Broadcom just dropped a bombshell forecast that could reshape the AI chip landscape. CEO Hock Tan told investors the company expects AI chip revenue to hit 'significantly' above $100 billion next year, riding a wave of demand that already pushed quarterly AI sales up more than 100%. The projection positions the semiconductor giant as a dark horse challenger in a market dominated by Nvidia's data center empire.
Broadcom CEO Hock Tan just gave the clearest signal yet that the AI chip boom extends far beyond Nvidia. Speaking to investors, Tan projected AI chip revenue will reach 'significantly' above $100 billion in 2027, a staggering forecast built on momentum that saw AI sales more than double in the company's most recent quarter.
The projection, reported by CNBC, marks a dramatic acceleration for a company that's been quietly building custom AI chips for tech giants like Google and Meta. While Nvidia dominates headlines with its general-purpose GPUs, Broadcom's made itself indispensable by designing application-specific integrated circuits that hyperscalers use to power everything from search algorithms to recommendation engines.
'Significantly' above $100 billion is doing some heavy lifting in that guidance. If Broadcom hits even the low end of that range, it would represent one of the fastest revenue ramps in semiconductor history. The company's total revenue across all segments was around $51 billion in fiscal 2025, meaning AI alone could potentially eclipse the entire company's current sales within two years.
The doubling of AI revenue in the latest quarter suggests Tan's not just being optimistic. Hyperscalers are pouring capital into custom silicon as they race to build proprietary AI infrastructure that doesn't depend entirely on Nvidia's roadmap. Google's TPU chips and Amazon's Trainium processors both rely on Broadcom's manufacturing expertise and networking technology to tie massive chip clusters together.
But Broadcom's play goes deeper than just manufacturing chips. The company's also become the go-to provider for the networking fabric that connects AI accelerators in data centers. Those Ethernet switches and optical components are seeing explosive demand as AI training clusters scale from thousands to hundreds of thousands of GPUs. According to industry analysts, networking can represent 20-30% of total AI infrastructure costs, a margin-rich business Broadcom dominates.
The timing of Tan's forecast is notable. It comes as Microsoft, Meta, and other hyperscalers signal they're not slowing capex spending despite broader economic uncertainty. Microsoft alone pledged over $80 billion in AI infrastructure investments, much of which will flow to chip makers and their supply chains.
There's a competitive edge here too. While Nvidia has captured mind share with its H100 and upcoming Blackwell GPUs, custom ASICs offer hyperscalers better performance per watt for specific workloads and, critically, reduce dependence on a single vendor. That strategic diversification is music to Broadcom's ears and a direct threat to Nvidia's margin fortress.
Still, hitting $100 billion won't be automatic. Broadcom faces execution risk on multiple fronts. Custom chip design cycles stretch 18-24 months, meaning today's orders determine 2027 revenue. Any slowdown in hyperscaler spending, shifts in AI architecture, or manufacturing hiccups at partners like Taiwan Semiconductor Manufacturing Company could derail the trajectory.
The enterprise AI buildout also remains a question mark. While hyperscalers are proven buyers, Broadcom's betting that corporations will follow suit with private AI clouds. That requires enterprises to move beyond experimenting with AI to deploying it at production scale, a transition that's happening but not universally.
Wall Street's already pricing in aggressive AI growth for chip makers, but Tan's forecast suggests even bullish analysts might be underestimating the total addressable market. If custom chips and networking capture a meaningful slice of the estimated $1 trillion AI infrastructure spend over the next decade, Broadcom's positioned to claim an outsized share.
What makes this projection particularly credible is Broadcom's track record. Tan's built a reputation for conservative guidance and consistent execution, traits that stand out in a semiconductor industry known for boom-bust cycles. The company's integration of VMware also gives it software leverage to bundle infrastructure sales, creating stickier customer relationships.
The broader implication is that AI infrastructure spending isn't concentrating in one company's hands. While Nvidia gets the glory for training chips, Broadcom's quietly building an empire in custom inference accelerators and the networking glue holding AI clusters together. That diversification is healthier for the ecosystem and suggests the AI buildout has more room to run than skeptics believe.
Broadcom's $100 billion-plus AI revenue target for 2027 isn't just a big number, it's a signal that the AI chip market is fracturing into specialized segments where custom silicon and networking infrastructure matter as much as raw GPU horsepower. If Tan delivers on this forecast, it validates the thesis that AI's infrastructure layer will create multiple mega-winners beyond the obvious names. For investors and industry watchers, the question now shifts from whether AI infrastructure spending is sustainable to how that spending gets distributed across an increasingly diverse supplier base. Broadcom's betting it can claim a piece nearly as large as the entire company is worth today.
