Library/Coinbase Cuts AI Spend by 50% | Kalshi's $40B Valuation & Impending IPO | The Year for SaaS Roll-Ups

20VC with Harry Stebbings

Coinbase Cuts AI Spend by 50% | Kalshi's $40B Valuation & Impending IPO | The Year for SaaS Roll-Ups

1h 18mJul 2, 2026
Key Themes
AI ROI pressureopen-source competitionAI policymarket structureprediction marketsSaaS roll-upsenterprise AIIPO dynamics
Summary

AI budgets tighten, frontier-model politics intensify, and software roll-ups get a fresh look

This episode traces a broad shift in the AI and software landscape: companies are moving from experimental AI spending to hard ROI scrutiny, frontier-model vendors are facing more competition and policy risk, and legacy software roll-ups are resurfacing as a possible strategy in an AI-transformed market. The conversation also covers Microsoft’s AI positioning, the rise of prediction markets, IPO timing for major private companies, and whether Claude inside Slack could become a meaningful enterprise wedge.

1
AI spend is shifting from enthusiasm to proof

A central thread of the episode is that companies are no longer satisfied with broad AI experimentation. They want measurable savings, revenue lift, or operational impact before increasing budgets, which changes the economics for vendors and buyers alike.

2
Open-source and geopolitical concerns are becoming intertwined

The discussion goes beyond model quality and into policy, with the speakers suggesting that debates over distillation, Chinese models, and enterprise security could shape access to AI systems. That makes the competition landscape as much political as technical.

3
Market leadership in AI now depends on more than exposure

Microsoft is used as the example of why owning an AI partner is not the same as having a differentiated AI story. The episode suggests that scale alone is no longer enough if growth is slowing or the company does not control a proprietary, compelling product layer.

4
Prediction markets may be moving from novelty to category

Kalshi, Polymarket, and related products are discussed as part of a broader market that could be large if it ties into culturally familiar behaviors like sports and financial betting. The episode treats that category as one where product-market fit may matter more than ideological novelty.

5
Legacy software can still be valuable if the customer base is sticky

The Bending Spoons discussion shows that aging products with recurring revenue can remain attractive if they have durable users and room for operational improvement. But the episode also stresses that cost-cutting alone is rarely enough in today’s market; reinvention and leadership matter.

6
Enterprise AI may enter through workflow context, not standalone tools

Claude in Slack is presented as interesting because it sits where work already happens and can observe organizational context in real time. The hosts think that context capture could be strategically important, even if the immediate disruption risk is still uncertain.

7
Venture markets are demanding sharper fundraising standards

The final segment argues that the bar for Series A has risen and that founders need to understand where their growth profile sits relative to scarce capital and better alternatives. The episode favors candor, but also acknowledges that tone matters when delivering hard truths.

Select any chapter text to Deep Dive with AI
01AI Spend Cuts, ROI Pressure, and Anthropic vs. Open Source

The episode opens with Coinbase’s reported 50% AI spend cut despite continued usage, which becomes a proxy for a broader shift from exuberant experimentation to disciplined AI ROI scrutiny. The speakers question whether reduced spending signals weaker economics for frontier model vendors or simply better cost management, and they critique companies that advertise AI adoption without clear business outcomes. The chapter closes with a sharp discussion of Anthropic’s complaint about Chinese model distillation, framed as a mix of IP conflict, legal ambiguity, and geopolitical tension.

Coinbase’s AI spend reportedly fell by 50% while usage stayed high.
The discussion frames AI adoption as entering a more disciplined ROI phase.
The speakers criticize performative AI messaging without measurable business impact.
They distinguish between cost savings and true revenue or productivity lift.
Anthropic’s claims about Chinese models are discussed as an IP and geopolitical issue.
The conversation suggests distillation disputes could escalate beyond contract law.
02Anthropic vs. Chinese Open Source Ban, then Microsoft and AI Market Structure

The conversation broadens from model theft into the politics of AI regulation, with the speakers weighing whether Anthropic could be pushing U.S. restrictions on Chinese open-source models. They distinguish between punishing alleged distillation and broadly limiting open-source competition, then argue that policy may be shaped as much by regulatory capture and national-security framing as by technical nuance. The segment ends with a market-structure debate about whether the AI economy resembles an oligopoly that governments will protect to preserve strategic advantage and high prices.

Anthropic is described as potentially lobbying for restrictions on Chinese open-source models.
The speakers separate distillation/IP theft from national-security concerns.
They argue a broad open-source ban is unlikely, but narrower restrictions are plausible.
The discussion frames this as possible regulatory capture by frontier labs.
Enterprises may be nudged away from foreign open-source systems even without an explicit ban.
The chapter ends with a debate over AI market structure, competition, and price erosion.
03Microsoft AI Position, Kalshi, SpaceX IPO Volatility, and Bending Spoons

The hosts criticize Microsoft’s AI posture, arguing that ownership of OpenAI does not substitute for a truly differentiated frontier-model story, especially as Azure growth slows. The discussion then turns to Kalshi’s rapid valuation expansion and the larger prediction-market category, before shifting to how volatility around a future SpaceX IPO could influence sentiment for other private-market listings. The chapter closes by contrasting AI-native businesses with Bending Spoons’ roll-up model, which still commands a strong valuation despite being rooted in mature software assets.

Microsoft is seen as lacking a true standalone frontier-model story.
Azure growth deceleration is treated as a warning sign.
Kalshi’s valuation jump signals strong interest in prediction markets.
The hosts see sports and financial betting as the biggest TAMs.
SpaceX IPO volatility could influence Anthropic/OpenAI listing timing.
Bending Spoons is presented as an 'anti-AI' IPO with a premium valuation.
04Bending Spoons B2B Roll-Up Strategy and Chamath’s CEO Move

This chapter explores whether a Bending Spoons-style acquisition playbook can be extended across B2B software, especially for aging but sticky products with room for operational cleanup. The speakers argue that future winners will likely need more than cost cuts and pricing optimization; they may need to rebuild products with AI-driven features and sharper leadership. The segment then shifts to Chamath Palihapitiya’s decision to raise capital and serve as CEO of an AI startup, which the panel respects for ambition but questions as a sustainable operating model.

Bending Spoons is treated as a template for acquiring and fixing software assets.
Sticky customer bases make declining software businesses appealing targets.
AI may be required to create new value, not just cut costs.
Roll-up success depends heavily on leadership quality and execution.
Chamath Palihapitiya’s CEO move is seen as ambitious but hard to sustain.
The panel doubts VCs can match the grind of full-time founder-CEOs.
05VC Honesty, Series A Reality, and Claude in Slack

The final chapter debates Harry’s blunt tweet about turning down a founder whose revenue growth profile no longer clears today’s bar for a strong Series A. The panel agrees the underlying point matches the current venture environment, but they wrestle with the line between useful honesty and unnecessarily abrasive communication. The conversation then shifts to Claude inside Slack, which is treated as both an intriguing enterprise wedge and a potential Trojan horse for broader workflow capture, even as the hosts expect incumbent platforms to respond quickly.

Series A standards are perceived to be higher in the AI era.
The panel debates blunt founder feedback versus tactful communication.
Capital efficiency and opportunity cost matter more than before.
Slow early growth can still lead to large outcomes.
Claude in Slack is framed as a possible enterprise wedge.
Incumbent platforms like Salesforce and Slack are expected to react quickly.