The business of GPU preorders
- 01🖥️ Theme 1: GPU Compute Scarcity Is Restructuring the Neocloud Market
- 02🏦 Theme 2: Neocloud Debt Is Safer Than the Market Believes
- 03🌍 Theme 3: Sovereign AI & Geographic Diversification of AI Infrastructure
- 04🌱 Theme 4: Africa VC Is Maturing Beyond Its Payments Roots
- 05🛡️ Theme 5: Defense Tech VC Is Surging, Even Into Unprofitable Companies
1. Key Themes
🖥️ Theme 1: GPU Compute Scarcity Is Restructuring the Neocloud Market
The shortage of GPU compute is so acute that neocloud providers like TensorWave are running eBay-style auctions for future access, fundamentally shifting market power toward suppliers.
"It's so scarce that whoever wants to pay the most is gonna get it." — Piotr Tomasik, co-founder and COO, TensorWave
Contract durations are lengthening dramatically as a direct result, with multi-year deals becoming the new norm.
"If you had asked for a six-year contract just two years ago, people would laugh you out of the room, but that's not uncommon nowadays." — Kevin Cochrane, CMO, Vultr
Implication for investors: Neocloud providers with locked-in long-term contracts are accreting structural revenue advantages. Smaller AI startups, meanwhile, face a worsening compute access problem.
🏦 Theme 2: Neocloud Debt Is Safer Than the Market Believes
The mainstream narrative frames neocloud debt facilities as dangerously leveraged, but a contrarian cohort of investors argues the signed-contract backlogs make these loans structurally sound.
"You had abnormally high growth, abnormally high customer concentration, and abnormally high debt. That was the cocktail investors had to navigate through." — Antolin Garza, partner, Karmel Capital
"The market has persistently underweighted the value of the signed-contract backlogs." — Antolin Garza, Karmel Capital
Even in default scenarios, the contracts themselves are structured to protect lenders:
"The terms by which one of these contracts can be cancelable are super small—like force majeure. Even if a neocloud fails, most of these contracts are being written in ways to still protect the lenders." — Piotr Tomasik, TensorWave
🌍 Theme 3: Sovereign AI & Geographic Diversification of AI Infrastructure
Governments are actively building national AI capabilities, signaling a broader geopolitical fragmentation of AI infrastructure investment.
"The UK launched a $675M sovereign AI fund to reduce its dependence on technology from other countries."
Implication: This creates a durable, government-backed investment theme in AI infrastructure outside the US, particularly in Europe and emerging markets.
🌱 Theme 4: Africa VC Is Maturing Beyond Its Payments Roots
Africa's VC market is broadening its sectoral base, suggesting the ecosystem is entering a new phase of maturation that may attract more generalist global capital.
"Africa's VC market has started the year with a bang as investment spreads across more sectors—signaling the market's maturation beyond its payments industry roots."
🛡️ Theme 5: Defense Tech VC Is Surging, Even Into Unprofitable Companies
Venture capital continues to pour into defense tech regardless of profitability, with last quarter registering near-record deal volumes.
"VCs haven't been shy about investing in unprofitable defense tech companies. Last quarter saw the second-highest VC deal total on record in the industry, and much of which went to companies spending heavily."
2. Contrarian Perspectives
Neocloud Debt Is a Mispriced Opportunity, Not a Systemic Risk
The consensus treats neocloud leverage as a red flag. A small group of investors, backed by direct portfolio exposure, argues the market has systematically mispriced the value of forward-contract backlogs as collateral. The combination of GPU scarcity and near-unbreakable contract terms (cancelable only under force majeure) means lenders are better protected than commonly understood.
"The market has persistently underweighted the value of the signed-contract backlogs." — Antolin Garza, Karmel Capital
"The terms by which one of these contracts can be cancelable are super small—like force majeure. Even if a neocloud fails, most of these contracts are being written in ways to still protect the lenders." — Piotr Tomasik, TensorWave
Evidence: Karmel Capital has deployed this thesis via investments in CoreWeave, Lambda Labs, and Crusoe.
LPAC Membership Is an Underappreciated Due Diligence Edge in Secondaries
The industry treats LP Advisory Committee (LPAC) membership primarily as a governance mechanism. Pantheon's experience suggests it is actually a significant information asymmetry tool in secondary transactions — providing diligence depth that dramatically exceeds standard LP reporting.
"This GP is well known for the fact that if you're an LP in their funds, you basically get only a paragraph [on how they're performing], on each company, every quarter. But if you're on the LPAC, you get pages per company." — Amyn Hassanally, Pantheon
OpenAI Is Now a Life Sciences Competitor, Not Just an AI Lab
The assumption that frontier AI labs would stay in their lane is being challenged directly. OpenAI's GPT-Rosalind release into drug discovery is a direct threat to specialized life sciences AI players.
"OpenAI aims at drug discovery with GPT-Rosalind, a new frontier model unveiled yesterday. The release challenges the assumption that life sciences are too specialized for AI labs to explore."
3. Companies Identified
| Company | Description | Why Mentioned | Key Quote |
|---|---|---|---|
| TensorWave | AI neocloud startup selling GPU compute access | Case study of GPU scarcity dynamics and auction-based preorder model | "It's so scarce that whoever wants to pay the most is gonna get it." |
| Vultr | GPU cloud compute provider | Illustrates market-wide shift toward longer-term compute contracts | "If you had asked for a six-year contract just two years ago, people would laugh you out of the room." |
| CoreWeave | Neocloud operator | Cited as a Karmel Capital portfolio investment validating the debt thesis | Referenced alongside Lambda Labs and Crusoe as investable neoclouds |
| Lambda Labs | Neocloud operator | Karmel Capital portfolio company | Referenced as part of the contract-backed debt thesis |
| Crusoe | Neocloud operator | Karmel Capital portfolio company | Referenced as part of the contract-backed debt thesis |
| OpenAI | AI frontier lab | Entering drug discovery with GPT-Rosalind, threatening specialized AI life sciences players | "The release challenges the assumption that life sciences are too specialized for AI labs to explore." |
| Plata | Mexico-based fintech | Raised $405M Series C at $5B valuation — signals large emerging market fintech appetite | Series C led by Bicycle Capital |
| Glydways | Autonomous urban transit startup | $170M Series C for autonomous urban transportation networks | Led by Suzuki Motor Corp, ACS Group, and Khosla Ventures |
| X-Energy | Small modular nuclear reactor company | Targeting $7.5B IPO valuation; backed by Amazon and ARK Investment Management | IPO in progress |
| Wayve | London-based autonomous driving startup | Raised $60M from AMD, Arm, and Qualcomm Ventures — notable chip-company backing | Semiconductor strategic investors signal conviction in AV |
| Resolve AI | Software production operations startup | $40M Series A extension at $1.5B valuation; led by DST Global and Salesforce Ventures | Rapid valuation milestone |
| Accel | Major VC firm | Raised $5B across latest funds, including $4B late-stage vehicle | Signals continued mega-fund appetite for late-stage tech |
| Pantheon | PE secondaries firm, active since 1980s | Case study on secondaries deal flow, LPAC advantages, and quality filtering in current market | "2026 has gotten off to an even busier start than 2025." |
| Mazama Energy | Geothermal energy startup | Raising $100M round — signals continued investor interest in alternative energy infrastructure | Reported by Axios |
| Hypercard | Expense management / employee credit cards | Backed by Sam Altman; agreed to be acquired by American Express | Notable exit to a major financial incumbent |
| Spektr | Copenhagen-based AI compliance agent for financial services | $20M Series A led by NEA | AI agents entering regulated financial services |
4. People Identified
| Person | Description | Why Mentioned | Key Quote |
|---|---|---|---|
| Piotr Tomasik | Co-founder and COO, TensorWave | Primary source on GPU scarcity, contract structure, and neocloud debt safety | "The terms by which one of these contracts can be cancelable are super small—like force majeure." |
| Antolin Garza | Partner, Karmel Capital | Contrarian voice on neocloud debt being underpriced; direct investor in CoreWeave, Lambda, Crusoe | "The market has persistently underweighted the value of the signed-contract backlogs." |
| Kevin Cochrane | CMO, Vultr | Confirms industry-wide shift to longer GPU compute contracts | "If you had asked for a six-year contract just two years ago, people would laugh you out of the room." |
| Amyn Hassanally | Global Head of PE Secondaries, Pantheon | Expert on secondaries deal flow quality, LPAC information advantages, and continuation fund conflicts | "If you're on the LPAC, you get pages per company." |
| Dimitri Zabelin | Senior Analyst, PitchBook | Published defining report on sovereign AI following UK's $675M fund announcement | Cited as author of sovereign AI report |
| Steve Feinberg | Former CEO, Cerberus Capital Management | Now applying PE portfolio management discipline to Pentagon defense contractors | Referenced via WSJ: treating defense contractors "like portfolio companies of the Pentagon" |
| Sam Altman | CEO, OpenAI; backer of Hypercard | Noted as backer of Hypercard, which was acquired by American Express | Notable investor/exit connection |
5. Operating Insights
For Neocloud Operators: Prioritize Long-Term Contracts as Both Revenue and Collateral Strategy
Lenders are now underwriting neocloud debt against signed-contract backlogs rather than physical chip assets. Operators who secure longer-duration contracts not only stabilize revenue but unlock better debt terms and lower perceived risk. Actively steering customers toward multi-year deals is now a capital strategy, not just a sales preference.
"The company would pick a customer seeking a longer contract over a shorter one." — Kevin Cochrane, CMO, Vultr
For Secondary Investors: LPAC Seats Are a Structural Due Diligence Advantage Worth Pursuing
In secondary transactions, LPAC membership translates into qualitatively superior portfolio intelligence — potentially the difference between a paragraph and pages of company-level data per quarter. Investors structuring secondary programs should prioritize LPAC access as a core part of their underwriting infrastructure.
"It gives us incredible insight into the assets we're buying... if you're on the LPAC, you get pages per company." — Amyn Hassanally, Pantheon
For Secondary Market Participants: Selectivity Is the Edge When Deal Flow Overwhelms Capital
The 2026 secondaries market has more supply than capital to absorb it. The operational discipline is not sourcing — it is filtering. Pantheon explicitly rates managers as "A" or "B" and concentrates on portfolios dominated by "A" managers.
"We're being extremely selective about what we're leaning into because there's much more deal flow than there is capital to absorb it." — Amyn Hassanally, Pantheon
6. Overlooked Insights
Smaller AI Startups Are Being Structurally Squeezed Out of Compute Access
While the GPU scarcity narrative focuses on upside for neocloud providers, the article briefly surfaces a significant downstream consequence: smaller startups are being priced and prioritized out of compute access as providers favor larger, longer-term customers. This dynamic could widen the competitive moat for well-capitalized AI incumbents and create a structural barrier to entry for early-stage AI companies.
"Unfortunately, these dynamics are not favorable for smaller startups looking to access compute power from these providers."
2017 Vintage Global Funds-of-Funds: Low DPI Despite Decent IRR
The Daily Benchmark data shows 2017 vintage global funds-of-funds posting a median IRR of 14.32% but a median DPI of only 0.47x — meaning LPs have received back less than half their invested capital nearly a decade in. This is a meaningful liquidity signal for LPs evaluating the realized return profile of fund-of-funds structures in a prolonged exit drought.
"Median IRR: 14.32% | Median DPI: 0.47x" — PitchBook Daily Benchmark, 2017 Vintage Global Funds-of-Funds