One day before NVIDIA (NASDAQ:NVDA | NVDA Price Prediction) reports fiscal second-quarter results after today’s close, roughly 4:20 to 4:30 p.m. ET, OpenAI announced its first in-house accelerator beats Nvidia’s Blackwell-generation systems on inference. The chip, codenamed Jalapeño, was co-developed with Broadcom on silicon and networking and with Celestica on systems integration, a product of the October 2025 deal to co-develop 10 gigawatts of custom AI accelerators. The timing coincides with earnings, but that alone does not signal impact on tonight’s numbers.
What OpenAI Claims, and What It Doesn’t
OpenAI’s self-reported benchmarks show Jalapeño delivering 1.5x to 1.9x more AI work at peak throughput versus Nvidia Blackwell-generation systems, with 1.7x to 3.6x lower end-to-end latency and 2.1x to 4.1x faster ultra-low-latency interactive inference across GPT-OSS-120B, DeepSeek R1 and Kimi K2.5. Each rack packs 128 accelerators, 1.7 exaFLOPS of 4-bit compute, 27.5 TB of HBM4 and just under 2 petabytes/sec of memory bandwidth. VP of Hardware Richard Ho said the chip “achieves high throughput and low latency simultaneously, a first in the industry.”
Key caveats: Jalapeño handles inference only; Nvidia’s training dominance remains untouched. The comparison targets Nvidia’s GB200 NVL72 and GB300 NVL72 racks that launched in 2024 and 2025, not the upcoming Vera Rubin platform. Tests excluded speculative decoding, and OpenAI did not disclose system-level power, so this measures throughput, not performance per watt. Deployment trickles out later in 2026, with volume production in 2027. AMD’s and Nvidia’s latest racks still deliver 1.46x to 2x more compute and up to 12% more memory than OpenAI’s rack, at roughly 85% of its memory bandwidth. That’s a tradeoff.
Tonight’s Expectations for Nvidia
Analyst consensus clusters around revenue of $91 billion to $92 billion, roughly double year over year, with EPS around $2.09. Nvidia’s prior guide was $91 billion plus or minus 2%. Wedbush’s Matt Bryson expects a beat on both lines, with a $330 price target, noting Nvidia has “consistently exceeded consensus” the last three quarters “yet the stock is roughly unchanged from October of last year.” Shares last settled at $213.05, down 3.04% on the week, up 14.37% year to date. Nvidia’s stock has fallen after four of its last five reports despite beats, a pattern tonight will test.
Custom Silicon as a Slow-Burn Risk
An Investing.com pre-earnings roundup lists hyperscalers building their own chips as one of six named risks facing Nvidia, with a roughly 4% anticipated decrease in GPU market share, framed as a three-to-five-year structural drag. The landscape is crowded: Google’s TPU 8t/8i launched in April 2026, with Morgan Stanley modeling $84 billion to $108 billion in TPU-related Google Cloud revenue for 2027 to 2028; Amazon’s Trainium3 launched December 2025, reportedly past a $25 billion annual run rate; Meta’s MTIA 300 launched March 2026. Broadcom (NASDAQ:AVGO), the vehicle underneath much of this, told investors AI semiconductor bookings hit over $30 billion in a single quarter against $10.8 billion shipped, and guided Q3 AI revenue to $16 billion, up over 200% year on year.
Reasonable people split. CNBC’s Jim Cramer has dismissed the threat, saying he sees “no real competitors” to Nvidia year after year. SemiAnalysis’s Dylan Patel called Jalapeño “huge news,” unusual for first-generation custom silicon to beat Blackwell at all. Keep this distinct from the separate “circular financing” concern about how OpenAI pays for compute; that is a balance-sheet question about financing.
What to Watch After the Close
Tonight’s numbers were locked in well before Jalapeño’s reveal. Orders reflected in the earnings report were placed months ago, and OpenAI’s chip is not deployed at scale. Watch three things: any softening in data-center guidance beyond the ongoing China exclusion, hyperscaler concentration commentary, and whether Jensen Huang addresses custom silicon on the call unprompted. Nvidia’s biggest customers are increasingly its competitors, and Broadcom is the pick-and-shovel supplier to that revolt.
Contact [email protected] for any questions or corrections.