully about where to position next. There’s no one-size-fits-all choice in investing — and the same goes for Tiger Merch. This month’s hot picks are in, featuring the Tiger Toiletry Bag, Universal Travel Adapter, Tiger Umbrella and more favourites chosen by fellow Tigers.

Explore the Monthly Hot Picks in Tiger Coin Mall, now 12% OFF for a limited time.

https://laohu8.com/J/redeemGift?goodID=100561&type=delivery