Broadcom's $30 Billion Signal From the Future

Broadcom's latest earnings report revealed a tension at the center of its AI business. The company disclosed more than $30 billion in quarterly AI bookings, expanded its core customer count to six and said it now has visibility into customer demand through 2028. Yet management maintained its forecast of more than $100 billion in annual AI semiconductor revenue by 2027.

The decision stood out because Broadcom first introduced the $100 billion target in March, when it reported $8.4 billion in quarterly AI semiconductor revenue and outlined a path to surpass that level by 2027. Since then, the company has added another core customer and announced new commitments involving Google, Anthropic and Meta.

Investors were looking for evidence that Broadcom's expanding AI business would translate into a higher long-term target. Instead, the company left its forecast unchanged and reported weaker-than-expected software revenue, contributing to a post-earnings selloff.

"Our visibility runs all the way to 2028 right now," CEO Hock Tan said during the call. "Just three months ago, I can tell you visibility ran pretty much [to] 2027. Today, it runs to 2028."

The statement raises a question that runs through the rest of the earnings report: if customer commitments, bookings and planning horizons have all expanded, why has Broadcom's long-term revenue forecast remained the same?

The Deals Behind Broadcom's 2028 Visibility

Broadcom's confidence is tied to agreements and commitments that stretch well beyond the current fiscal year.

In April, the company signed a long-term agreement with Google to develop and supply future generations of Tensor Processing Units (TPUs) and networking technology through 2031, according to Reuters. That agreement expanded a relationship that has already produced multiple generations of Google's custom AI chips.

The same month, Anthropic expanded its use of Google Cloud and TPU infrastructure. Broadcom said on its earnings call that it is providing access to more than one gigawatt of TPU-based compute for Anthropic this year and entered an agreement in April to provide access to an additional five gigawatts of next-generation TPU-based compute beginning in 2027. Anthropic separately confirmed the expansion of its TPU partnership with Google.

Broadcom also outlined the next phase of its work with OpenAI. Tan said the company has already delivered silicon and remains on track for production in late 2026. Tan said Broadcom has a contractual commitment to deploy 1.3 gigawatts of compute in 2027 as part of what he described as a broader 10-gigawatt agreement extending through 2029.

Meta represents another multi-year commitment. Reuters reported in April that Broadcom and Meta expanded their partnership through 2029 to develop multiple generations of Meta's custom AI accelerators.

The customer list continues to grow. Broadcom now describes six core AI customers, up from five earlier this year. Tan said the company's two other AI customers are expected to begin shipments in late 2026 and accelerate deployments during 2027. To date, Broadcom has received purchase orders totaling $6 billion from those two customers.

Viewed together, these commitments provide Broadcom with a clearer view of future deployment schedules than it had earlier this year. Broadcom did not disclose how much of that visibility is tied to firm contractual commitments versus customer forecasts and deployment plans.

AI Infrastructure Is Being Planned in Gigawatts

The June earnings call offered a glimpse into how AI infrastructure planning is changing.

Rather than discussing chip volumes, Broadcom repeatedly framed future demand in terms of gigawatts of compute capacity. The company outlined deployment commitments spanning multiple gigawatts across Anthropic, OpenAI, Meta and Google.

Broadcom expects to ship approximately 10 gigawatts of compute capacity during 2027. Tan said those deployments are expected to be concentrated in the second half of the year, creating momentum that could carry into 2028.

The company's order book reflects that planning cycle. Tan said customers are placing orders further in advance because compute deployments now require coordination across multiple parts of the infrastructure stack.

"These six customers realize that getting access to compute requires lead time[s]," Tan said. "They are placing their orders early ... and they are placing orders in fairly huge demand."

The company also pointed to infrastructure requirements outside semiconductors. During the question-and-answer session, Tan referenced power availability, memory supply and broader infrastructure planning as factors customers must align before deploying large AI clusters.

Broadcom announced an AI XPU Platform with Apollo, Blackstone and other investors that it said is intended to support more than 20 gigawatts of compute capacity. Tan said the first tranche of the platform, valued at $35 billion, is already being launched.

Despite discussing networking infrastructure, financing platforms and large-scale AI deployments throughout the call, Broadcom continued to define its role primarily through semiconductors. That point became clear during an exchange with analysts about rack-scale AI systems and their potential effect on margins.

"No racks. No. No. No. No racks. It is all a chip business only. We [are] only chips. Only chips," Tan said.

The remark stood out because Broadcom spent much of the call discussing networking infrastructure, financing platforms and multi-gigawatt deployments. Yet Tan consistently framed the company's long-term opportunity as a semiconductor business rather than a rack-scale systems business.

Broadcom reported that networking represented almost 40% of AI revenue during the quarter. Tan said he expects networking to normalize closer to 30% of total AI revenue over time as custom AI accelerators become a larger part of the mix.

Management reiterated its expectation that AI semiconductor revenue will reach $56 billion during fiscal 2026 and exceed $100 billion in 2027.

Broadcom says it can now see customer demand extending into 2028. Whether that longer planning horizon ultimately translates into revenue beyond the company's existing forecast remains one of the key questions following the earnings report.