Key Takeaways
- GF Securities upgraded Broadcom (AVGO) price target to $450 following increased Google TPU shipment projections
- Revised TPU shipment estimates now stand at 4.5M units for 2026 and 7.9M for 2027, compared to prior 6M forecast
- Broadcom’s Ironwood and Sunfish chips currently lead as the sole SKUs available for customer evaluation
- CEO Hock Tan projects visibility into exceeding $100B in AI-related revenue by 2027
- Wall Street analysts propose the $100B target could be understated, with calculations suggesting $180B–$200B potential
Broadcom (AVGO) shares traded at $320.55, gaining 0.71%, during Wednesday’s morning session.
The semiconductor giant is capturing renewed market focus following GF Securities’ price target increase and mounting speculation that Broadcom‘s self-imposed $100B AI revenue projection for 2027 might significantly underestimate actual potential.
GF Securities analyst Alicia Xia elevated her price objective on the stock to $450 from a previous level, maintaining her Buy recommendation. This adjustment followed her revised upward estimates for Google’s Tensor Processing Unit (TPU) shipments.
Xia’s updated projections anticipate total TPU volumes reaching 4.5M units during 2026, expanding to 7.9M in 2027. The 2027 estimate represents a substantial increase from her previous 6M unit forecast, primarily attributed to anticipated external sales expansion.
Broadcom-specific TPU volume projections stand at 4.1M units in 2026, rising to 5.8M in 2027. The company’s Ironwood and Sunfish chip platforms currently represent the exclusive SKUs prepared for customer validation. MediaTek’s competing Zebrafish chip hasn’t entered testing phases yet, providing Broadcom with a competitive timing advantage.
Xia also anticipates average selling prices will continue their upward trajectory, with Broadcom’s forthcoming Pumafish chip potentially exceeding $20,000 in 2027 owing to its enhanced architectural complexity.
Debating the $100B Projection
The more compelling narrative centers on whether Broadcom‘s internal AI revenue projection substantially underrepresents future performance.
During the company’s March 4 earnings conference call, CEO Hock Tan indicated Broadcom maintains “line of sight” toward surpassing $100B in AI-generated revenue during 2027. While that figure appears substantial, analysts participating in the call challenged the estimate — not to reduce it, but to suggest it’s too conservative.
Bernstein analyst Stacy Rasgon attempted to establish a more granular projection using gigawatts — representing data center power capacity — as a measurement framework. He tallied Broadcom’s documented customer commitments: 3 GW from Anthropic, 1 GW from OpenAI, a minimum 2 GW from Meta, and at least 3 GW from Google. His calculations point to approximately 9–10 GW for 2027.
Tan validated this assessment, confirming Broadcom is “seeing it getting close to 10 gigawatts” in 2027.
Rasgon subsequently applied his revenue-per-gigawatt calculation of approximately $20B. Bank of America analyst Vivek Arya corroborated this methodology, referencing Broadcom’s 2026 Anthropic deployment — where 1 GW is projected to generate $20B in revenue.
Using $20B per gigawatt across 10 GW, the mathematics yields a range between $180B and $200B — substantially exceeding the official guidance.
Tan didn’t dispute the analysis aggressively. He acknowledged that per-gigawatt revenue varies across customers, but concurred the estimate is “not far off.”
Component Supply Secured Beyond Forecast Horizon
Broadcom has additionally locked in supplies of advanced-node wafers, high-bandwidth memory modules, substrates, and T-glass extending through 2028 — a full year beyond its current projection timeframe.
Melius analyst Ben Reitzes observed that Broadcom was likely the first semiconductor company to secure these critical components with such extended visibility.
The consensus price target among analysts covering AVGO currently stands at $435.30, suggesting approximately 36% upside potential from present levels, based on 33 analyst ratings.


