Significantly raised NVIDIA's target price, the reason from this major bank: TSMC's capacity allocation far exceeds expectations, OpenAI's "closed-loop trading"

Amid the market's clamor over the prospects of AI, HSBC has suddenly shifted from a wait-and-see approach to an aggressive bullish stance, significantly raising its target price for NVIDIA and providing two key reasons.

On October 16, according to Hard AI news, HSBC stated in its latest research report that it has upgraded NVIDIA's rating to "Buy," with the target price soaring from $200 to $320, representing an upside potential of 78%. This aggressive judgment is supported not by vague narratives, but by two key signals.

Frank Lee, the head of global technology hardware and semiconductor research at HSBC, stated in the report that NVIDIA's allocation of advanced CoWoS packaging capacity at Taiwan Semiconductor has significantly increased, with the expected quota for fiscal year 2027 jumping from 480,000 wafers to 700,000 wafers, a year-on-year increase of 140%; and the "closed-loop transaction" structure formed with OpenAI and the Stargate project is expected to bring $251 billion to $400 billion in GPU revenue opportunities.

The bank noted in the report that it has raised its fiscal year 2027 data center revenue expectation for NVIDIA to $351 billion, becoming the highest forecast on Wall Street, which is 36% higher than the market consensus of $258 billion. The adjusted fiscal year 2027 earnings per share expectation is $8.75, also significantly ahead of the market consensus of $6.48. This indicates that NVIDIA's profit growth potential may be severely underestimated.

Taiwan Semiconductor's Capacity Allocation Soars 140%: Strong Signals from the Supply Chain

HSBC's research shows that NVIDIA's allocation of CoWoS advanced packaging capacity at Taiwan Semiconductor has undergone a historic turnaround. After experiencing a downward trend in the first half of 2025, the capacity expectation for fiscal year 2027 saw a significant 76% upward adjustment in October.

Specifically, HSBC raised NVIDIA's CoWoS wafer quota for fiscal year 2027 from the previous estimate of 480,000 wafers to 700,000 wafers, indicating a year-on-year growth of 140%. This is the first time since the fourth quarter of 2025 that we have seen a resurgence in capacity allocation momentum.

The report emphasizes that this capacity expansion will not only drive data center revenue for fiscal year 2027 to $351 billion, which is 36% higher than the market consensus of $258 billion, but more importantly, it "reconfirms the argument that the overall addressable market (TAM) for AI GPUs is continuously expanding, surpassing reliance on traditional cloud service providers' capital expenditures."

"We are now seeing the aggressive nature of NVIDIA's forecast for Taiwan Semiconductor's CoWoS wafers for fiscal year 2027, which will not only lead to an upward adjustment of data center revenue for fiscal year 2027 beyond consensus forecasts but also reiterates our argument about the growing TAM for AI GPUs that surpasses traditional Chip Scale Package."

HSBC stated that in an optimistic scenario, if the capacity allocation reaches 800,000 wafers, data center revenue for fiscal year 2027 could reach $390 billion, corresponding to earnings per share of $9.68, which is 11% higher than the baseline scenario This forecast is based on recent feedback from the supply chain, indicating that NVIDIA is confident in the continued growth of demand in the AI GPU market.

Historically, the expansion of CoWoS capacity has been a leading indicator of NVIDIA's performance exceeding expectations. In 2023, the continuous increase in capacity quotas brought unprecedented profit growth; this trend is expected to continue in 2024. The further expansion of current capacity allocation signals the beginning of a new growth cycle.

The Astonishing Scale of OpenAI's "Closed-Loop Transactions"

The report reveals an intriguing triangular relationship: NVIDIA invests up to $100 billion in OpenAI, OpenAI signs a five-year cooperation agreement worth over $300 billion with Oracle to deploy Stargate capacity, while Oracle purchases GPUs from NVIDIA.

HSBC estimates through bottom-up analysis that the 18.3GW deployment committed by Stargate and OpenAI alone could create revenue opportunities of $251 billion to $400 billion for NVIDIA, nearly double its current forecast for fiscal year 2027.

Specifically, based on the power consumption differences of various chip generations:

If all Blackwell B300 chips (power consumption of 1820W) are used, approximately 10 million chips are needed, resulting in total revenue of $400 billion at a unit price of $40,000;

If all Rubin chips (power consumption of 2340W) are used, approximately 8 million chips are needed, resulting in total revenue of $336 billion at a unit price of $42,000;

If all Rubin Ultra chips (power consumption of 4176W) are used, approximately 4.4 million chips are needed, resulting in total revenue of $251 billion at a unit price of $57,000.

HSBC believes that this "closed-loop transaction" structure, while raising market concerns about TAM transparency, also effectively transforms demand from assumptions into a rigid support for capacity allocation