Balderton, Altimeter, TVG, Coatue, Alpha & Harmonic: Investment in AI Who Gets Funded and Why

Investment Landscape & Stack Economics

• Current generative AI revenue is estimated at ~$200B total, split as:
  • ~$160B at the semiconductor layer (dominated by NVIDIA).
  • ~$30B at the infrastructure/cloud layer.
  • ~$20B at the application layer.
• This inverted "pyramid" structure contrasts with the cloud era, where applications ($400B) outperformed infrastructure ($200B) and semis (~$40–50B).
• The industry is expected to eventually invert, with applications capturing the most value, though the timeline is uncertain; AWS took eight years (2004–2012) to invert its own revenue triangle.
• Early-stage AI startups now require significantly larger capital injections ($50M–$100M), with 60–80% of funds often spent immediately on infrastructure (training/inference) rather than R&D.

Infrastructure Bottlenecks & Capital Requirements

• The primary bottleneck for data center build-outs is power capacity and grid transmission queues; in the US and Europe, queues for megawatts are congested by opportunistic actors seeking to resell capacity.
• Older data centers face a "second-order" obsolescence risk as new NVIDIA/AMD SKUs require higher rack densities that legacy facilities cannot support, necessitating a new CAPEX cycle for infrastructure upgrades.
• High TCO and CapEx intensity in the current GPU stack create a temporary scarcity premium, with value concentrated among buyers of cutting-edge models rather than the "long tail" of smaller apps.
• Margins for current AI applications are significantly lower than traditional SaaS margins due to high compute costs; convergence to SaaS-like margins is anticipated only when cheaper "Generation N-1" models and improved inference efficiency become available.

VC Strategy & Ecosystem Dynamics

• Venture capital economics are shifting toward earlier, larger checks with higher conviction; "average" investments are no longer viable given the Pareto curve where 0.3% of companies (approx. 15 of 5,000) generate 90% of gross profits.
• European startups face capital constraints only at the growth stage; early-stage availability is comparable to the US, but top European firms increasingly rely on US growth capital to scale.
• Investors are expanding beyond software to include deep infrastructure and energy, such as nuclear fusion (e.g., Proxima Fusion), viewing power as a critical strategic asset.

Founder Criteria & Speed to Market

• "Speed" has become the primary moat, but the definition has shifted from mere release velocity to the ability to identify high-value problems without prior learning time.
• Ideal founders now require a dual competence: deep domain expertise to solve specific industry problems and deep technical literacy regarding AI model limitations and capabilities.
• Go-to-market (GTM) strategies for enterprise AI differ from traditional SaaS, requiring heavy change management, on-site implementation, and upskilling of clients rather than simple seat-based sales.
• "AI-first" companies are increasingly expected to operate like consumer brands, using their own AI products internally to drive efficiency before selling them to the market.

Software Development Impact

• Generative AI raises both the "ceiling" (best engineers) and the "floor" (average engineers) of software productivity, increasing overall output but potentially widening the gap between exceptional and average products.
• While the cost of creating functional prototypes approaches zero, the cost of building scalable, compliant, and reliable enterprise systems remains high due to complex architectural and regulatory requirements.
• Future development value will shift toward product management, design, and systems thinking, as basic coding tasks become increasingly automated.

Market Disruption Predictions (Rapid Fire)

• Google: Identified as high-risk due to the shift from search queries to AI chat interfaces, threatening the "10 blue links" search paradigm.
• Apple: Flagged as vulnerable due to perceived lag in AI infrastructure adoption and ongoing regulatory headwinds.
• Microsoft (MAG7): Implied to be less vulnerable due to its heavy integration of AI into existing enterprise workflows and infrastructure.

Notable Examples & Company Metrics

• Higgsfield: Demonstrated extreme speed by releasing 10 products and reaching double-digit millions in ARR within six months.
• Lovable: Raised under $25M to achieve hundreds of millions in revenue, exemplifying a high-return, capital-efficient application model.
• Envelope: Cited as a leading AI-first company in cybersecurity insurance.
• AI-First Accounting: Reported a 4x improvement in average margins by automating tax return filings internally before productizing the service.