Broadcom Hits $2 Trillion Market Cap as Google Reveals New AI Chips

Key Points

• Broadcom's market capitalization momentarily rose above $2 trillion, driven up by a recent announcement from Google.
• Google's next-generation TPU includes two separate chips, confirming past statements made by Broadcom.
• See the benefits that two-chip deployments can bring to the semiconductor giant.

As Broadcom NASDAQ: AVGO has surged to new all-time highs, the firm has a key partner to thank: Google parent company Alphabet NASDAQ: GOOGL.

On April 22, the semiconductor giant moved to what was at the time a record closing high near $422. In turn, Broadcom briefly eclipsed a $2 trillion market capitalization.

This was the result of a new artificial intelligence (AI) chip announcement from Google, Broadcom’s top customer. The stock rose 5% that day, although shares have traded down slightly since.

Google’s announcement provides proof of Broadcom’s past predictions and indicates that the relationship between these companies is only growing stronger.

Google’s 8th Gen TPU: 2 Chips Instead of 1

Google co-develops its tensor processing units (TPUs) with Broadcom, with the two having partnered on the technology for around a decade. Now, Google has announced its eighth generation of the product, but with a key distinction this time around.

The eighth generation will feature two chips, not one, which specialize in different use cases. This marks a significant shift from the norm, as the majority  of prior TPU generations only encompassed one chip. Google notes that TPU 8t will be a “training powerhouse," while TPU 8i will be a “reasoning engine” serving the most latency-sensitive inference workloads.

Google’s announcement lines up almost exactly with what Broadcom CEO Hock Tan has said in the past. In the company’s last earnings call, Tan said that customers will “start to develop two chips each year simultaneously, one for training, one for inference to be specialized.” These statements came in early March. For Broadcom, this is a key win for multiple reasons.

Why Google’s 2-Chip Roadmap Benefits Broadcom

First off, by splitting the eighth-generation TPU into two chips, Broadcom can likely benefit more as general AI adoption increases. Adoption aligns with inference usage, or AI models answering questions and performing tasks. With a customized inference chip in TPU 8i, Google can scale inference workloads more efficiently.

Google notes that TPU 8i delivers 80% better performance per dollar compared to the seventh-generation TPU. Google says this allows businesses to serve nearly twice the inference volume at the same price. With TPU 8i reducing costs, inference demand should increase. This should thereby increase demand for the chips themselves, a clear positive for Broadcom.

Furthermore, by making two specialized chips, Broadcom’s hardware expertise is becoming increasingly important. In single-chip deployments, companies can optimize the chip to be better at certain workloads using software. However, the optimization gains possible using a software approach are far less than those from optimizing at the hardware level. As Google introduces two chips, it is aligning with this view.

For Broadcom, optimization moving away from software and to hardware is a good thing, as hardware is Broadcom’s specialty. This shift could allow the firm to capture more value per chip.

Additionally, developing two chips requires more work than just developing one. Broadcom should expect to receive compensation for this added work, likely allowing the company to generate more revenue from the chip design process.

Lastly, the announcement is but another proof point of exactly what Hock Tan previously said. Meta Platforms NASDAQ: META, another Broadcom customer, announced its custom chip roadmap in mid-March. Meta titled the announcement “Four MTIA [Meta Training and Inference Accelerator] Chips in Two Years: Scaling AI Experiences for Billions.” This also aligns with the two chips per year framework, with the Google and Meta announcements clearly adding legitimacy to Broadcom’s statements.

Importantly, Broadcom says it is seeing these two-chip-a-year roadmaps across five customers. With two customers confirming this, the level of confidence that all five will do so increases. This could allow the benefits discussed from Google’s two-chip deployment to extend across the rest of Broadcom’s hyperscaler customer base.

Broadcom Soars as Hyperscaler Earnings Approach

Overall, Broadcom continues to show that customers are deepening their relationships with the company. As investors notice Broadcom’s increasing importance within the AI buildout, shares are moving to unprecedented levels.

All the while, Broadcom trades at a depressed forward price-to-earnings (P/E) ratio compared to the recent past. With earnings expectations now more aligned with the stock’s valuation, Broadcom’s ability to hold recent gains, or potentially move higher, improves.

A potential slowdown in hyperscaler AI spending remains a risk, but one that has not surfaced at this point. Fast-approaching earnings from multiple Magnificent Seven companies should provide more data on the spending outlook.