Marvell Technology shares fell roughly 8% in premarket trading Friday after chief executive Matt Murphy told investors that the semiconductor company's marquee custom AI chip agreement with Google will not become materially meaningful until fiscal 2029, well after some Wall Street models had priced in.
Deal Worth Up To $120 Billion Through 2033
Marvell recently secured an agreement with Google that could generate as much as $120 billion in revenue through fiscal 2033 and potentially make Alphabet one of Marvell's largest shareholders. On the earnings call, Murphy said the ramp will be back-loaded and that fiscal 2029 marks the point at which the Google program becomes a substantial revenue contributor, later than some investors had assumed even as the company posted stronger overall results.
Custom Silicon Becomes A Long, Capital-Intensive Bet
Google, Amazon, Microsoft, Meta and OpenAI are all designing specialized processors rather than relying only on off-the-shelf GPUs, and companies like Marvell and Broadcom sit at the center of that shift. But hyperscaler custom silicon takes years to design, validate, manufacture and deploy, and investors appear increasingly unwilling to value every large announcement as near-term revenue. The trend echoes patterns visible across the sector, from Cerebras' wafer-scale roadmap to Intel's Diamond Rapids Xeon push.
NVIDIA Still Dominant, But Second Pillar Is Growing
Marvell's Google deal illustrates a broader shift: NVIDIA remains dominant in AI training compute, but custom accelerators and networking silicon are becoming a strategically important second pillar of hyperscale compute. The market reaction shows how sensitive share prices have become to the exact timing of that build-out, particularly for firms with concentrated exposure to a single anchor customer — a dynamic also visible in the AWS-NVIDIA 2 million GPU order.
Reporting based on Reuters coverage.