Tech P/E Gap Hits 7-Year Low as Cerebras Files $35B IPO

◆ DEEP DIVES

1. 01

   Cerebras $35B IPO: The Warrant-for-Compute Model That Resets AI Infrastructure

   <h3>Why This Matters Now</h3><p>Cerebras is filing IPO paperwork <strong>as soon as today</strong>, targeting a <strong>$35 billion+ valuation</strong> — a 60% premium to its $22B private round just two months ago. The IPO aims to raise more than <strong>$3 billion</strong>. But the valuation anchor is extraordinary: OpenAI has committed <strong>$20-30 billion over three years</strong> for Cerebras-powered servers, with ~$1B in data center funding and equity warrants that could give OpenAI <strong>up to 10% of Cerebras</strong> as spending scales.</p><h3>The Structural Innovation</h3><p>This isn't just a chip company going public — it's a <strong>new financial architecture for AI supply chains</strong>. OpenAI is vertically integrating into its compute supplier through spending-linked equity rather than M&A. Every AI infrastructure deal in your pipeline will be measured against this template. Sources converge on the implication: expect anchor customers to demand <strong>5-15% equity participation</strong> through warrants in future AI infra term sheets. This changes dilution math and ownership economics for every infrastructure startup in your portfolio.</p><h4>The Bull-Bear Framework</h4><table><thead><tr><th>Dimension</th><th>Bull Case</th><th>Bear Case</th></tr></thead><tbody><tr><td><strong>Revenue anchor</strong></td><td>$20-30B committed = unprecedented S-1 narrative</td><td>Customer concentration above 50% carries 20-30% IPO discount historically</td></tr><tr><td><strong>NVIDIA disruption</strong></td><td>First demand-side defection at scale from NVIDIA</td><td>NVIDIA inference demand growth may outpace diversification</td></tr><tr><td><strong>Warrant structure</strong></td><td>Creates aligned incentives between buyer and supplier</td><td>OpenAI holds renegotiation leverage; long-term margin risk</td></tr><tr><td><strong>Valuation precedent</strong></td><td>Sets public market benchmark for AI chip companies</td><td>$35B+ on pre-commercial revenue is highly contingent</td></tr></tbody></table><h3>Cross-Source Intelligence</h3><p>Multiple sources confirm this deal signals <strong>deliberate diversification away from NVIDIA</strong>. Jensen Huang's emotional response on China chip restrictions — calling nuclear proliferation comparisons <em>"lunacy"</em> — reveals strategic pressure from both supply-side restrictions and demand-side defection. The NVIDIA bull case now requires inference demand growth to <strong>outpace customer diversification</strong> — a tighter thesis than six months ago.</p><p>Simultaneously, <strong>xAI is renting excess GPU capacity to Cursor</strong> at below-hyperscaler rates, creating an entirely new compute arbitrage layer. AI model companies becoming cloud providers is collapsing the infrastructure stack faster than expected. Portfolio companies spending $1M+ annually on cloud compute should be evaluating non-traditional providers — the short-term savings could reach <strong>30-50%</strong>.</p><blockquote>The Cerebras IPO creates a binary signal: if it prices at $35B+, every private AI infra deal in your pipeline reprices upward overnight. If it struggles, customer concentration risk gets repriced across the sector.</blockquote>

   Action items

   • Model Cerebras IPO scenarios at $25B, $35B, and $40B and map implications for every AI infra company in your pipeline before the pricing window closes
   • Audit portfolio company compute contracts for warrant/equity kicker structures this quarter
   • Evaluate xAI/alternative compute providers for portfolio companies currently on AWS/Azure/GCP spending $1M+/year

   Sources:Cerebras' $35B IPO resets AI chip valuations · Anthropic just fired a shot at Figma's $20B moat · Cerebras IPO, $20B OpenAI deal, and compute-as-kingmaker · AI's pricing inflection is here · Anthropic just declared war on Figma

2. 02

   AI's Pricing Phase Transition: Hidden Cost Increases, Budget Exhaustion, and Who Builds Real Moats

   <h3>The Monetization Inflection</h3><p>The AI industry just crossed from the <strong>subsidy phase to the monetization phase</strong>, and three data points prove it. First, Anthropic has shifted large enterprise customers to <strong>consumption-based pricing</strong> — and customers are staying despite higher costs, because productivity gains justify the spend. Second, Uber's CTO disclosed that <strong>Claude Code usage maxed out the company's full-year AI budget just months into 2026</strong>. Third, Morgan Stanley reports <strong>37% of enterprises now report quantifiable AI benefits</strong>, up 23% quarter-over-quarter — the fastest adoption acceleration since cloud computing.</p><h3>The Hidden Cost Trap</h3><p>But here's what the market hasn't absorbed: Opus 4.7's new tokenizer <strong>inflates token counts by up to 35%</strong> depending on content type. List pricing is unchanged at $5/$25 per million tokens — but the same API call now costs more. Anthropic claims reasoning efficiency improvements offset this for complex tasks, but <strong>for non-reasoning workloads like document processing, it's a pure cost increase</strong>.</p><table><thead><tr><th>Processing Method</th><th>Cost per Page</th><th>Quality</th></tr></thead><tbody><tr><td>Opus 4.7 (direct)</td><td>~7¢</td><td>Charts: 55.8%, Content: 90.3%</td></tr><tr><td>LlamaIndex (agentic)</td><td>~1.25¢</td><td>Comparable on content</td></tr><tr><td>LlamaIndex (cost-effective)</td><td>~0.4¢</td><td>Lower but adequate</td></tr></tbody></table><p>This <strong>5-17x cost gap</strong> for document processing confirms that specialized middleware retains a durable economic moat even as frontier models improve. The value accrues to the orchestration layer, not the model layer.</p><h3>The Enterprise Budget Crunch</h3><p>The Uber signal is a double-edged sword. <strong>Bull case</strong>: AI budgets must expand 2-3x to accommodate demand, creating massive TAM expansion. <strong>Bear case</strong>: CFOs who didn't plan for this will impose spending freezes, creating 1-2 quarters of revenue volatility for AI vendors. Companies with consumption-based contracts where spending increases automatically are best positioned — which is precisely why Anthropic's pricing transition is strategically important.</p><p>The consumption-pricing filter now works as a <strong>diligence binary</strong>: companies that have successfully transitioned to usage-based billing without material churn have real moats. Companies still on flat-fee models either lack confidence that customers will pay for actual usage or are still in the subsidy phase. Use this as a portfolio triage signal.</p><h3>The Inference Deflation Counter-Signal</h3><p>GPT-4-level inference has collapsed from <strong>$20 to $0.40 per million tokens</strong> in 3.5 years — a 50x decline. Yet a 5-8x gap persists between naive and optimized deployments. The companies that master multi-layer inference optimization achieve structurally superior unit economics; those that don't face widening cost disadvantages. Fine-tuned 7B models now match 70B models on narrow domains, meaning vertical AI companies that self-host have <strong>10x cost advantages</strong> over competitors calling frontier APIs.</p><blockquote>The diagnostic question for every AI portfolio company at the next board meeting: 'Walk me through your inference cost per unit of customer value delivered, and how that's changed in the last two quarters.' If they can't answer precisely, their margins are on a timer.</blockquote>

   Action items

   • Audit all portfolio companies with >20% of COGS from Claude API for effective cost impact from the new tokenizer — the 35% inflation is a hidden margin squeeze that won't show until this billing cycle closes
   • Use consumption-based pricing as a binary diligence filter for all AI deal flow — companies that transitioned without churn have real moats; those on flat-fee are still in subsidy mode
   • Source 3-5 deals in AI inference optimization: semantic caching, intelligent model routing, and structured RAG preprocessing

   Sources:AI's pricing inflection is here · Anthropic reclaims SOTA with Opus 4.7 · Anthropic's IPO calculus just got harder · LLM inference costs fell 50x in 3.5 years · Three strategic shifts in frontier AI just redrew the value chain

3. 03

   Europe's Six-Week Fuel Cliff: The Macro Tail Risk Lurking Behind the Tech Re-Rating

   <h3>The Hard Numbers</h3><p>IEA Executive Director Fatih Birol has declared the Strait of Hormuz closure <strong>"the largest energy crisis we have ever faced."</strong> Europe — the biggest recipient of jet fuel transiting the strait — has roughly <strong>six weeks of supply remaining</strong>. ACI Europe warns shortages could begin in three weeks. Jet fuel costs have doubled since the Iran war began. And the recovery math is brutal: even with a deal tomorrow, Birol says it could take <strong>up to two years to normalize</strong> oil flows.</p><p>Unplanned liquid fuel outages are running at <strong>4.5x the typical month</strong> — one of the largest supply disruptions in modern history. Meanwhile, the world still consumes ~37 billion barrels annually at ~$3 trillion in value.</p><h3>The US-Europe Divergence Trade</h3><p>The competitive asymmetry is stark. The US produces most of its own jet fuel and is the <strong>world's largest net exporter</strong>. European carriers face an existential supply problem; US carriers face a manageable cost problem.</p><table><thead><tr><th>Carrier</th><th>Region</th><th>Exposure</th><th>Risk Level</th></tr></thead><tbody><tr><td>Ryanair</td><td>Europe</td><td>On track for June shortages</td><td>Critical</td></tr><tr><td>easyJet</td><td>Europe</td><td>70% summer fuel covered; 30% gap at 2x</td><td>High</td></tr><tr><td>Lufthansa</td><td>Europe</td><td>Grounding 40 planes, cutting long-haul</td><td>High</td></tr><tr><td>Delta</td><td>US</td><td>Owns refinery; domestic supply</td><td>Moderate</td></tr><tr><td>Spirit Airlines</td><td>US</td><td>Potential bankruptcy risk</td><td>Critical</td></tr></tbody></table><h3>Why Tech Investors Should Care</h3><p>The Nasdaq is on a <strong>12-day winning streak</strong> — the longest since 2009 — amid what the IEA calls the worst energy crisis in history. This divergence between equity optimism and commodity-market stress is a setup for volatility. The catalyst is a known date: <strong>when European fuel reserves hit zero</strong>.</p><p>Energy is the <strong>only sector matching tech's earnings growth trajectory</strong>, which tells you the market is pricing in sustained disruption. Your tech thesis needs a Hormuz stress test. The 94% decline in oil intensity since 1965 provides structural resilience — but a prolonged closure reprices macro risk across asset classes. The rare combination of tech earnings growth plus multiple expansion that the valuation gap implies could evaporate if energy shock cascades into demand destruction.</p><h4>The Geopolitical Trajectory</h4><p>The US-Iran ceasefire remains tenuous. VP Vance's negotiations in Pakistan failed. Defense Secretary Hegseth renewed threats to attack Iran's civilian infrastructure. An Israel-Lebanon 10-day ceasefire started April 16, but every escalation signal <strong>extends the energy crisis timeline</strong> and makes Birol's 2-year estimate look optimistic.</p><blockquote>The tech re-rating thesis only works if the macro doesn't break first — and Europe's fuel clock is the most clearly defined catalyst for a macro break on the calendar right now.</blockquote>

   Action items

   • Stress-test all portfolio exposure to European aviation, travel, and tourism against a 2-year energy normalization scenario — not a 2-month recovery
   • Evaluate long positions in US refining/midstream assets as a macro hedge for tech-heavy portfolios
   • Add a Hormuz scenario to every tech portfolio company's macro stress test

   Sources:Europe has 6 weeks of jet fuel left · Tech P/E compressed to 2018 levels while earnings growth hits 43%