Bean counting for AI
1. Key Themes
Theme 1: AI Cost Management is a Growing but Likely Unsustainable Standalone Business
Enterprise AI spend has become painful enough that Fortune 10 CEOs are calling startups directly for relief, spawning a wave of "tokenomics" optimization companies — but experienced investors doubt this is a durable standalone category.
"Every company wants to understand and put metrics around spend, and it's just gonna be a normal piece of finance. ... Do I think [there's] gonna be like a $10 billion company that helps companies manage AI spend? I don't." — Joanne Chen, GP at Foundation Capital
"Atlas Cloud is part of a growing category of companies all selling the allure of managing runway 'tokenomics' to businesses rattled by massive AI spending. But managing such spending may not be a durable business to invest in on its own."
Theme 2: AI Spend is Maturing from Experimentation to ROI Discipline
The market is moving past promiscuous token consumption toward rigorous measurement of returns — a structural shift in how enterprises justify AI budgets.
"What's changed, according to Chen, is a shift from 'please use more tokens' experimentation to instead trying to define and measure the ROI from that spending."
Theme 3: Tech-Focused PE Fundraising is Sharply Bifurcated
Capital is concentrating among proven tech PE managers; the broader category is in freefall.
"Francisco Partners' haul comes as the overall pace of capital allocation to technology-oriented strategies has cooled. Tech-focused PE funds collected $26.24 billion through June, well off the pace of the $97 billion raised last year."
"The $75 billion firm amassed $21 billion across both its flagship vehicle, Francisco Partners VIII, and Francisco Partners Agility IV."
Theme 4: VC Deal Valuations Have Surpassed 2021 Peak Across Every Stage
Median deal values have not just recovered — they have blown past prior highs at every series, signaling a new valuation regime driven largely by AI.
"Median VC deal values have blown past 2021 highs across every series, with pre-seed, seed and Series D+ more than doubling. Series C, A and B are up 95%, 89.9% and 83.3%, respectively."
Theme 5: Defense and National Security AI is Attracting Tier-1 VC Capital at Scale
The most prominent VC firms are writing large checks into military-adjacent cyber and AI startups, signaling a durable investment theme beyond commercial AI.
"Cathedral, a cyber operations startup focused on US military capabilities, raised a $160 million round led by Andreessen Horowitz and Sequoia Capital at a $1.4 billion valuation."
2. Contrarian Perspectives
Perspective 1: AI Cost-Optimization Startups Have No Historical Precedent for Building a Breakout Company
While the pain is real and demand exists today, the venture community's bet against this category is grounded in a decade of failed analogues in infrastructure cost-cutting — not just theoretical skepticism.
"'In the last 10 years, there really was no breakout success,' said Theory Ventures' GP Tomasz Tunguz of prior waves of infrastructure cost-cutting startups. 'A company showing up saying we will help you optimize your cost is probably not the right manifestation.'"
The pattern of consolidation reinforces this: "Stripe and ClickHouse acquired startups specializing in this," suggesting acqui-hire outcomes rather than independent scaled businesses.
Perspective 2: AI Disruption of Incumbent Software is Real but Adoption Will Be Slow — Incumbents Have Time
While the consensus narrative is that AI will rapidly upend software incumbents, Francisco Partners is betting the opposite — that in key verticals, incumbents have meaningful runway to adapt before disruption hits.
"'Don't confuse the pace of innovation vs. the pace of adoption,' Brown said."
"The firm sees AI as less immediately disruptive in some of its target sectors, which include healthcare IT, education technology and fintech. While incumbent software providers will inevitably have to adapt, they have time to fully take advantage of what AI can offer."
Perspective 3: Blackstone Sees No Oversupply Risk in AI Infrastructure Returns — Near Term
Despite widespread concern about data center and AI infrastructure overcapacity, Blackstone's COO is publicly dismissing the risk to near-term returns.
"Blackstone does not expect growing competition to drive down returns on its new investments in AI infrastructure, at least not anytime soon, said COO Jon Gray."
3. Companies Identified
- Description: AI inference startup focused on image and video generation
- Why mentioned: Lead case study in the AI cost-optimization trend; fielding inbound calls from Fortune 10 CEOs seeking 10–20% AI bill reductions
- Quote: "Atlas Cloud is part of a growing category of companies all selling the allure of managing runway 'tokenomics' to businesses rattled by massive AI spending."
- Description: Cyber operations startup focused on US military capabilities
- Why mentioned: Raised $160M at a $1.4B valuation led by a16z and Sequoia — the largest deal in the newsletter and a signal of tier-1 VC conviction in defense tech
- Quote: "Cathedral, a cyber operations startup focused on US military capabilities, raised a $160 million round led by Andreessen Horowitz and Sequoia Capital at a $1.4 billion valuation."
Francisco Partners
- Description: $75B technology-focused private equity firm
- Why mentioned: Raised $21B across two funds despite a broader tech PE fundraising freeze; offers a contrarian view on AI adoption pace in healthcare IT, edtech, and fintech
- Quote: "Francisco Partners' $21B haul defies tech PE fundraising chill."
- Description: Corporate spend management platform
- Why mentioned: Cited as an example of AI cost-optimization becoming a native feature within broader platforms, rather than a standalone product — a structural threat to point-solution startups in this space
- Quote: "Ramp, which helps companies manage their spending, announced access to its router tool to help clients save money by routing requests to different models."
- Description: AI agents for phishing defense
- Why mentioned: Raised $36M Series A led by Battery Ventures; notable as a purpose-built AI security company at Series A scale
- Quote: "AegisAI, which builds AI agents to defend against phishing attacks, raised a $36 million Series A led by Battery Ventures."
- Description: AI-native creative design platform (San Francisco)
- Why mentioned: Raised $34M Series A led by Accel and Iconiq; represents continued top-tier investor conviction in AI-native creative tools
- Quote: "Paper, the developer of an AI-native creative design platform, raised a $34 million Series A led by Accel and Iconiq."
- Description: Data infrastructure for physical AI and robotics (Singapore)
- Why mentioned: Raised $30M pre-A — notable for being in the physical AI/robotics data layer, an emerging infrastructure category
- Quote: "Ropedia, which builds data infrastructure for physical AI and robotics, raised $30 million in pre-A funding."
- Description: AI-native machine sensors
- Why mentioned: Raised $21M led by Innovation Endeavors and Xora; sits at the intersection of physical AI and industrial IoT
- Quote: "Elio, which builds AI-native machine sensors, raised a $21 million round led by Innovation Endeavors and Xora."
Mubadala Capital
- Description: Sovereign wealth-backed investment firm
- Why mentioned: Tokenizing its evergreen private markets fund on blockchain in partnership with Coinbase and KAIO — a notable institutional step toward on-chain private markets access
- Quote: "Mubadala Capital teamed with Coinbase and KAIO to tokenize its evergreen private markets fund, giving investors access to a blockchain-native version of the vehicle."
Partners Group
- Description: Global private markets asset manager
- Why mentioned: Closed $5.5B in infrastructure secondaries; articulated a value-creation-over-discount investment philosophy
- Quote: "Partners Group closed $5.5 billion in infrastructure secondaries — its case: Returns come from what happens after you buy, not the discount you obtained to get in."
4. People Identified
Jerry Tang
- Description: CEO of Atlas Cloud
- Why mentioned: Primary industry source illustrating the intensity of enterprise AI cost pressure — fielding calls from Fortune 10 CEOs personally
- Quote: "Tang, whose startup Atlas Cloud handles AI inference for image and video generation, said calls from prospective customers, looking to cut their bills by 10% or 20%, are coming in at a steady clip these days."
Joanne Chen
- Description: General Partner, Foundation Capital
- Why mentioned: Provided the most direct bear case on AI cost-management as a standalone investment category, while also articulating the broader ROI-measurement shift
- Quote: "Do I think [there's] gonna be like a $10 billion company that helps companies manage AI spend? I don't."
Tomasz Tunguz
- Description: General Partner, Theory Ventures
- Why mentioned: Offered historical evidence against the durability of infrastructure cost-cutting as a standalone venture category
- Quote: "'In the last 10 years, there really was no breakout success,' said Theory Ventures' GP Tomasz Tunguz of prior waves of infrastructure cost-cutting startups."
Andrew Brown
- Description: Global Head of Fundraising and Marketing, Francisco Partners
- Why mentioned: Articulated the firm's differentiated thesis on the gap between AI innovation speed and enterprise adoption speed — the strategic rationale for betting on incumbent software in healthcare IT, edtech, and fintech
- Quote: "'Don't confuse the pace of innovation vs. the pace of adoption,' Brown said."
Jon Gray
- Description: COO, Blackstone
- Why mentioned: Publicly dismissed AI infrastructure oversupply concerns as a near-term threat to returns — a high-stakes conviction statement from the world's largest alternative asset manager
- Quote: "Blackstone does not expect growing competition to drive down returns on its new investments in AI infrastructure, at least not anytime soon, said COO Jon Gray."
5. Operating Insights
1. Build AI Cost Optimization as a Feature, Not a Company The investor consensus is that standalone AI spend management tools will be absorbed into broader platforms. Operators should integrate cost-routing and ROI measurement natively rather than relying on point solutions — and founders in this space should consider whether their product is better positioned as a feature acquisition target than a standalone business.
"Some companies are even adding cost-optimization features natively. Ramp...announced access to its router tool to help clients save money by routing requests to different models." "Stripe and ClickHouse acquired startups specializing in this."
2. Reframe AI Investment Conversations Around ROI, Not Usage Enterprise buyers have moved past the exploration phase. Sales and product leaders should lead with measurable ROI frameworks rather than token consumption metrics — the language of the budget conversation has changed.
"What's changed, according to Chen, is a shift from 'please use more tokens' experimentation to instead trying to define and measure the ROI from that spending."
6. Overlooked Insights
1. Midwest Local Governments Are Extracting Concessions from Data Center Developers A briefly noted item suggests a new dynamic in data center site selection: communities are gaining negotiating leverage over tech giants, with implications for site costs, permitting timelines, and community relations strategies for infrastructure developers.
"Several local governments across the Midwest are in talks with tech giants to invest back into schools, healthcare, teachers, housing and more."
2. Japan's Interest Rate Hits 1% for the First Time Since 1995 — a Generation Has No Memory of What Follows Mentioned only in passing, this macro development could have significant ripple effects on global capital flows, carry trades, and LP behavior — particularly for funds with Japanese institutional investors.
"Japan's benchmark interest rate has reached 1% for the first time since 1995, and an entire generation of business leaders, bankers and households has no memory of what comes next."