SoundHound vs. NVIDIA After Earnings: Which AI Stock to Buy Now?

NVIDIA reported fiscal Q2 2027 revenue of $96.2 billion, up 106% year over year, led by $89 billion in Data Center revenue (+117%), and guided fiscal Q3 revenue to about $108 billion, plus or minus 2%. Its non-GAAP gross margin rose 250bps to 75%, while Vera Rubin entered full production, reinforcing its AI-infrastructure leadership. SoundHound posted Q2 revenue of $61.9 million (+45%), raised 2026 revenue guidance to $230 million-$260 million, and narrowed its GAAP net loss to $42.8 million from $74.7 million; however, the article favors NVIDIA because of its scale, profitability, and diversified AI ecosystem.

Analysis

The relevant dispersion is not AI exposure but monetization quality. NVDA converts incremental accelerator demand into cash earnings and ecosystem lock-in, whereas SOUN must prove that its deployments translate into durable, high-margin recurring revenue rather than project-heavy implementation revenue. For SOUN, a modest miss on growth or an increase in customer-acquisition/inference costs could re-rate the equity sharply because profitability remains distant; for NVDA, the nearer-term debate is whether hyperscaler capex can remain above already elevated expectations.

Over the next 1-3 months, NVDA's setup depends more on supply availability, customer concentration and the pace at which customers absorb new platform capacity than on headline AI enthusiasm. A clean ramp of the next architecture supports not only NVDA but optical/interconnect beneficiaries AVGO, ANET and CRDO; conversely, any signs of utilization normalization would pressure the entire AI infrastructure complex before it meaningfully affects application-layer vendors. SOUN is a higher-beta expression of enterprise AI adoption, but faces competitive substitution from platform vendors and cloud providers that can bundle voice/agent capabilities into broader enterprise contracts.

Consensus may be too binary in treating NVDA as the "safe" AI choice and SOUN as simply speculative. NVDA's earnings quality is superior, but its multiple is increasingly exposed to even small decelerations in sequential data-center growth or gross-margin dilution during product transitions. SOUN's upside requires evidence that gross-margin progress persists while operating expenses grow materially slower than revenue; absent that proof, revenue growth alone is not a sufficient long thesis.

Falsifiers: reduce NVDA exposure if next-quarter revenue guidance or data-center sequential growth falls materially below the current high-single-/low-double-digit growth regime, or if gross margin declines without a credible mix explanation. For SOUN, require sustained gross-margin expansion, improving operating-loss leverage, and evidence of multi-year contracted backlog; a renewed widening of quarterly cash burn would invalidate a turnaround narrative.

Key Decisions for Investors

• Maintain an overweight NVDA versus SOUN for the next 1-3 months, preferably as a pair trade long NVDA / short SOUN sized beta-neutral. The trade captures the spread between infrastructure earnings conversion and application-layer execution risk; reassess after each company's next guidance update.
• Do not initiate a standalone SOUN long solely on revenue momentum. Upgrade to a tactical long only if the next report shows continued gross-margin expansion, operating-expense growth below revenue growth, and no deterioration in cash burn; otherwise, SOUN remains a candidate short on AI-risk-on spikes.
• Use NVDA strength to add selectively rather than chase: buy shares or 3-6 month call spreads after a broad semiconductor pullback, with upside tied to sustained platform ramp execution. Hedge sector-level capex risk through a partial SOXX or SMH hedge rather than shorting NVDA outright.
• Monitor AVGO, ANET and CRDO as confirmation trades: upward revisions or order commentary around AI networking/interconnect would validate broadening infrastructure spend and support adding to NVDA. Weak order cadence from these suppliers would be an early warning that accelerator demand is becoming more supply-push than customer-pull.
• Set an alert for a material downward revision to hyperscaler capex plans or evidence of lower AI-cluster utilization. Either development would likely compress NVDA's valuation multiple quickly and remove the main justification for maintaining the long leg of the NVDA/SOUN pair.