Significantly raising Nvidia’s target price, this major bank’s reasons: TSMC’s capacity allocation far exceeds expectations; OpenAI’s “closed-loop transactions.”

Significantly raising Nvidia’s target price, this major bank’s reasons: TSMC’s capacity allocation far exceeds expectations; OpenAI’s “closed-loop transactions.”

```

Amid the market buzz surrounding AI prospects, HSBC has suddenly shifted from a wait-and-see attitude to bullish optimism, sharply raising Nvidia's target price and giving two key reasons.

On October 16th, according to Hard AI news, HSBC upgraded Nvidia's rating to "Buy" in its latest research report, raising the target price from $200 to $320, a potential upside of 78%. This aggressive assessment is not based on vague narratives, but on two crucial signals.

Frank Lee, global head of technology hardware and semiconductor research at HSBC, stated in the research report that Nvidia's allocation of TSMC's CoWoS advanced packaging capacity has increased substantially, with the 2027 fiscal year quota expected to soar from 480,000 wafers to 700,000 wafers, an annual increase of 140%; and a "closed-loop transaction" structure with OpenAI and the Stargate project, which could generate $251–$400 billion in GPU revenue opportunities.

The bank's report said it has raised Nvidia's projected 2027 fiscal data center revenue to $351 billion, the highest forecast on Wall Street, 36% higher than the consensus estimate of $258 billion. The revised 2027 fiscal year earnings-per-share projection is $8.75, which also far exceeds the consensus of $6.48. This implies Nvidia's earnings growth potential may be seriously underestimated.

TSMC Capacity Allocation Surges 140%: Robust Signals from the Supply Chain

HSBC research shows that Nvidia's CoWoS advanced packaging quota at TSMC has reached a historic turning point. After a downward trend in the first half of 2025, the 2027 fiscal year capacity forecast saw a hefty 76% upward revision in October.

Specifically, HSBC raised Nvidia's 2027 CoWoS wafer quota from an earlier estimate of 480,000 to 700,000, meaning 140% year-on-year growth. This is the first time since Q4 2025 that upward momentum in capacity allocation has reappeared.

The report notes that this capacity expansion will not only drive 2027 fiscal data center revenues to $351 billion, 36% higher than market consensus of $258 billion, but more importantly, "confirms again the thesis of an ever-expanding addressable market (TAM) for AI GPUs, transcending reliance on traditional cloud service provider capital expenditure."

"We now see the aggressiveness of Nvidia's forecast for TSMC's CoWoS wafers in the 2027 fiscal year. This will not only drive 2027 fiscal year data center revenue projections beyond consensus, but also reaffirms our thesis that AI GPU TAM continues to grow and surpasses traditional Chip Scale Package."

HSBC said that in an optimistic scenario, if capacity allocation reaches 800,000 wafers, 2027 fiscal data center revenue could hit $390 billion, with EPS of $9.68, 11% above the baseline scenario. This forecast is based on the latest feedback from the supply chain, showing Nvidia's confidence in growing demand for AI GPUs.

Historically, CoWoS capacity expansion has been a leading indicator of Nvidia's earnings beating expectations. In 2023, continued upward adjustments in allocation led to unprecedented earnings growth; the same trend continued in 2024. The current round of capacity expansion signals the start of a new growth cycle.

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

The report reveals a striking triangular relationship: Nvidia invests up to $100 billion in OpenAI, OpenAI signs a five-year cooperation deal worth more than $300 billion with Oracle to deploy Stargate capacity, and Oracle buys GPUs from Nvidia.

Based on bottom-up analysis, just the 18.3GW deployment promised by Stargate and OpenAI could create $251–$400 billion in revenue opportunities for Nvidia, nearly double its current 2027 forecast.

Specifically, considering the power consumption differences of various chip generations:

If all use Blackwell B300 chips (power consumption 1820W), about 10 million chips would be needed, at $40,000 each, for a total revenue of $400 billion;

If all use Rubin chips (power consumption 2340W), about 8 million chips would be needed, at $42,000 each, for a total revenue of $336 billion;

If all use Rubin Ultra chips (power consumption 4176W), about 4.4 million chips would be needed, at $57,000 each, for a total revenue of $251 billion.

HSBC believes this kind of "closed-loop transaction" structure, though sparking concerns about TAM transparency, also solidly translates demand from hypothesis into rigid support for capacity allocation.

This article is from WeChat Official Account "Hard AI". For more cutting-edge AI news, please click here

Risk Warning and DisclaimerThe market entails risk, and investments should be made cautiously. This article does not constitute individual investment advice nor does it take into account the special investment objectives, financial situation, or needs of any particular user. Users should consider whether any opinions, views, or conclusions in this article are appropriate to their specific circumstances. Investing accordingly is at your own risk. ```