Broadcom’s 2027–2028 Gigawatt Outlook and Implied Revenue per GW

Broadcom’s 2027–2028 Gigawatt Projections and Implied Revenue per Gigawatt

Gigawatt projections

The gigawatt figures described on the call are broadly consistent with the investor’s calculation:

The customer-level disclosures support the Anthropic and OpenAI totals: Anthropic is expected to deploy another 5 GW in 2027 and an incremental 10 GW in 2028, while OpenAI is expected to deploy 1.3 GW in 2027 and more than 5 GW in 2028. ⁴ The remaining gigawatts come from Broadcom’s other XPU customers, including Meta, for which Broadcom has line of sight to deploy 3 GW through 2028. ⁶

The important qualification is that these are demand and deployment-visibility figures, not necessarily completed production shipments within each fiscal year. Management explicitly said that the 30 GW represents what customers could deploy if data-center shells and other infrastructure are ready, but that it expects the amount actually coming into production over the two years to be somewhat lower. ² Accordingly, the gigawatt figures should not be interpreted as a precise annual shipment denominator.

Revenue guidance and mechanical revenue per gigawatt

Broadcom provided AI-semiconductor revenue guidance of approximately $115 billion for fiscal 2027 and $230 billion for fiscal 2028, or approximately $350 billion cumulatively over the two years. ⁵⁶

Using the headline gigawatt figures as a simple arithmetic denominator produces the following result:

Thus, the investor’s estimate of roughly $11 billion–$12 billion per gigawatt is mathematically reasonable if the 10 GW and 20 GW figures are used as the respective denominators. ¹²⁵ However, it should be viewed as an implied blended ratio, not as Broadcom’s stated dollar content per deployed gigawatt, because management cautioned that the 30 GW will not necessarily be fully deployed or shipped within fiscal 2027–2028. ²⁵

Management’s stated content outlook

Management’s direct answer was materially higher than the simple $11 billion–$12 billion calculation: Broadcom expects its content to be approximately $20 billion–$30 billion per gigawatt, and expects that level to remain broadly sustainable. ⁷

The apparent discrepancy is primarily a denominator and timing issue:

• The $11 billion–$12 billion figure divides annual AI-semiconductor revenue guidance by the full gigawatt pipeline attributed to that year. ¹²⁵
• The $20 billion–$30 billion figure is management’s stated content estimate for the Broadcom products associated with a deployed gigawatt, while management separately warned that some of the identified gigawatts may not yet be in production or shipped during the relevant fiscal year. ²⁷
• Therefore, dividing $115 billion by 10 GW or $230 billion by 20 GW likely overstates the effective gigawatt denominator for revenue recognized in those fiscal years if deployment is delayed, while the precise mix of XPU, networking, optical and other semiconductor content is not disclosed in the excerpts. ²⁵⁷

The best interpretation is that $11 billion–$12 billion per headline pipeline GW is an implied revenue ratio from the guidance, whereas $20 billion–$30 billion per deployed GW is management’s normalized content framework. Those figures should not be treated as directly interchangeable. ²⁵⁷

Trend as XPUs evolve

Management does not expect dollar content per gigawatt to rise proportionally with XPU performance. The expected mechanism is:

1. Performance and XPU ASPs increase. Each successive XPU generation becomes more powerful, uses more leading-edge and expensive silicon, and therefore carries a higher average selling price. ⁷
2. Power consumption also increases. Higher-performance XPUs consume more power per chip. ⁷
3. Fewer advanced XPUs fit within one gigawatt. Because each chip consumes more power, a fixed one-gigawatt installation contains fewer chips in later generations. ⁷
4. The effects approximately offset at the gigawatt level. Broadcom therefore expects dollar content per gigawatt to remain in the $20 billion–$30 billion range, rather than continuously increasing with every XPU generation. ⁷

In practical terms, the expected trend is:

• Per-XPU ASP: upward. ⁷
• Performance per XPU: upward. ⁷
• Power per XPU: upward. ⁷
• Number of XPUs per GW: downward. ⁷
• Broadcom dollars per deployed GW: broadly stable at approximately $20 billion–$30 billion. ⁷
• Total industry gigawatts: expected to accelerate as customers require exponentially more compute. ⁷

This framework implies that Broadcom’s growth is expected to come less from steadily increasing dollar content within each gigawatt and more from the rapid expansion in the number of gigawatts deployed. Management specifically said that the dollar content should remain relatively stable even as the number of gigawatts grows substantially. ⁷

The content opportunity also extends beyond the XPU itself: Broadcom identified AI networking, Ethernet switching, PCI Express switching, optical DSPs, EMLs, VCSELs and continuous-wave lasers as part of its broader AI-semiconductor portfolio, and expects AI-networking revenue to grow as fast as XPU revenue over the next few years. ⁶ That broader product content is important when interpreting the $20 billion–$30 billion figure, because it indicates that “content per gigawatt” encompasses more than just the accelerator silicon. ⁷⁶

Bottom line

Broadcom effectively confirmed a pipeline of approximately 10 GW in 2027 and 20 GW in 2028, including roughly 6 GW and 15 GW, respectively, from Anthropic and OpenAI. ¹²³⁴ Against AI-semiconductor revenue guidance of $115 billion and $230 billion, the mechanical calculation is approximately $11.5 billion per headline GW in each year, or approximately $11.7 billion per GW cumulatively. ⁵

That calculation is directionally useful but not a clean measure of realized content because the 30 GW pipeline will not necessarily be fully deployed or shipped during the two fiscal years. ²⁵ Broadcom’s own normalized framework is $20 billion–$30 billion of content per deployed gigawatt, with that range expected to remain relatively stable as newer XPUs become more expensive but also consume more power and therefore require fewer chips per gigawatt. ⁷