Jensen Huang and Hock Tan Address AI Slowdown Fears as NVDA and AVGO Say Demand for Computing Power Remains Strong

TradingKey - On September 12, Eastern Time, Anthropic CEO Dario Amodei called for controlling the pace at which frontier AI model capabilities advance, heightening market concerns over data center investment and chip demand.

Nvidia (NVDA) CEO Jensen Huang subsequently publicly opposed slowing down AI development, while Broadcom (AVGO) CEO Hock Tan stated that demand for computing infrastructure from frontier model R&D and inference workloads remains strong.

Amodei proposed introducing independent third-party evaluators, encouraging frontier AI enterprises in democratic nations to establish common safety standards, and strengthening international coordination. OpenAI CEO Sam Altman supported introducing an independent evaluation mechanism, and Elon Musk also voiced agreement with Amodei's stance.

These discussions prompted investors to reassess long-term demand for AI infrastructure, leading to adjustments in AI supply chain stocks such as Nvidia and Broadcom on Monday. Nvidia closed down about 3.4%, Broadcom fell over 4%, and the Philadelphia Semiconductor Index plunged 5.9%.

Broadcom expects Anthropic to become its largest XPU customer in fiscal year 2027 and maintain that position in fiscal year 2028; consequently, the impact of the pace of frontier model development on Broadcom's long-term revenue outlook has drawn significant attention.

Hock Tan Maintains $115 Billion and $230 Billion AI Revenue Outlook

On September 14, Hock Tan stated in an interview with CNBC that discussions of an AI development slowdown will not prompt the company to adjust its AI semiconductor revenue forecasts for fiscal years 2027 and 2028. Demand for computing infrastructure driven by frontier model R&D and product inference remains strong and persistent.

During its fiscal third-quarter 2026 earnings call held on September 2, Broadcom projected that its AI semiconductor revenue will reach approximately $115 billion in fiscal 2027 and increase to about $230 billion in fiscal 2028.

The company has secured the supply of advanced process wafers, substrates, and HBM required to support these forecasts, with customer demand exceeding its current revenue outlook. Actual revenue remains dependent on data center construction, power supply conditions, supply chain deliveries, and customer deployment schedules.

Hock Tan expressed greater confidence in inference demand. He noted that training demand remains difficult to gauge, but demand for inference computing power is expected to stay strong once AI models are integrated into commercial products.

Broadcom's AI semiconductor revenue for the third quarter of fiscal 2026 reached $16.7 billion, up 221% year-over-year and accounting for 56.4% of its total quarterly revenue of $29.591 billion. The company expects fourth-quarter AI semiconductor revenue to rise to $21.7 billion, representing a year-over-year increase of 236%.

Jensen Huang Opposes Coordinated Slowdown in AI Development

On September 15, Jensen Huang opposed a coordinated slowdown in AI development at the Dreamforce conference. He stated that safety is the top priority, and companies should not release products if they cannot verify their functionality and safety; if R&D or products are found to be losing control, they should also pause and complete rectifications. Huang believes that development speed and product safety can be balanced simultaneously.

Nvidia's revenue for the second quarter of fiscal 2027 was $96.221 billion, up 106% year-on-year; data center revenue stood at $89.023 billion, up 117% year-on-year and accounting for approximately 92.5% of total revenue.

The strong quarterly performance also supported Jensen Huang's judgment on long-term computing power investments. On September 10, he reiterated at the Goldman Sachs Communacopia + Technology Conference that annual global AI infrastructure spending is expected to reach $3 trillion to $4 trillion by 2030. The development of generative computing and the slowing of Moore's Law will continue to drive growth in computing, networking, and data center investments.

In the earnings report, Huang stated that AI has reached a critical inflection point, computing power is translating into revenue, and demand continues to accelerate. While infrastructure construction was mainly driven by a single AI lab a year ago, there are now more new AI labs and startups, with multiple frontier labs expanding simultaneously, while open models and physical AI are also broadening computing demand.

Will Slowing AI Development Impact Computing Power Demand?

Controlling the pace of capability enhancements in frontier models may increase the time required for training environment checks, model testing, and external evaluations, but it does not mean halting model training.

To date, no public information indicates that major AI companies are cutting data center investments as a result. Model training, safety testing, inference services, agent operations, and enterprise applications all require computing power support.

The latest financial reports, guidance, and management commentary from Nvidia and Broadcom have not shown weakening demand for AI chips. Nvidia projects third-quarter fiscal 2027 revenue of approximately $108 billion; Broadcom expects AI semiconductor revenue to reach about $115 billion and $230 billion in fiscal 2027 and 2028, respectively, noting that customer demand exceeds current forecasts.

Going forward, key areas to monitor include cloud service providers' capital expenditures, data center construction progress, and whether Nvidia and Broadcom adjust their supply arrangements or revenue guidance.