The AI budget trap | PitchBook News Summary

1. Key Themes

Theme 1: The AI Budget Trap — Corporate Spending Is Not Delivering on Promised ROI

Despite massive and growing AI budgets, measurable productivity gains remain elusive for most corporate adopters, creating a feedback loop of spending without returns.

"40% of the companies that tracked their spending recorded cost savings of less than 10% from their AI initiatives. Still, 90% of those surveyed whose AI investments underdelivered plan to increase their AI budgets next year."

"In reality, it is a circular bet with a structural leak." — Bain & Company report authors

Real-world examples compound the concern: Amazon shut down an internal AI activity leaderboard after finding employees were gaming it with unnecessary bots, and Uber "blown through its entire 2026 AI budget in the first four months."

Theme 2: Investor Scrutiny of AI is Hardening — ROI Has Replaced Roadmaps as the Dominant Narrative

Wall Street analysts are shifting from celebratory to interrogative on AI, signaling a maturation in how public markets assess AI-exposed companies.

"A PitchBook analysis of 186 AI-related questions on B2C earnings calls found that 90% were probing and investigative in tone, and that ROI and financial returns have overtaken product roadmaps as the dominant focus."

This shift creates meaningful risk for AI application startups that cannot demonstrate clear unit economics, as churn risk rises from enterprise customers reassessing spend.

Theme 3: Private Credit's Hidden Stress — EBITDA Is Masking a Cash Flow Crisis

A Kroll Bond Rating Agency analysis of 2,400+ sponsor-backed middle-market borrowers reveals that strong reported earnings are increasingly disconnected from actual liquidity.

"For a typical company, EBITDA has grown at an impressive rate of 27% over the last two years through Q1 2026, yet cash flow from business operations rose at an anemic 8% over the same period."

"About 48% of borrowers in the dataset reported negative operating cash flow, a hallmark of stressed credits. That share was 39% about two years ago."

The median EBITDA-based interest coverage ratio sat at 1.6x, but the cash-flow-based ratio collapses to just 0.3x — meaning the typical borrower has almost nothing left after debt service.

Theme 4: Defense Tech and European Cybersecurity Are Attracting Record Early-Stage Capital

Two sectors are drawing outsized and accelerating VC attention driven by structural tailwinds: geopolitical tensions (defense tech) and AI-native threat surfaces (cybersecurity).

"Record capital is flooding a sector riding geopolitical tailwinds and a $839 billion US defense budget." (defense tech)

"VCs backing European cybersecurity are writing bigger checks for younger companies — $1 billion landed in Q1 alone, with AI-native startups hoovering up early-stage rounds that are now outpacing late-stage deal values globally."

Theme 5: AI Infrastructure Investment Continues at Sovereign Scale

Despite enterprise ROI concerns at the application layer, infrastructure-level AI commitment from major players — including sovereign wealth and global conglomerates — shows no signs of slowing.

"SoftBank Group will commit up to €75 billion to develop 5 gigawatts of AI data center capacity in France."

"Anthropic just filed its S-1, chasing a nearly $1 trillion valuation. The growth is real and the profits are arriving, but our AI Quality Gap research shows the highest-valued AI companies carry the widest gap between price and business quality."

2. Contrarian Perspectives

Perspective 1: Continued AI Budget Increases Despite Consistent Underdelivery Are Irrational — and Structurally Dangerous

The consensus view is that AI investment is justified by future potential. The contrarian read: the data shows a self-reinforcing spending loop with no corrective mechanism. Bain found that 90% of executives whose AI investments underdelivered still plan to increase budgets — not reduce or reallocate them. This isn't optimism; it's institutional inertia. Bain's own authors labeled it plainly:

"In reality, it is a circular bet with a structural leak."

The Amazon and Uber examples illustrate how AI spend can generate negative operational value (gaming rankings, blowing annual budgets in four months), yet neither triggered spending pullback — only internal corrections.

Perspective 2: EBITDA as a Credit Metric Is Dangerously Misleading in the Current Environment

The conventional wisdom is that EBITDA growth signals health in private credit portfolios. The KBRA data suggests the opposite: EBITDA is flattering borrower quality while actual cash generation deteriorates sharply.

"The median ratio of operating cash flow to EBITDA dropped to 21% from 33% two years ago."

"The median EBITDA-based interest coverage ratio sat at 1.6x as of Q1, while the same ratio based on operating cash flow is a median of just 0.3x."

Shane Olaleye of KBRA noted: "As of right now, for these smaller companies, it helps watching their actual usage of cash, which might be an earlier signal that something might go wrong." For investors in private credit funds, reported EBITDA growth of 27% may be actively obscuring a portfolio under significant stress.

Perspective 3: AI Augmentation, Not Replacement, May Be the Higher-Value Enterprise Strategy

Amid widespread narratives about AI replacing workers, at least one major industrial company is betting on the opposite approach.

"Businesses stand to gain more by using AI to augment workers than by using it to replace them, Schneider Electric wants to prove, even as companies are laying off thousands of employees."

This is a minority position in corporate AI strategy today but is supported by the broader evidence that replacement-oriented AI deployments are producing low savings (under 10% for 40% of companies tracked) while augmentation models remain underexplored as a ROI strategy.

3. Companies Identified

Anthropic Description: AI lab / large language model developer Why mentioned: Filed S-1 targeting ~$1 trillion valuation; cited as a case study of the "AI Quality Gap" — high valuation with a wide spread between price and business quality

"Anthropic just filed its S-1, chasing a nearly $1 trillion valuation. The growth is real and the profits are arriving, but our AI Quality Gap research shows the highest-valued AI companies carry the widest gap between price and business quality."

Anysphere (Cursor) Description: VC-backed AI coding agent startup Why mentioned: Cited as evidence that AI application-layer growth is real even amid broader enterprise skepticism

"Anysphere, the VC-backed company behind coding agent Cursor, hit $3 billion in annualized revenue in April, driven by enterprise demand from software development teams."

Amazon Description: Global technology and e-commerce conglomerate Why mentioned: Cited as a case study in AI budget mismanagement and employee gaming of AI activity metrics

"Amazon shut down an internal leaderboard tracking employees' AI activity after the company found staffers were running unnecessary autonomous bots to climb the rankings."

Uber Description: Ride-sharing and logistics platform Why mentioned: Illustrates the speed at which enterprise AI budgets are being consumed without commensurate returns

"Uber said in May that it had blown through its entire 2026 AI budget in the first four months."

Schneider Electric Description: Global energy management and automation company Why mentioned: Cited as a counterexample pursuing AI augmentation over replacement as a value-creation strategy

"Businesses stand to gain more by using AI to augment workers than by using it to replace them, Schneider Electric wants to prove."

Runway Description: New York-based AI lab focused on generative video/media Why mentioned: Announced European HQ in London with a $100M+ UK AI ecosystem commitment; backed by notable investors including Nvidia and Qatar Investment Authority

"Runway...announced a new European headquarters in London with plans to invest more than $100 million in the UK AI ecosystem over the next 18 months."

Mecka AI Description: Startup collecting human motion and gesture data for physical AI Why mentioned: Raised both a $25M Series A and $35M follow-on, both led by Framework Ventures — a signal of high conviction in physical AI training data

"A startup collecting data on human motions and gestures for use in physical AI, raised a $25 million Series A and a $35 million follow-on investment."

Radionor Communications Description: Norwegian tactical broadband radio maker Why mentioned: Exploring a $3B–$4.5B sale, fielding interest from both buyout firms and defense groups — a defense tech investment signal

"A Norwegian tactical broadband radio maker, is exploring a sale at a valuation of $3 billion to $4.5 billion, as the company fields interest from both buyout firms and defense groups."

Liftoff Mobile Description: Marketing platform for mobile app developers; Blackstone-backed Why mentioned: Targeting up to $3.66B valuation in IPO — a notable PE-to-public exit in the martech space

"Blackstone-backed Liftoff Mobile...is targeting a valuation of up to $3.66 billion in its IPO."

Gigascale Description: Climate-tech focused VC firm founded by former Meta CTO Mike Schroepfer Why mentioned: Raised $250M for its latest fund — notable given founder's pedigree and the climate-tech investment thesis

"Former Meta CTO Mike Schroepfer's VC firm Gigascale raised $250 million for its latest fund focused on climate-tech investments."

SoftBank Group Description: Japanese multinational technology and investment conglomerate Why mentioned: Committing up to €75B for AI data center capacity in France — a landmark scale of sovereign-adjacent AI infrastructure investment

"SoftBank Group will commit up to €75 billion to develop 5 gigawatts of AI data center capacity in France."

Phosphorus / Dragos Description: Phosphorus is a cybersecurity startup (IoT/OT security); Dragos is an industrial cybersecurity firm Why mentioned: Acquisition signals consolidation in the OT/IoT security segment

"Cybersecurity startup Phosphorus...agreed to be acquired by Dragos."

Encosa Description: Munich-based battery storage startup Why mentioned: Raised a €25M seed round — notable seed size in the energy storage space

"Munich-based battery storage startup Encosa raised a €25 million seed investment led by Realyze Ventures."

Sekai Description: Developer of a platform to create small apps using text prompts Why mentioned: Raised $20M Series A led by Khosla Ventures and Connect Ventures — signals continued VC conviction in no-code/AI-native app creation

"The developer of a platform to create small apps using text prompts, secured a $20 million Series A."

ERock Description: Houston-based modular power system developer; backed by Energy Impact Partners Why mentioned: Seeking up to $642M IPO — a significant exit in the energy infrastructure space

"Houston-based ERock, a modular power system developer backed by Energy Impact Partners, is seeking to raise up to $642 million in its IPO."

4. People Identified

Shane Olaleye Description: Head of corporate credit assessment at Kroll Bond Rating Agency (KBRA) Why mentioned: Provided expert commentary on the divergence between EBITDA and operating cash flow as an early warning signal in private credit

"I still think EBITDA [works] as a proxy for cash flow over a long period of time. But as of right now, for these smaller companies, it helps watching their actual usage of cash, which might be an earlier signal that something might go wrong."

Mike Schroepfer Description: Former CTO of Meta; founder of Gigascale VC Why mentioned: His climate-tech VC firm raised $250M for its latest fund, drawing on his high-profile tech leadership background

"Former Meta CTO Mike Schroepfer's VC firm Gigascale raised $250 million for its latest fund focused on climate-tech investments."

Sam Altman Description: CEO of OpenAI Why mentioned: Subject of a new lawsuit filed by the Florida attorney general alleging OpenAI ignored safety warnings

"Once again, Sam Altman is getting sued. This time, the Florida attorney general filed a lawsuit alleging that OpenAI ignored safety warnings and put the state's residents at risk."

Rosie Bradbury Description: Senior Venture Capital Reporter at PitchBook News Why mentioned: Author of the AI budget trap lead story

Madeline Shi Description: Senior Private Equity Reporter at PitchBook News Why mentioned: Author of the private credit EBITDA story

5. Operating Insights

Insight 1: Track Operating Cash Flow, Not Just EBITDA, When Evaluating Portfolio Company Health

The KBRA data is a direct operational warning for PE/credit investors and operators managing leveraged businesses. At a median cash-flow-based interest coverage of 0.3x, companies have nearly no buffer for reinvestment or unexpected shocks. Operators need to stress-test liquidity independently of EBITDA.

"The median EBITDA-based interest coverage ratio sat at 1.6x as of Q1, while the same ratio based on operating cash flow is a median of just 0.3x. At that level, the typical borrower has almost no cash left after servicing its debt, leaving little buffer against an unforeseen shock."

Insight 2: AI Application Startups Face a Two-Sided Risk — Churn from Existing Customers and Harder New Customer Acquisition

As corporates reassess AI ROI, startups selling AI tools into enterprise will face increasing headwinds on both retention and new growth. The strategic implication: focus on demonstrable, measurable outcomes — not just deployment — and build churn defenses now.

"AI tools underperforming on productivity while dragging on costs represent a one-two punch for many AI application startups: They risk high churn from existing corporate users and may find recruiting new customers increasingly difficult."

Insight 3: AI-Generated "Workslop" Is a Reputational and Operational Risk Worth Naming and Managing

A new category of workplace inefficiency has been formally identified by BetterUp Labs and Stanford's Social Media Lab: low-quality, AI-generated content that looks polished but lacks substance. For operators deploying AI across teams, this is a quality control risk that requires governance, not just deployment.

"It's also created a new genre of workplace inefficiency, which research by BetterUp Labs and Stanford's Social Media Lab dubbed 'workslop': low-quality content that looks slick, but has little substance."

6. Overlooked Insights

Insight 1: Physical AI Training Data Is Emerging as a Distinct, Investable Category

Mecka AI's dual fundraise — a $25M Series A and a $35M follow-on, both led by the same investor — signals very high conviction in a niche that doesn't get the same attention as language model training data: human motion and gesture capture for physical AI applications (robotics, embodied AI). This is a data infrastructure play, not a model play, and is easy to overlook in the broader AI noise.

"Mecka AI, a startup collecting data on human motions and gestures for use in physical AI, raised a $25 million Series A and a $35 million follow-on investment, both led by Framework Ventures."

Insight 2: Rare Earth Mining's Environmental and Reputational Liability Is a Latent Risk for AI/Defense Supply Chains

Briefly mentioned in Side Letters but significant: the rare earth metals required for AI hardware and defense technology are generating environmental and public health crises in mining regions — with potential downstream regulatory, ESG, and supply chain implications for investors across these sectors.

"Mining for industry-critical rare earth metals is releasing potentially radioactive waste. 'It's killing our children and killing everything around the mine,' a fisherman from Madagascar said."