Mira Murati creating Her with Thinking Machines, year after she broke up with OpenAI and Sam Altman

Mira Murati's Thinking Machines, founded in 2025, unveiled demos of a new real-time interactive AI system trained from scratch to handle audio, video, and text natively. The company is reportedly seeking investment at a valuation above $50 billion, signaling strong investor interest in next-generation AI. The article is largely about product innovation and venture momentum rather than immediate financial results.

Analysis

This is less a product launch than an escalation in the competitive frontier for AI interfaces. If real-time, multimodal interaction works even at a partial level, it shifts value away from pure model quality toward latency, orchestration, and user retention — a structural advantage for the platform that can make AI feel persistent rather than episodic. That dynamic is favorable to the incumbent with the largest distribution surface and enterprise footprint, because the winning layer may be the one that owns the session, memory, and default workflow, not necessarily the smartest standalone model.

The immediate beneficiary is likely Microsoft, but not because it needs to win the consumer companion race outright. The second-order effect is that any credible threat to ChatGPT-style interaction increases the strategic value of owning the OS, productivity suite, and cloud inference stack, where MSFT can bundle, route, and monetize agentic experiences across work contexts before a startup can acquire comparable distribution. This also pressures competitors to spend more on multimodal inference and always-on context, which is margin-dilutive in the near term and may force a wave of model-led feature parity launches over the next 3-12 months.

The market may be underestimating the commercialization risk: demos that feel magical in controlled settings often break on reliability, safety, and user tolerance once scaled. The first-order enthusiasm can last weeks; the real gating issue is whether this becomes a consumer habit or just another enterprise pilot, and that should be judged over 2-4 quarters. A failure mode is regulatory scrutiny around always-on audio/video and emotional dependency, which would slow adoption sharply and favor players with deeper compliance infrastructure.

Contrarianly, this is bullish not just for AI leaders but for the infrastructure layer that sells picks-and-shovels into every arms race. More interactive models mean more inference-heavy workloads, more GPU demand, and higher attach rates for cloud, data, and orchestration tooling; those economics are better for the hyperscaler than the startup unless the startup can monetize at scale very quickly. In that sense, the headline looks like a threat to incumbents, but the cash flow winner over the next year may still be the platform provider that hosts everyone else's experimentation.

Key Decisions for Investors

• Long MSFT vs. short a basket of AI application startups over 3-6 months: the asymmetric advantage is distribution + enterprise bundling, while startups face higher burn to match multimodal parity; target 1.5-2.0x relative performance if enterprise copilots re-rate.
• Buy MSFT Jan-2027 calls on pullbacks: use a delayed catalyst window because the monetization of real-time interaction is likely a 2-4 quarter story, not an immediate revenue inflection; calls capture optionality with limited downside if the market overreacts to startup hype.
• Add long positions in hyperscaler inference beneficiaries over 6-12 months: the thesis is that always-on multimodal agents increase GPU and cloud utilization regardless of who wins the interface layer, creating a second-order tailwind for infrastructure economics.
• Avoid chasing private-market AI enthusiasm at this stage: if valuation headlines for the startup keep expanding without proof of retention or unit economics, expect a sentiment air pocket in 1-2 quarters when product reality collides with scaling costs.
• If available, structure a call spread on MSFT into any AI-feature launch selloff: the downside is muted by existing enterprise monetization, while upside comes from the market assigning higher strategic value to owning the interaction layer.