Uber burned through its entire 2026 AI budget in four months. Now its COO is questioning whether it’s worth it

Uber executives are questioning whether rapid AI adoption is producing enough consumer-facing value to justify rising spending, after reports the company exhausted its 2026 AI coding tools budget in four months. CEO Dara Khosrowshahi said about 10% of committed code is built by autonomous agents, while Uber spent $951 million on R&D in Q1 2026, up nearly 17% year over year. The article signals cost pressure and a more cautious stance on enterprise AI economics, though it is unlikely to move the stock materially on its own.

Analysis

The key signal is not that AI adoption is slowing, but that enterprise buyers are discovering the economics are more elastic than the vendors implied. Once usage moves from assistive tooling to agentic workflows, token consumption scales nonlinearly with task complexity, so the budget blowout is a preview of margin pressure for any company trying to operationalize AI at the team level. That creates a second-order winner/loser split: software vendors with usage-based pricing benefit near term, while enterprise customers face a creeping opex line item that is harder to map to KPIs.

For UBER, the issue is less the headline AI spend and more the governance implication: management is implicitly acknowledging that AI adoption is still a cost center before it is a measurable productivity engine. In a company already investing heavily in autonomy and mapping/logistics, this raises the hurdle rate for internal AI experiments and could shift capital toward initiatives with clearer payback, especially autonomous driving. The market should watch whether the company tightens tool access and replaces broad employee adoption incentives with centralized procurement and usage controls over the next 1-2 quarters.

MSFT is the cleaner read-through. If large enterprises are rationalizing away standalone coding-agent licenses, it suggests channel friction for premium AI seats and a push toward bundled copilots inside existing workflows, which compresses ARPU upside even as gross consumption rises. DUOL’s relevance is indirect but important: it is a warning that AI enthusiasm can outrun product economics, and management reversals on AI narrative tend to hit multiple-expansion names hardest. The contrarian view is that the current skepticism may be too early: if AI tools lift developer throughput only modestly, firms may still pay up because the bottleneck shifts from coding speed to shipping cadence and decision velocity, which is harder to measure but still valuable.

Key Decisions for Investors

• Short MSFT vs long a basket of usage-metered AI beneficiaries for 1-3 months: the risk/reward favors a modest underweight in premium-seat assumptions as enterprises renegotiate licensing; cover if Microsoft reaccelerates Copilot attach in the next earnings cycle.
• Reduce exposure to narrative-heavy, productivity-story software names such as DUOL on rallies over the next 2-6 weeks; the setup is vulnerable to any management commentary implying AI is not yet monetizing end-user outcomes.
• Hold UBER, but hedge with short-dated downside puts into the next quarterly print if AI efficiency metrics are highlighted without matching consumer-level ROI; the market may punish capex/opex creep before autonomy upside is recognized.
• Long picks-and-shovels infrastructure beneficiaries on any enterprise AI spend rotation, especially semis/cloud infra names with metered consumption exposure, as higher token usage is likely to favor the layer selling compute rather than the layer paying for seats.