Nvidia shares slipped while Broadcom surged after OpenAI's GPT-6 Astra debut. Jim Cramer's latest call spotlights the shifting power struggle among AI chipmakers and what it means for investors betting on the next wave of AI infrastructure
The launch of OpenAI's GPT-6 Astra model sent ripples through Wall Street's AI chip sector, sharply dividing the outlook for Nvidia and Broadcom. On September 8, Nvidia shares fell nearly 2% to $225.80, while Broadcom rose almost 3% to $368.17. This divergence was driven by OpenAI's significant infrastructure investment and renewed public analysis from Jim Cramer, who identified both companies as major beneficiaries-though for distinct reasons.
Broadcom's AI semiconductor revenue soared 221% year over year in fiscal Q3 2026, reaching $16.7 billion and making AI the company's main growth engine in the quarter.
Broadcom, in contrast, is focusing on the inference stage-the operational deployment of trained AI models. Its custom Jalapeno chip, developed with OpenAI, is designed for efficiency and speed in real-world applications. OpenAI reports that Jalapeno achieves 1.5 to 1.9 times more AI work per watt than Nvidia's GB200 and GB300 chips, with lower latency across several benchmarks. While this challenges Nvidia's dominance, it does not represent a straightforward replacement.
Jim Cramer maintains that Nvidia remains the primary stock for those investing in the AI buildout, citing its established ecosystem and continued demand from OpenAI as central strengths. However, Cramer's charitable trust also holds Broadcom, Intel, and Micron, reflecting his view that the AI infrastructure expansion benefits multiple players rather than a single winner.
AI infrastructure arms race
Broadcom's AI strategy is centered on custom silicon for large customers rather than general-purpose GPUs, and its Q3 2026 report showed total revenue of about $29.6 billion for the quarter ended August 2, 2026.
Anthropic, another leading AI company, is projected to surpass Google as Broadcom's largest XPU customer in 2027, with OpenAI expected to be second. Anthropic plans to deploy 5 gigawatts of Broadcom TPU 8i-designed chips in 2027 and 10 gigawatts in 2028, while OpenAI is set to use 1.3 gigawatts of Jalapeno capacity in 2027. These figures represent concrete infrastructure commitments that will influence the competitive landscape for years to come.
Some analysts remain cautious about Broadcom's trajectory. KeyBanc's John Vinh, for instance, noted the risk that Google could shift more chip business to MediaTek-based silicon, and he continues to favor Nvidia as the top pick in the sector. The market's split response to the GPT-6 Astra launch-Nvidia declining, Broadcom rising-illustrates how investors are weighing these competing dynamics in real time.
Investor calculus and market mechanics
For both retail and institutional investors, the implications are immediate. The AI buildout is not solely about technological superiority; it also concerns which companies secure early revenue, control the ecosystem, and leverage customer relationships. OpenAI's infrastructure decisions carry significant weight due to its capital requirements and competition from Anthropic and open-source alternatives.
Cramer's diversified portfolio-holding Nvidia, Broadcom, Intel, and Micron-reflects the reality that no single supplier dominates the entire stack. The market's reaction to the GPT-6 Astra launch underscores that even with a strong AI thesis, the timing and sequence of returns can vary widely. Investors focusing only on headline customers or recent benchmarks may overlook the underlying mechanics of revenue flow within the supply chain.
According to The Globe and Mail's coverage of Broadcom's Q3 earnings, AI chip revenue growth has become the company's primary growth driver. Management has raised its fiscal 2026 AI semiconductor revenue outlook to approximately $58 billion, with projections of $115 billion in 2027 and $230 billion in 2028.
As of September 2026, the Philadelphia Semiconductor Index (SOX) has risen 18% year to date, outpacing the S&P 500's 9% gain over the same period. Nvidia's market capitalization remains above $1.2 trillion, while Broadcom's has surpassed $600 billion. These numbers reflect the scale of investor confidence-and the potential for rapid reversals if infrastructure investments or customer relationships weaken.
Investors pursuing the AI chip surge should recognize that the market does not reward all suppliers equally, even when serving the same end customer. The GPT-6 Astra rollout demonstrated how quickly sentiment can shift between training and inference, and between established leaders and emerging competitors. Ultimately, the most successful companies will be those that secure both technological leadership and customer loyalty, converting infrastructure spending into lasting profitability. Investing in the AI wave requires understanding where value is created and captured as the industry evolves.
The distinction between training and inference chips is fundamental to the AI hardware competition. Training chips, such as Nvidia's Grace Blackwell, are optimized for building and refining large models, demanding significant computational power and memory bandwidth. Inference chips, like Broadcom's Jalapeno, are engineered for efficiency and speed in deploying models at scale. For investors, identifying which companies lead in each segment-and tracking shifts in customer preferences-can determine who benefits most from the next phase of AI growth.
