Broadcom CEO Bets $100B+ on AI Chips by 2027

One number from Broadcom's earnings call sent shockwaves through Silicon Valley: $100 billion. That's what CEO Hock Tan expects AI chip revenue to hit by 2027—"significantly in excess of $100 billion," to be precise.

The bold prediction crushed even Wall Street's most bullish estimates, sending Broadcom shares up 5% and forcing analysts to scramble their calculators. JPMorgan now thinks the company could hit $120 billion or more in AI revenue, assuming $12-15 billion per gigawatt of capacity.

The Copper vs. Optical War

But Tan's confidence did more than boost his own stock. It created instant winners and losers across the semiconductor supply chain.

Amphenol and Credo—makers of copper connectivity technology—surged 4% and 10% respectively. Meanwhile, optical technology companies Lumentum and Coherent tumbled over 4% each. The reason? Broadcom's bet that hyperscalers will choose copper over optical connections for AI servers.

It's a fascinating technical divergence that could reshape how AI data centers are built. Copper is cheaper and more established, but optical promises higher speeds and lower power consumption. Broadcom's backing of copper suggests cost still trumps cutting-edge performance in the AI infrastructure race.

Memory Secured Through 2028

Tan addressed one of the industry's biggest concerns head-on: the high-bandwidth memory shortage that's been plaguing AI chipmakers. His solution? Lock up supply years in advance.

"We have secured memory and leading-edge wafer supply through 2028," Tan told analysts. This forward planning puts Broadcom in stark contrast to competitors scrambling for HBM chips from Samsung and SK Hynix.

The company has already signed up six customers for nearly 10 gigawatts of AI chip capacity—a massive commitment that suggests hyperscalers are serious about custom silicon over Nvidia's off-the-shelf solutions.

Taking Aim at Nvidia's Crown

Tan's most intriguing comments weren't about revenue—they were about strategy. While Nvidia dominates with general-purpose AI chips, Broadcom is betting on custom silicon designed for specific workloads.

"You cannot afford to have a chip that is just good enough," Tan explained. "You need the best chips around because you're competing against other LLM players, and most of all, you're also competing against Nvidia."

It's a direct challenge to Nvidia's one-size-fits-all approach. As AI models become more specialized and companies seek competitive advantages, custom chips could offer better performance per dollar—if Broadcom can deliver.

The $100 Billion Question

Broadcom's confidence contrasts sharply with Nvidia's recent reception. Despite blowout earnings, Nvidia's stock struggled to maintain momentum as investors questioned sustainability. Broadcom seems to have avoided that skepticism, at least for now.

Goldman Sachs analysts noted that Broadcom's "leadership in AI networking and custom silicon enables the lowest inference cost for its hyperscaler customers." That cost advantage could be crucial as AI companies face pressure to make their massive investments profitable.