Broadcom shares are rising after announcing a long-term agreement with Google to develop chips powering faster and more efficient artificial intelligence models, reinforcing the scale of global AI infrastructure demand.
BNN Bloomberg spoke with Nicholas Mersch, portfolio manager at Purpose Investments, who said the deal — alongside rapid growth at Anthropic — points to accelerating enterprise adoption and sustained investment in compute capacity.
Key Takeaways
- Broadcom’s agreement to supply Google’s AI chips reflects continued expansion in large-scale computing tied to enterprise AI adoption.
- Anthropic’s sharp revenue growth signals a shift from experimentation to meaningful enterprise spending on AI tools.
- Planned compute capacity reaching multiple gigawatts underscores that AI investment is now operating at infrastructure scale.
- The partnership strengthens alternatives to Nvidia, though its dominance in supply chains and pricing remains significant.
- Canadian firms such as Celestica are positioned to benefit by supplying critical data centre infrastructure tied to AI growth.
Read the full transcript below:
ANDREW: Shares in Broadcom are getting a lift today. The semiconductor company has announced a long-term agreement with Google to develop and supply devices responsible for running faster and more energy-efficient large AI models. The companies are also working with Anthropic to power that company’s growing operations.
We’re joined by Nicholas Mersch, portfolio manager at Purpose Investments. Nicholas, always great to see you. Thanks very much.
So these are Tensor Processing Units. Google apparently invented them, and they help machines learn faster.
NICHOLAS: Yeah, that’s right. This industry is filled with acronyms. And this headline this morning is really exciting — it’s a great time to be investing in many of these technology companies.
There are several important details in this announcement. First, Anthropic’s revenue growth. It was at a $9 billion annualized run rate at the end of last year, and over the past three months that has increased to $30 billion. That’s a significant jump and a strong signal of demand for artificial intelligence.
We’re seeing widespread adoption of these tools, with companies actively implementing them. That’s driving increased usage and inference, which in turn requires more computing capacity.
The second key point is the scale of infrastructure. The buildout is expected to expand from one gigawatt in 2026 to 3.5 gigawatts by 2027. That suggests there is no slowdown in demand for AI infrastructure.
It also highlights growing competition to Nvidia. Google’s TPUs, supported by Broadcom, offer an alternative for companies looking to build custom silicon and reduce reliance on Nvidia’s pricing power.
ANDREW: Anthropic is not public yet, but could it be a major IPO?
NICHOLAS: Yes, it could be. There are a few large IPOs being discussed, including Anthropic and OpenAI.
Anthropic’s positioning has been strong, particularly with enterprise use cases such as coding. Meanwhile, OpenAI has faced some challenges around strategic direction and capital allocation.
As a result, Anthropic is gaining momentum and appears to be competing effectively among leading large language model developers.
ANDREW: Have you been a buyer of Broadcom lately?
NICHOLAS: Yes. Broadcom is an attractive name in this space. The company is positioning itself as an alternative provider of custom AI chips for large customers.
It already has relationships with major players such as Google, Anthropic, Meta and OpenAI, along with additional customers that have not been publicly identified.
Nvidia remains dominant, but there are efforts to expand capabilities on the inference side. Nvidia has also moved in that direction through partnerships, including with companies focused on inference performance.
At the same time, Google’s stack, supported by Broadcom, is gaining traction. So you’re seeing competing ecosystems develop.
ANDREW: Is that Elon Musk’s Grok?
NICHOLAS: No, this is different. Elon Musk’s Grok is separate. This refers to Groq, which focuses on hardware designed for fast inference.
The industry is shifting more toward inference — using models repeatedly — because that’s where efficiency gains are being realized.
ANDREW: We’re tight for time. What about Celestica? The stock is down about 16 per cent from November. Would you be a buyer?
NICHOLAS: Celestica fits well into this theme. It’s one of the clearer ways to gain exposure to AI infrastructure from a Canadian perspective.
The company builds key components such as switches, racks and interconnects that support data centres. These are essential even if they receive less attention than chips.
Celestica has confirmed it is a manufacturing partner for Google’s TPU systems. That places it directly within this ecosystem.
As infrastructure investment increases, that demand flows through to companies like Celestica. Given the recent pullback, it’s an interesting opportunity.
ANDREW: Thanks very much, Nick. Always great speaking with you. Nicholas Mersch, portfolio manager at Purpose Investments.
---
This BNN Bloomberg summary and transcript of the April 7, 2026 interview with Nicholas Mersch are published with the assistance of AI. Original research, interview questions and added context was created by BNN Bloomberg journalists. An editor also reviewed this material before it was published to ensure its accuracy and adherence with BNN Bloomberg editorial policies and standards.
