Everyone Is Counting GPUs. The Real Chokepoint Is a Laser Die
The AI scaling race is quietly being decided not by GPUs but by Indium Phosphide laser dies, a III-V crystal with 15-50% yields that no amount of capital can rush.
9 min read
Latest · June 21, 2026
Everyone's Watching Training Move. Inference Can't Follow.
xAI built a 100,000-GPU cluster in Memphis in 122 days because the city had power and could deliver it fast. That choice marks a permanent split in AI compute into two machines that obey different laws of physics.
9 min read
Read article
Because these models run on our edge network, inference happens close to your users, which reduces latency.
95% of AI Pilots Show No P&L Impact. Trust, Not Models, Is Why
Generating an AI answer now costs fractions of a cent. Trusting it still costs human labor. That gap decides who captures value in enterprise AI.
9 min read
From the archive
The Credit Gate: When Power Scarcity Became a Balance-Sheet FilterPower scarcity has become a credit filter in the AI buildout. Landlords with energized land now demand investment-grade balance sheets, so cash-rich neoclouds end up holding GPUs they can't turn on.
Why the Revolt Against Nvidia Made Broadcom the Real Toll CollectorEvery hyperscaler tried to escape Nvidia's pricing power by building custom chips. They didn't tear down the toll road, they split it in two and handed the bigger booth to Broadcom.
OpenAI Isn't Building an Empire. It's a Panicked Tenant.Every AI giant owns an OS, a device, or a browser, except OpenAI. That single gap reframes its product blitz as a scramble for distribution it doesn't own.
Google Zero: What the Open Web Loses When AI Answers Replace the ClickPublishers blocking AI crawlers think they're losing. They're actually pulling the foundation out from under the AI industry and triggering a feedback loop that ends in model collapse.
How Pure Storage Won a Hyperscaler by Selling Back MegawattsA top-four cloud provider just swapped its hard drives for Pure Storage's proprietary DirectFlash. The trigger wasn't cheap flash; it was AI running the data center out of power. Here's why that should terrify Seagate.
Probabilistic COGS: Why AI ARR Is Disguised ConsultingMuch of what AI startups report as recurring revenue is consulting work in disguise, and when the forward deployed engineers leave, the ARR leaves with them.
Second-Order Effects of AI Agents Becoming Software's Primary UsersWhen AI agents become software's primary users, seat-based pricing collapses, engagement metrics invert, and B2B marketing shifts from persuading humans to optimizing for machine comprehension.
The Interconnection Moat: Why Energy Is AI's New Barrier to EntryThe grid interconnection queue has replaced chip allocation as AI's binding constraint. Microsoft, Google, and Amazon are becoming energy companies because whoever controls the electrons controls the frontier.
The Supercharger Gambit: How Tesla Traded a Hardware Moat for Platform DominanceTesla's decision to open its Supercharger network wasn't generosity, it was a calculated platform play that transformed a hardware moat into an ecosystem that extracts value from every competitor vehicle on the road.
The Day Shopify Admitted Software Can't Eat EverythingShopify's sale of its logistics business to Flexport reveals the hard boundaries where software companies meet physical reality, and the second-order effects reshaping e-commerce strategy.
Toxic Virality: When Your Best Growth Asset Starts Billing YouMidjourney once shut off its own free trials to survive its own virality. That inversion explains why freemium is dying, and who gets to inherit it.
Netflix Called Everyone's Bluff, And Rewrote the Rules of StreamingNetflix's password crackdown delivered record sign-ups and proved its utility status, but the second-order effects on Gen Z subscription habits, competitor viability, and streaming's business model may reshape the industry for a decade.Contact
Found a mistake, or have something to share?
Email admin@secondorderlabs.com. We read what comes in and reply when we can.