Breakfast News: Week in Review

AI economics are getting cheaper: about 1M tokens cost $0.97, a record low (less than half the summer high), pressuring AI monetization models even as usage needs to grow. Nvidia confirmed a $13B purchase of Hugging Face (expected 1H 2027), after posting ~$96B revenue and guiding for ~70% revenue growth by 2028, while Uber plans to cut ~10% of staff (~3,300 jobs) to fund a $10B+ autonomous-vehicle push but also lobbies for stricter robotaxi rules. In software, MongoDB beat estimates with revenue up 30% YoY and adjusted operating margin rising to 24% (from 15%), yet shares fell ~15% after management forecast a slower quarter.

Analysis

The key shift is that AI is moving from a scarcity story to a distribution story: when inference gets cheaper faster than demand grows, the economic rent migrates away from whoever sells the answer and toward whoever owns the interface, workflow, or compute rail. That is constructive for MSFT if Azure dollars per customer rise with usage, but bearish for any standalone model/API business whose cost base is fixed while price per unit keeps falling. Over the next 1-3 months, the market will likely reward companies that can prove volume elasticity; over 6-18 months, it will penalize any AI business that cannot defend gross margin with product lock-in.

Nvidia’s platform moves read as defensive capital allocation, not just expansion. Buying influence around developer tooling and chip design lowers the odds that customers fully abstract away from its hardware, but it also signals the market that the easy monopoly-like phase is ending and future growth will require more capital and more ecosystem control. That can cap multiple expansion even if earnings stay strong. By contrast, UBER’s regulatory posture is a near-term moat for the existing network, because slowing autonomous penetration preserves dispatch economics and driver supply balance; the tradeoff is that it delays the optionality embedded in its own AV spend.

MongoDB looks more interesting tactically than structurally: the selloff reflects duration risk, not obvious demand destruction. If booked demand keeps converting, the market will eventually have to re-rate the name on collected revenue rather than headline growth slope. The contrarian miss is that AI price compression is not uniformly bearish—it's bullish for application layers with sticky data gravity, while the real losers are the vendors trying to monetize model access as a standalone product.

Key Decisions for Investors

• Tactically buy MDB on further weakness over the next 1-2 weeks; target a 3-6 month mean reversion if conversion of contracted revenue remains intact. Falsify if the next quarter shows another guide-down or material slippage in backlog conversion.
• Stay long MSFT into the first Azure dollar-revenue print; use it as the cleanest test of whether AI usage is outrunning token deflation. Add on confirmation of dollar growth acceleration; fade if growth decelerates while AI traffic rises.
• Avoid chasing NVDA at current levels; the Hugging Face/MediaTek moves are ecosystem defense, not a new demand cycle. If long, consider trimming into strength and wait for a better entry after any multiple compression tied to capex scrutiny.
• Relative-value long UBER / short GOOG or GOOGL for a 1-3 month window if robotaxi rulemaking tightens in key states. The thesis is that regulation extends UBER’s cash-flow runway while delaying autonomous competition; exit if state-level rules fail or Waymo expands faster than expected.
• For more aggressive risk-taking, pair long MDB / short SNOW on the idea that the market will pay for collected revenue and operating leverage, not just AI adjacency. Falsify if SNOW’s growth re-accelerates while MDB’s conversion stalls.