Jim Cramer calls Nvidia earnings a referendum on AI trade

Reflection AI, a New York City startup valued at $25 billion, is rumored to be on the verge of releasing its first open-source AI model, according to a report from Axios. Founded about 2.5 years ago by two former Google DeepMind researchers who worked on the foundations of Google Gemini, the company came out of stealth at a $545 million valuation and has since raised $2 billion from Nvidia, which led its next round. Reflection AI has committed upwards of $7 billion on compute before its first model ships, including roughly $150 million per month paid to Elon Musk's SpaceX AI for compute access, and it does not own any of its own chips. The company's pitch centers on an "AI factory" model in which enterprises pay for model weights, download them, and run them privately on their own infrastructure so proprietary data never reaches a centralized lab. The launch would mark a rare Western attempt to compete in open-source AI, where Chinese models now account for 58 to 80 percent of usage among US AI startups, and where Reflection AI's $7.4 billion compute spend amounts to only about 1.4 percent of the $0.5 trillion Anthropic plans to spend on compute over the next 5 to 10 years.

AMD CEO Lisa Su said overall data center demand remains "very strong" and that she is seeing "strong demand signals" across the company's product portfolio during a visit to Seoul on Wednesday to meet with Samsung Electronics and SK Hynix, according to a Reuters report, a day after she predicted in Taipei that chip demand would stay strong for years. AMD shares are trading at a record high and are up 200% year to date. Citi analyst Atif Malik raised his price target on AMD to $800 from $575 on Tuesday, citing demand for Meta's new AI agent known as Muse and calling AMD the primary beneficiary of a CPU renaissance, with Meta among the largest customers of AMD's server business. Malik said he expects the CPU total addressable market to expand from $29 billion in 2025 to $300 billion in 2030, a 60% CAGR, with AMD the key beneficiary and Intel a secondary one.

Jim Cramer said on the September 28 episode of Mad Money that Advanced Micro Devices' push into artificial intelligence has opened another growth opportunity as the industry shifts from model training toward inference and agentic AI, pointing to the stock's roughly 13% gain the previous week and its brief move above a $1 trillion market capitalization. AMD's second-quarter revenue rose 50% year over year to $11.5 billion, with Data Center revenue more than doubling to $6.7 billion and accounting for 58% of total revenue, and the company expects approximately $13 billion of third-quarter revenue, implying roughly 41% year-over-year growth and 13% sequential growth at the midpoint. Cramer credited CEO Lisa Su, who took over in October 2014, with AMD's turnaround, saying the stock has rallied an astonishing 18,433% since she arrived, while noting NVIDIA remains ahead in high-end AI accelerators. AMD also announced on September 28 that it agreed to acquire AI model developer World Labs for approximately $8.2 billion in an all-stock transaction, with co-founder Fei-Fei Li expected to join as executive vice president and chief scientist after the deal's expected closing by the end of 2026. Cramer cautioned that the latest leg higher was in part a stroke of good luck tied to the market's shift toward inference and agentic AI, and AMD's own filings flag that U.S. export restrictions affect its MI308 products and that China sales depend on customer demand, import-control rules and licensing, while its 54% GAAP gross margin in the second quarter compared with 40% a year earlier that included $800 million of inventory and related charges tied to those export controls. AMD trades at approximately 39.5x forward earnings, above NVIDIA at nearly 25x but below Intel at 63.3x, and 164 hedge funds held the stock in the second quarter, up from 134 in the first, with short interest at 2.46% of the public float heading into the second half of September.