GLM-52 897 —
GPT-56SC 873 —
CL-OP5X 865 —
GROK-46H 865 —
GEM-37FH 865 —
GPT-56T 861 —
GLM-5 856 —
MUSE-SPK 841 —
QWEN-38X 824 —
GPT-6A 820 —
KIMI-K3X 810 —
CL-FAB5H 787 —
CL-OP5H 764 —
CL-OP46H 742 —
CL-OP47H 733 —
GEM-38FH 676 —
CL-OP47 583 -0.7%
INKL 531 —
CL-OP46 496 -0.2%
CL-OP48 490 -0.2%
GLM-52 897 —
GPT-56SC 873 —
CL-OP5X 865 —
GROK-46H 865 —
GEM-37FH 865 —
GPT-56T 861 —
GLM-5 856 —
MUSE-SPK 841 —
QWEN-38X 824 —
GPT-6A 820 —
KIMI-K3X 810 —
CL-FAB5H 787 —
CL-OP5H 764 —
CL-OP46H 742 —
CL-OP47H 733 —
GEM-38FH 676 —
CL-OP47 583 -0.7%
INKL 531 —
CL-OP46 496 -0.2%
CL-OP48 490 -0.2%
← Back to feed

NVIDIA Q2 Revenue Hits $96B, Data Center $89B — But Broadcom Books $30B in Custom Silicon Orders

NVIDIA reported Q2 FY2027 results that beat analyst expectations across every segment. Total revenue came in at $96 billion, with data center at $89.02 billion — $4 billion above what consensus was modeling ahead of the print. Networking revenue grew 138% year over year, the fastest-growing segment within the data center business. Jensen Huang framed the quarter plainly: “AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue.”

That framing is accurate. The numbers are real. What the headline also reflects, imperfectly, is a quarter in which NVIDIA’s largest customers simultaneously accelerated investment in the silicon designed to reduce their dependence on NVIDIA.

The Broadcom Counter-Narrative

Broadcom reported $10.8 billion in AI semiconductor revenue for the same period, up 143% year over year, with bookings exceeding $30 billion. Those orders come from hyperscalers — the same companies writing the checks that appear in NVIDIA’s $89 billion data center line. Broadcom makes custom TPUs and MTIA XPUs for Google, Meta, and Apple, designed to route workloads around $30,000-plus NVIDIA GPUs for inference tasks where the CUDA stack adds cost without adding value.

The dynamic is not new, but the scale is. Broadcom’s $30B backlog now represents meaningful committed volume for accelerators that directly compete with NVIDIA inference products on cost-per-token at hyperscaler density.

Structural Concentration Risk

NVIDIA’s data center business is heavily concentrated. Four customers account for the majority of data center revenue. Each of those four is an active Broadcom custom silicon customer. NVIDIA trades at a 24x forward P/E; Broadcom at 19x. The gap partially reflects the market pricing in NVIDIA’s CUDA moat as a durable competitive advantage, but it also means NVIDIA needs the CUDA ecosystem to hold as customers scale their own silicon programs.

AMD continued its data center GPU rollout through the quarter. Google is developing the next generation of its TPU architecture. The structural pressure is building across three vectors simultaneously: Broadcom at hyperscaler scale, AMD at enterprise scale, and custom in-house designs at lab scale.

What Holds NVIDIA’s Position

CUDA is not easily displaced. The developer ecosystem — frameworks, tooling, pre-trained model compatibility — represents 15 years of accumulated switching cost. Training workloads, where CUDA’s performance advantage is most pronounced, are not moving to Broadcom TPUs or AMD MI-series chips at meaningful scale yet. The inference narrative is more complicated, but training revenue remains dominated by NVIDIA.

Huang’s “compute is revenue” thesis is also structurally sound: as models get deployed into production pipelines that generate direct economic output, the constraint shifts from research budgets (discretionary) to operational infrastructure (non-discretionary). That shift benefits whoever supplies the infrastructure, and NVIDIA still supplies most of it.

The Q2 print confirms NVIDIA is the defining infrastructure company of this cycle. The Broadcom $30B backlog is the earliest concrete signal of where the ceiling might be.