Daily Newsletter Digest

Why AI productivity gains aren't translating to company profits

May 31st 2026

Today's News

Equity analysts systematically underestimate AI-driven earnings revisions, invalidating consensus forecasts

Azeem Azhar, Exponential View

Micron's FY2026 EPS consensus revised from $18.25 in December 2025 to $58 in May 2026, while Google's median analyst view grew 40% year-over-year. JP Morgan consensus forecasts show capex and R&D spending as a share of revenues will crash 25–40% in 2027, a prediction that assumes either imminent AI revenue slowdown, major technical breakthrough, or economic shock. These aren't one-time misses but systematic failures to model regime change during exponential technology adoption, similar to decades of IEA underestimation of solar cost declines. The pattern reveals analyst blindness to how hyperscaler capex growth (20% annually through 2030) vastly outpaces revenue growth (15%), creating potential shareholder value destruction if the underlying demand assumptions prove wrong.

SoFi launches bank-issued stablecoin to 14.7M retail customers, marking first US bank stablecoin in consumer app

The Defiant

SoFi Technologies made SoFiUSD available inside its banking app for 14.7 million members, enabling users to buy, sell, hold, and convert the token directly within the app. SoFiUSD is backed 1:1 by cash held at the Federal Reserve and issued by SoFi Bank (FDIC-insured), running on Ethereum and Solana with additional networks planned. No U.S. national bank has previously offered a publicly redeemable stablecoin inside a consumer banking app; JPMorgan's JPM Coin is restricted to institutional clients, and Société Générale's stablecoin is issued from France. The launch follows SoFi's March partnership with Mastercard to use SoFiUSD as a settlement currency across Mastercard's global payments network, with SoFi's Galileo platform (160+ million accounts) expected to offer issuing banks the option to settle card transactions using the stablecoin. SoFi posted $1.1 billion in adjusted net revenue in Q1 2026, up 41% year-over-year, with membership up 35% year-over-year.

ICE chairman calls Hyperliquid 'bigger than Nasdaq' as institutional capital flows into decentralized perpetuals exchange

The Defiant

Jeff Sprecher, founder and CEO of Intercontinental Exchange (owner of NYSE), called decentralized exchange Hyperliquid "bigger than Nasdaq" at the Bernstein Strategic Decisions Conference, praising the team as "extremely smart" and noting the platform's wealth creation with a small 11-person team. Hyperliquid is the largest decentralized perpetual futures exchange with $5.9 billion in total value locked, running its order book entirely onchain on its own Layer 1 network. HYPE token has gained 70% over the past 30 days and touched an all-time high of $67, reaching the 11th-largest cryptocurrency by market value at $15 billion, driven partly by US spot ETF inflows exceeding $100 million in combined net inflows within the first 10 trading sessions. Sprecher highlighted Hyperliquid's synthetic SpaceX perpetual market as "really interesting to watch," suggesting the onchain price discovery "could be bigger than the IPO" when SpaceX lists. Despite the praise, Sprecher flagged regulatory concerns, questioning why traditional exchanges face restrictions that crypto-native rivals do not, and noting the risk of up-to-100x leverage offered on the platform.

Worth Reading

Why AI isn't showing up on your bottom line

Exponential View

I read the full essay. Main takeaways: - Individual AI productivity gains (Claude Code users producing more PRs/code) are not translating to firm-level ROI; only 27% of executives say AI has met expectations despite widespread adoption - AI adoption mirrors Paul David's three-stage electrification model: Stage 1 (lightbulb—individual task speedup), Stage 2 (group drive—workflow optimization on existing processes), Stage 3 (unit drive—firm-level decision speed and organizational restructuring) - Most deployed AI today is Stage 1-2; firms experience 'congestion' when individual/team output accelerates but decision-making layers remain unchanged, creating backlogs rather than ROI - Stage 3 requires AI to make autonomous decisions without human intermediaries (e.g., feature requests flowing directly from customer support agent → codebase decision → implementation in hours, not weeks) - The bottleneck is managerial oversight and sign-off cycles, not execution speed; firms must rebuild decision architecture around workflow speed, not individual task speed

Read more

Articles Read