Broadcom’s Q2 AI surge meets a provocative Mizuho TPU forecast aimed at 2028

Broadcom Inc. (NASDAQ: AVGO) reported record fiscal second-quarter results on June 3, with total revenue rising 48% year over year to $22.2 billion and artificial-intelligence related semiconductor revenue jumping 143% to $10.8 billion. The update underscored Broadcom’s growing role supplying specialized silicon and infrastructure components to hyperscale AI data centers, even as the market appears to be debating how fast near-term expectations can be sustained.

In its quarterly outlook, Broadcom guided consolidated third-quarter revenue of approximately $29.4 billion, an 84% year-over-year increase, while also projecting continued expansion in AI semiconductor demand. Chief executive Hock Tan said momentum was driven by growing demand for custom AI accelerators and AI networking, and management expected Q3 semiconductor revenue from AI to grow to $16.0 billion, which would be more than 200% higher than the prior year period.

The new focus comes from a separate argument made by Mizuho, as relayed by 24/7 Wall St., that growth in Google tensor processing unit, or TPU, deployments could create a far larger revenue runway for Broadcom through 2028. TPUs are Google-designed AI chips optimized for accelerating machine-learning workloads, and their mass deployment tends to require a supply chain that includes networking, packaging, and application-specific silicon partners.

According to the Mizuho view summarized in the 24/7 Wall St. piece, TPU-related opportunities could scale to more than 35 million units by 2028 as hyperscalers build out capacity. The note also framed a scenario where Broadcom’s annual AI revenue could rise above $100 billion by 2028, alongside a cumulative opportunity labeled at more than half a trillion dollars, depending on market share assumptions and revenue-per-gigawatt estimates. Those are forecasts, not commitments, and Broadcom did not endorse those specific multi-year figures in its earnings release.

The TPU thesis aligns with how Broadcom describes its positioning across semiconductor solutions. Rather than chasing the general-purpose GPU market, Broadcom has been building customized chip designs and supporting infrastructure that can be tuned to specific AI systems. In the 24/7 Wall St. summary, this custom ASIC approach, where ASIC stands for application-specific integrated circuit, is presented as a way to win repeat deployments tied to large AI platforms.

Recent reporting also points to Broadcom’s involvement in TPU supply chains beyond just chip design. Tom’s Hardware reported that Broadcom agreed to supply Anthropic with roughly 3.5 gigawatts of Google TPU capacity starting in 2027, and that Broadcom’s role includes providing networking and other components for next-generation AI racks, alongside a longer-term collaboration arrangement. It also described Broadcom as a silicon implementation partner that helps convert Google’s TPU architecture into a manufacturable ASIC layout.

Even with the strong Q2 print, the gap between corporate guidance and long-dated analyst scenarios is where the uncertainty remains. Broadcom’s public numbers in the quarter emphasize near-term direction, including Q3 revenue guidance and the expectation for AI semiconductor revenue growth, but it did not provide detailed 2028 commitments or TPU unit forecasts in the materials reviewed here. The Mizuho math, as summarized, should therefore be treated as a directional model for what could happen if hyperscaler capex and TPU ramp schedules follow through.