2 Stocks to Buy as Google Starts Selling Its AI Chips

Broadcom's AI semiconductor revenue rose 221% year over year to $16.7 billion in fiscal Q3 and is guided to roughly $21.7 billion in Q4, while total revenue increased 86% to $29.6 billion and free cash flow nearly doubled to $13.7 billion. Alphabet began recognizing TPU system-sale revenue in Q2, with most contracted revenue expected in 2027; Google Cloud revenue grew 82% to $24.8 billion and operating margin expanded to about 36% from 21%. Alphabet's cloud backlog increased by more than $50 billion sequentially to $514 billion, although its 2026 capex guidance of $195 billion-$205 billion and hardware-related costs remain key investment risks. Both companies trade below their 52-week highs, with Broadcom down about 31% and Alphabet about 16%.

Analysis

External TPU systems create a meaningful strategic change only if they convert Google from a captive accelerator user into a credible merchant alternative for frontier-model labs. That would pressure NVIDIA's pricing power at the margin, but the nearer-term effect is more likely incremental AI infrastructure spend: customers adopting TPUs still require high-bandwidth networking, optical interconnect, rack integration, and power/cooling. AVGO is advantaged because it monetizes accelerator content without bearing the end-customer cloud-utilization risk; the offset is that its valuation multiple will remain highly sensitive to a small number of hyperscaler procurement decisions.

For GOOG, the important variable is not initial system-sale revenue but whether installed TPU customers subsequently purchase managed cloud, software, support, and networking services. Hardware mix can dilute consolidated gross margin and pull forward working-capital needs, so a market that rewards backlog growth without monitoring cloud operating margin and capex-to-revenue conversion could be complacent. Over the next 1-3 months, confirmation of signed external deployments and supplier-order visibility is a positive catalyst; over 6-18 months, utilization, repeat orders, and Cloud margin retention determine whether this deserves a platform multiple rather than being treated as capital-intensive equipment revenue.

Consensus may overstate the direct NVIDIA displacement risk. Large AI buyers generally multi-source accelerators to improve negotiating leverage and workload fit, while CUDA/software portability and ecosystem depth preserve NVIDIA's moat. The more investable second-order risk is that aggressive custom-silicon deployments reduce the scarcity premium on AI compute, shifting value from accelerator vendors toward networking, memory, power infrastructure, and cloud-service providers.

Key Decisions for Investors

• Initiate a 6-12 month long AVGO / short SMH relative-value position, sized market-neutral: AVGO has greater direct custom-ASIC and connectivity exposure, while SMH carries broader exposure to any moderation in merchant-GPU spending. Target 10-15% relative outperformance; exit if AVGO reports AI revenue growth decelerating below 30% quarter-over-quarter or a top customer materially cuts infrastructure guidance.
• Accumulate GOOG on post-earnings or capex-driven weakness rather than chase headline momentum; use a 12-18 month horizon. Underwrite only if Cloud operating margin remains broadly stable despite external hardware delivery and backlog converts into revenue; reassess if capex rises faster than revenue for two consecutive quarters or Cloud margin falls more than 300 bps.
• Do not short NVDA solely on TPU commercialization. Set an alert for evidence of merchant-GPU order cancellation, reduced NVIDIA data-center guidance, or sustained hyperscaler capex reallocation toward ASICs; absent those signals, TPU adoption is diversification rather than a thesis-breaking demand event.
• Monitor suppliers and adjacent beneficiaries—ANET, MRVL, VRT, and ETN—for order/commentary confirmation. A broad external TPU rollout should increase network and power-density demand; lack of corresponding backlog growth would indicate that deployments are largely internal reallocations rather than incremental industry spend.