NEW YORK — Broadcom held its footing Tuesday as the rest of the semiconductor sector retreated into NVIDIA earnings caution. Shares added $2.27 to close at $191.34 — a 1.2% gain — as investors leaned into the company’s positioning as the primary alternative to NVIDIA for hyperscale AI infrastructure.
The custom accelerator story is the one analysts have spent 2026 telling, and Broadcom is its main character outside of NVIDIA. The company designs application-specific integrated circuits — ASICs — for customers who want AI processing power without paying NVIDIA’s premium or accepting NVIDIA’s general-purpose architecture. Alphabet’s TPU (Tensor Processing Unit) series, Meta’s MTIA (Meta Training and Inference Accelerator), and two additional unnamed hyperscalers have all contracted Broadcom to design their custom silicon.
Broadcom CEO Hock Tan has been more specific in recent months than the company had previously been willing to be. On the most recent earnings call in June, Tan said the company’s AI revenue for fiscal year 2026 would reach $12 billion, up from a prior estimate of $10 billion and from $8.9 billion in fiscal 2025. He attributed the revision to larger-than-expected order volumes from existing customers and the addition of at least one new ASIC development contract. The backlog — which represents committed but not yet recognized revenue — is estimated by KeyBanc at approximately $12 billion, based on contract disclosures and typical development-to-deployment timelines.
The ASIC advantage Broadcom has built is not primarily technical; it is relational. Designing a custom accelerator takes eighteen months to three years of intensive co-development between Broadcom engineers and the customer’s AI research teams. Once that investment is made, switching costs become enormous. A hyperscaler that has deployed 100,000 TPUs in a Broadcom-designed architecture does not casually migrate to a different silicon supplier. The lock-in dynamic is structurally similar to the one TSMC has built in advanced foundry manufacturing, and it is the reason analysts have consistently upgraded their long-term AVGO price targets through 2026.
The networking business is the other AI-related segment that has driven Broadcom’s stock. The company’s Ethernet switching chips — sold under the Tomahawk and Trident brands — are the dominant networking silicon inside AI data centers globally. Connecting thousands of GPUs for distributed training requires ultra-low-latency networking at a scale that Broadcom’s chips handle more efficiently than alternatives. Network revenue was $5.8 billion in the most recent quarter, up 43% year-over-year.
NVIDIA’s earnings Wednesday matter to Broadcom, but less directly than to memory suppliers or pure-play GPU companies. The bull case for AVGO does not require NVIDIA to have a bad quarter. It requires only that hyperscalers continue investing in AI infrastructure — and that at least some of that investment flows to custom accelerators rather than exclusively to NVIDIA’s hardware.
Broadcom completed its acquisition of VMware in late 2023 for $69 billion, one of the largest technology acquisitions in history. The VMware integration has been complex — Broadcom’s decision to shift VMware’s licensing model from perpetual to subscription generated significant customer complaints and drove some enterprise clients to evaluate competing virtualization platforms. The financial results have been stronger than the customer relations. VMware’s contribution to Broadcom’s infrastructure software revenue in the most recent quarter was $5.3 billion, with subscription conversion running ahead of the company’s internal targets.
At $191.34, Broadcom trades at approximately 33 times forward earnings consensus for fiscal year 2027. The stock has risen roughly 68% in the past twelve months, driven by the AI custom silicon narrative and the VMware integration demonstrating higher-than-expected margins. The 52-week high is $207.45, set in mid-July. The gap from Tuesday’s close to that high represents the market’s uncertainty about whether fiscal 2027 AI revenue acceleration can meet increasingly ambitious Wall Street models.
The undisclosed hyperscaler customer is the largest open question in Broadcom’s near-term story. Tan referenced a new ASIC development contract without naming the company. Analysts have speculated about Amazon AWS, Microsoft Azure, or Apple as candidates — each of which has publicly discussed ambitions to reduce reliance on third-party AI chips. Whoever it is represents a multi-billion-dollar revenue stream that is not yet fully reflected in analyst models.
Related Posts
