Skip to main content
Insights•5 min read

The AI Funding Bubble Is Not a Prediction. It Is a Present-Tense Observation.

February 21, 2026 by Asif Waliuddin

AI Valuations
The AI Funding Bubble Is Not a Prediction. It Is a Present-Tense Observation.

The AI Funding Bubble Is Not a Prediction. It Is a Present-Tense Observation.

There is a category of claim that sounds like a prediction but is actually a description of the present. "AI valuations are disconnected from evidence" is one of those. It is not a forecast. It is arithmetic.

Here is one week of AI funding data, read as a unified signal rather than a series of individual announcements.

The Data

xAI: $20 billion raised at $230 billion valuation. Strategic investors include Nvidia, Cisco, and the Qatar Investment Authority. Product: Grok, which is functional but not demonstrably competitive with GPT-4 or Claude on standard benchmarks. The valuation prices in Elon Musk's distribution via X, his ability to attract strategic capital, and the narrative that xAI will become a major independent AI platform.

Thinking Machines Lab: reportedly in discussions at $50-60 billion valuation. Raised $2 billion at $12 billion in July 2025. Product: none public. No demos. No papers. Stealth mode. Founder: Mira Murati (ex-OpenAI CTO). The valuation prices in Murati's credibility and team.

Skild AI: tripled to $14 billion in a single SoftBank-led round. Robotics AI. For context: the entire robotics AI category received $13.8 billion in funding in 2025, up from $7.8 billion in 2024. Skild's valuation in a single round roughly equals the annual funding for its entire category.

Cerebras: $8.1 billion valuation, $1.1 billion raised. Ships wafer-scale engines. Has a $10 billion committed compute contract from OpenAI through 2028. Actual hardware in production serving actual customers.

Deepgram: $1.3 billion valuation, $130 million Series C. Enterprise voice AI. Revenue-backed growth in a specific vertical with demonstrated customer demand.

The Hype

The standard framing for each of these announcements is individual. "xAI raises big round." "Murati's startup soaring." "Robotics AI heating up." Each story gets its own analysis, its own bull case, its own justification.

When you read them individually, each one has a defensible narrative. xAI has Musk's distribution. Murati has elite talent. Skild is riding the robotics wave. Cerebras has a $10B contract. Deepgram has enterprise traction.

The problem is not that any individual narrative is impossible. The problem is what happens when you read them together.

The Reality

Arrange these companies by valuation:

  1. xAI: $230B
  2. Thinking Machines Lab: $50-60B
  3. Skild AI: $14B
  4. Cerebras: $8.1B
  5. Deepgram: $1.3B

Now arrange them by evidence of product capability:

  1. Cerebras: hardware shipping, $10B contract, measurable revenue
  2. Deepgram: enterprise customers, revenue, product-market fit in voice AI
  3. xAI: Grok exists, has users via X, not benchmark-competitive at the frontier
  4. Skild AI: robotics AI company with SoftBank backing; product details sparse
  5. Thinking Machines Lab: zero public output

The two rankings are almost perfectly inverted. The company with the strongest evidence of product and revenue (Cerebras) has the second-lowest valuation. The company with zero public evidence (Thinking Machines Lab) has the second-highest.

This is not a coincidence. It is the market telling you what it values: narrative scale, team pedigree, and strategic positioning -- not product, revenue, or technical evidence.

The Framework: Structural Floors vs. Narrative Premiums

Not all of these valuations carry the same risk. The useful distinction is between valuations with structural economic floors and valuations built entirely on narrative.

Structural floor valuations:

  • Cerebras at $8.1B: Has a $10B committed contract from OpenAI. Even in a downturn, that contract provides a revenue floor. The valuation is backed by hard economics.
  • xAI at $230B: The strategic investor base (Nvidia, Cisco, Qatar) means these investors have distribution and ecosystem incentives beyond pure financial return. Nvidia wants xAI to buy GPUs. Cisco wants to sell networking for xAI data centers. The capital has strategic anchoring.
  • Deepgram at $1.3B: Revenue-backed. Enterprise voice AI has measurable demand. The valuation is high relative to current revenue but has a demonstrable market.

Narrative premium valuations:

  • Thinking Machines Lab at $50-60B: Entirely narrative. The valuation is Murati's resume plus capital availability. There is no product floor, no revenue floor, no contractual floor. If the product disappoints, the entire valuation compresses.
  • Skild AI at $14B: SoftBank narrative premium (SoftBank's AI portfolio strategy systematically pays above-market valuations). The robotics AI market is real but the single-company valuation equals the category's annual funding.

The structural floor valuations will survive a correction. The narrative premium valuations will not -- not because the companies are bad, but because narrative premiums are the first thing to compress when capital tightens.

What This Means

The AI technology is real. The applications are real. The demand is real. None of that is in question.

What is in question is whether $230 billion for Grok and $50 billion for a stealth lab represent accurate pricing of future value, or whether they represent a market where capital availability has exceeded the supply of evidence-based investment opportunities.

Every prior technology cycle had this phase. The internet was real, and Pets.com was still overvalued. Cloud computing was real, and many cloud startups that raised at peak valuations did not survive to justify them. Mobile was real, and the app economy produced thousands of companies valued on download counts that never converted to revenue.

The pattern is consistent: the technology is transformative, the valuations are not sustainable at the margin, and the correction does not invalidate the technology -- it invalidates the companies that raised at peak narrative premium and could not deliver product to justify it.

The Bottom Line

One week of AI funding data, read as a unified signal: the market is pricing AI companies on narrative quality and team pedigree rather than product evidence and revenue. The valuations with structural economic floors (committed contracts, strategic investors with ecosystem incentives, demonstrated revenue) will likely survive a correction. The valuations built on narrative alone will likely compress.

This is not a prediction about timing. It is a description of the current market structure. When Cerebras with a $10B contract is valued at $8B and Thinking Machines Lab with no product is valued at $50-60B, the pricing model is narrative-driven, not evidence-driven.

That model works until capital tightens. Then it does not.