Robots don't usually fail because they don't work. Poor system integration and mismatched ROI expectations cause more than half of deployment failures. Inadequate change management adds another ~20%. The tech works. The deployment doesn't. That gap is a software problem. @navya_prab maps where physical AI's second wave is opening up: interplay.vc/blog/physical-a…

Main Street is planning to hire like it's 2020. The S&P is earning like it's 2021. Both are true right now. In May, the NFIB Small Business Optimism Index fell to 95.3, its ninth straight month below its 52-year average. Buried inside: a seasonally adjusted net 9% of small-business owners plan to add jobs in the next three months — down four points from April, and the lowest reading since May 2020. Two months later, the top-line labor number caught up. June nonfarm payrolls came in at just 57,000 — roughly half the 115,000 consensus, with downward revisions to the prior two months on top. Meanwhile, the other economy is having its best year in half a decade. S&P 500 companies are on pace for their strongest quarterly earnings growth since 2021, concentrated in the AI-adjacent names that drive capex and revenue simultaneously. → Small-business hiring plans: net 9% (lowest since 2020) → S&P 500 Q1 2026 earnings growth: strongest since 2021 The takeaway isn't that Main Street is broken. It's that the two economies are running on different fuel. Big-cap earnings are being pulled forward by an AI capex cycle that concentrates spending in a narrow set of buyers and sellers. Small-business hiring is being throttled by pricing pressure, labor costs, and demand uncertainty that doesn't show up in an index dominated by ten names. Any thesis that ignores which economy a portfolio company sells into is a thesis running on the wrong data. Full analysis → interplay.vc/blog/two-econom… #SmallBusiness #NFIB #Economy #VentureCapital #AI #Interplay

The manufacturing story on your feed says two things: factories are humming, or factories are dying. Neither is what the federal data actually shows. Employment is up 333,000 workers since 2015. It's also down 305,000 since January 2023. Output is 2% below its 2015 baseline. It's also up 3.4% off its October 2024 bottom. Both facts, same series. The 2015–19 productivity gap is closing right now: output per hour is up 3.9% since early 2023, on a workforce that's shrinking. And it's closing in specific segments the aggregate hides. High-tech equipment output is up 11.1% year over year. Consumer goods is down 1.2%. The average tells you nothing. With capacity utilization 2.5 points below its long-run norm, the industrial-tech pitch that sells "more units per hour" is selling into a factory that already has room to run harder. What buyers are actually paying for right now is different. We pulled the federal data. Full read: interplay.vc/blog/manufactur…

OpenAI took the first formal step last month toward the most anticipated tech listing since the dot-com era. It sits on an $852B private mark from its March round and has signaled ambitions to price the IPO north of $1 trillion. The valuation is the "easy" part — justifying it isn't. Set the valuation against the fundamentals and you get the defining tension of the filing: → Revenue is genuinely extraordinary — roughly $25B annualized as of early 2026, up from ~$20B at the end of 2025. → But the same business runs a 33% gross margin and is on track to burn ~$27B of cash in 2026, rising toward ~$63B in 2027. Internal forecasts pointed to a $14B loss in 2026 alone. That gap — between fast-climbing revenue and even faster-climbing cash burn — is what needs to be justified. A deck can frame a widening loss as investment in a category-defining platform — Amazon did exactly that. But an S-1 is a legal document. It has to reconcile an $852B mark with disclosed losses, a compute-cost structure that scales with usage, and partner economics — including Microsoft's revenue share — that most public investors will read closely for the first time. For founders, the lesson isn't "don't burn capital." It's that the story you tell in a deck and the story an S-1 forces you to tell diverge sharply the moment losses have to be quantified in a table. OpenAI is about to run the highest-stakes version of that exercise in public. What the market decides the unpriceable is worth will reset the comp set for every AI company behind it. Full analysis → interplay.vc/blog/pricing-th… #OpenAI #IPO #VentureCapital #AI #AIEconomy #TechIPO

America just stopped building offices for people. In April 2026, US data-center construction ran at a $50.7 billion annualized rate, passing general office construction at $43.8 billion for the first time on record. This is the cleanest single-chart refutation of the "services-led growth" narrative we have seen. The next investment dollar is no longer funding a software-company office in Austin. It's funding a hyperscale data-center campus (the giant cloud-compute halls run by Amazon, Microsoft, Google, and Meta) in northern Virginia, Phoenix, or central Texas — and the entire industrial supply chain underneath: electrical switchgear, transformers, liquid cooling, specialized cleanroom build-outs, concrete, copper, steel. Two things are converging. Raw computing power is now the binding constraint on every frontier tech company. And the office category itself is shrinking in real terms after a generation of remote/hybrid work and a wave of conversions of mid-tier ("Class-B") offices into apartments. The compute curve is doubling every five years; the office curve has flatlined. The opportunity for industrial and AI-infrastructure founders is wider than the headline. Anything that touches the data-center supply chain — grid interconnection, electrical equipment, liquid cooling, prefab construction, on-site power, fiber, security, system commissioning — sits on a multi-decade tailwind that does not depend on AI models continuing to get better. Even if the AI software layer becomes a commodity, the compute infrastructure has been built and someone has to run it. The $50.7B number being quoted is in the spotlight, but the real insight is the crossover. Read more here: interplay.vc/blog/america-st… #industrialAI #datacenter #venturecapital #infrastructure

The Semiconductor Industry Association (SIA) reports March 2026 chip sales hit $99.5B — up 79.2% YoY. Q1 totaled $298.5B. The industry is on track for the first $1T year in semiconductor history, ~2 years ahead of the most aggressive 2024 forecasts. The number matters less than the shape of the curve. For 40 years, semis were textbook cyclical: 18–24 month up-cycles (PCs, mobile, cloud), then 12–18 month corrections. Investors learned to underwrite the cycle, not the trend. The 2024–2026 data shows the cycle didn't correct. It accelerated. March's +79% follows a +28% comp in March 2025 — this isn't a base-effect bounce. One end-market explains nearly all of it: AI infrastructure. HBM is sold out through 2027. The constraint is no longer demand — it's fab capacity, packaging, and power. For VC, the takeaway isn't about chip companies themselves. It's that the cost of running AI — training and inference — is falling faster than any prior compute shift, because the underlying chip supply is compounding at 70%+ annually. That cost decline flows downstream to every company building on top of it. In the cloud era, infrastructure costs flattened within a few years; here, they're still dropping. Anything built on this base inherits that. The first trillion-dollar year isn't the story. The story is that it no longer looks cyclical. interplay.vc/blog/the-chip-i…

PitchBook-NVCA's Q1'26 data confirms what every seed investor has felt for six months: the seed round is no longer a category. It's two. The headline — a $3M median — looks unremarkable. The distribution under it isn't. It's bimodal. Peak 1: the classic $2–4M seed. Capital-efficient, team-and-thesis, 18–24 months of runway. Nothing new there. Peak 2: the $20M+ mega-seed. AI-native, pedigreed founders, functionally a Series A wearing a seed label. Different game. The valley in between — the old $7–12M "priced seed" — is where the worst 2026 rounds live. Too big to be scrappy. Too small to fund the AI play. In my experience, that's where down rounds incubate. So the right question for a founder raising now isn't how big the seed should be. It's which seed market they're in. There is no longer a default answer. Full piece: interplay.vc/blog/two-seed-m…

US manufacturing output is rising. Headcount is not. Q1 2026 BLS: factory productivity +3.6% — 5x the rest of the economy. Output up, hours down. That's the AI/automation story showing up in government data. interplay.vc/blog/manufactur…

"VC is back" is a misread. Q1 2026 hit an all-time record — and 43% of it was one round (OpenAI, $122B). Strip out four AI labs and the other 5,996 startups split what was left. There are two venture markets right now. Most founders live in one of them. interplay.vc/blog/two-ventur…