The Margin Squeeze
The Margin Squeeze
The most successful AI companies are discovering an uncomfortable truth: growth and profitability are pulling in opposite directions. OpenAI’s gross margin jumped to 70% in October, up from 52% at year-end 2024, signaling a sharp inflection where inference costs have become manageable and pricing power has materialized. Yet this margin expansion arrives at a moment when the entire industry faces mounting pressure from regulatory friction, geopolitical competition, and the brutal economics of the hardware race. The prosperity of the software layer depends entirely on keeping the cost of compute down, but that same pressure is squeezing hardware companies, robotics startups, and infrastructure plays into insolvency. The tech industry is rationalizing itself in real time: winners are consolidating margin while losers exit the game.
Deep Dive
OpenAI’s Margin Expansion Reveals the AI Profit Model’s Real Winners
OpenAI’s jump from 52% to 70% gross margin in ten months is not incremental progress, it is a structural break. The company has crossed from a growth-at-all-costs model into something resembling sustainable economics. This happens when three things align: inference efficiency improves (better models, better implementations), customer willingness to pay hardens (competitive moats form), and utilization scales (fixed costs spread across more transactions). For OpenAI, all three are now in motion.
The implication is stark: the software layer of AI is becoming a margin business, but only for companies that achieve scale and differentiation simultaneously. Smaller competitors face a squeeze. If OpenAI can achieve 70% gross margins while pricing aggressively to capture market share, what happens to Claude, Gemini, or open-source alternatives trying to compete on the same terms? This margin expansion is not a sign of health across the industry, it is a sign that the industry is bifurcating. Winners take margin. Everyone else fights for volume at razor-thin returns.
What matters next is whether this margin persists as a source of reinvestment (funding the next generation of models) or becomes extractive (dividends and shareholder returns). The answer determines whether OpenAI looks like Microsoft circa 2000 (incumbent defending its turf) or a genuinely transformative platform. If margins start compressing again in 2026 because competition forces price cuts or inference costs stop improving, the entire thesis inverts.
Robotaxi Fragmentation Will Determine the Next Phase of Autonomous Transport
Uber’s decision to trial Baidu’s Apollo Go RT6 robotaxis in London starting H1 2026 is significant not for what it reveals about Uber’s robotaxi strategy, but for what it reveals about the global AI and robotics market. Uber is not building robotaxis from scratch; it is licensing technology from the Chinese market leader and shipping it to the West. This is the opposite of the narrative tech companies have been selling for years. It suggests that the frontier of robotaxi capability may have shifted from Silicon Valley to Beijing.
Baidu has been operating robotaxis in Chinese cities at scale since 2023, accumulating millions of miles of autonomous driving data in dense urban environments. Apollo Go competes with local alternatives, not monopoly players. This competitive crucible may have produced better outcomes than the slower, more legally constrained trials happening in San Francisco or Phoenix. If Uber’s London deployment succeeds, it validates a global deployment model where Chinese robotaxi technology becomes the infrastructure layer for Western mobility companies.
The second-order implication is regulatory. London is not California. UK regulators may have different thresholds for autonomous deployment than US regulators, and Baidu’s success in a more permissive regulatory environment could accelerate global adoption. But there is also fragmentation risk: if Baidu powers Uber in London, Waymo controls San Francisco, and Chinese competitors dominate Asia, the world ends up with incompatible stacks, higher switching costs for cities and operators, and reduced innovation pressure. What looks like global robotaxi adoption might actually be regional balkanization.
The CHIPS Act’s Retreat Signals Weakness in US Industrial Policy
The Department of Commerce terminating its $285 million contract with the SMART USA Institute is not a budget cut, it is an admission of strategy failure. The SMART Institute was supposed to develop digital twins for chipmaking, a foundational capability for scaling US semiconductor manufacturing. Digital twins compress the feedback loop between design, process optimization, and yield improvement. Without them, US fabs run hotter, slower, and less efficiently than competitors using advanced simulation.
The termination reveals two problems. First, the Trump administration is cleansing government spending of initiatives it views as ideological or inefficient, and SMART USA looks vulnerable because its mission is diffuse and its deliverables are hard to measure in quarterly terms. Second, and more consequential, is the implication that US industrial policy lacks the staying power to compete with long-term commitments from other governments. If a five-year contract gets terminated mid-stream, manufacturers cannot rely on government backing for the decade-long infrastructure plays that semiconductor leadership requires.
This matters because digital twins are not a US advantage that can be rushed. China and Taiwan are building equivalent capabilities, and the race is for early-mover advantage in the integration of AI-driven simulation into fab operations. Losing institutional momentum at SMART USA may cost the US lead in this specific capability for years. The broader signal: the CHIPS Act was supposed to anchor US semiconductor independence through sustained investment, but it faces political fragmentation. What seemed like bipartisan consensus in 2022 is becoming contested territory in 2026.
Signal Shots
China’s Open AI Models Reach Parity with Western Competitors — Chinese AI models like DeepSeek and others have closed the capability gap with US benchmarks, and are increasingly available as open-weight models. This matters because adoption follows accessibility, and free or cheap Chinese models will undercut proprietary Western alternatives in price-sensitive markets. What to watch: whether Western regulatory barriers (export controls, sanctions) can slow adoption faster than Chinese ecosystem lock-in can accelerate it.
Spotify’s Metadata Breach Exposes Data Scraping at Scale — Activist group Anna’s Archive released 86 million music files and 256M rows of Spotify metadata in 300TB of torrents. This is not a database breach; it is authorized metadata extraction at an unprecedented scale. This matters because it shows that even companies with strong security can be undermined by public APIs and public data, and that the scraping economy for training AI models is unstoppable. What to watch: whether music labels use this as leverage to renegotiate streaming deals or licensing terms with both Spotify and AI companies.
iRobot’s Bankruptcy Reveals Hardware Economics Are Broken — iRobot filed Chapter 11 after Amazon’s acquisition was blocked by FTC pressure, a company that spent 30 years building a category now cannot survive standalone competition. This matters because it signals that hardware margins cannot compete with software, and that regulatory blocks on acquisitions destroy more value than they protect. What to watch: whether other hardware startups (robotics, consumer hardware) accelerate M&A or pivots to software-as-a-service models.
CISA Instability Compounds Cybersecurity Leadership Crisis — DHS is investigating whether CISA staffers misled the acting director into a failed polygraph, symbolizing broader dysfunction at the nation’s cybersecurity agency. This matters because CISA is supposed to coordinate critical infrastructure defense, but internal chaos suggests the organization cannot execute at scale. What to watch: whether this leads to leadership turnover that either stabilizes or further fractures the agency.
Echo’s $35M Series A Signals AI-Native Security Startups Are Fundable — Echo raised $35M to automate container vulnerability elimination at build time using AI agents, a model that bypasses traditional security scanning. This matters because it shows investors are willing to fund AI-first approaches to infrastructure problems, even if the underlying problem is not new. What to watch: whether Echo’s approach scales across different infrastructure domains or remains a niche play in containerization.
South Korea Mandates Facial Recognition for SIM Purchases — South Korea’s government announced facial scanning requirements for mobile SIM activation following massive data breaches. This matters because it represents a government choosing biometric identity verification as the solution to information security failure, a choice that trades privacy for fraud reduction. What to watch: whether other countries with large data breaches follow the same path, and whether this accelerates biometric infrastructure globally.
Scanning the Wire
SoftBank Scrambles to Close OpenAI Funding Commitment — Reuters reports SoftBank is racing to raise capital to close its $22.5B OpenAI commitment before year-end, signaling either confidence in the return or pressure from investors to deliver on promises.
Xbox Console Sales Fall Behind Original Nintendo Switch — CNBC analysis shows Xbox Series S/X sales are trailing the 2017 Nintendo Switch, evidence that Microsoft’s game console business has strategically retreated in favor of cloud and subscription models.
AI Cited in Over 50,000 Job Cuts in 2025 — CNBC reports Amazon, Microsoft, and other major tech firms explicitly attributed AI-driven restructuring to 50,000+ layoffs, the first year where AI is directly named as a cause.
Waymo Robotaxis Stalled During San Francisco Blackout — Waymo suspended service when a widespread power outage left many robotaxis unable to navigate, exposing dependency on grid-connected infrastructure and real-time cloud services.
Permira and Warburg Pincus Acquire Clearwater Analytics for $8.4B — PE consortium agrees to buy financial software maker Clearwater Analytics, signaling PE appetite for enterprise SaaS despite elevated valuations.
Apple and Google Allow Alternative App Stores in Japan — Apple and Google began offering third-party app store distribution in Japan under regulatory pressure, a significant crack in their closed ecosystem strategy that may spread globally.
Biren Technology Files for $623M Hong Kong IPO — Chinese GPU maker Biren plans Hong Kong listing, one of China’s “Four Little Dragons” in GPU design seeking capital for international expansion and AI chip competition.
David Sacks’ EO Push on AI Regulation Fractures Tech Lobby — Tech lobbyists report Trump’s AI czar is pushing executive action to block state AI laws, undercutting efforts by some companies to negotiate permanent federal frameworks instead of fragmented state rules.
Shield AI Valued at $5.6B Under New CEO Gary Steele — Military tech startup Shield AI aims to grow annual revenue from \(300M to \)1B by 2028, signaling confident growth plans in autonomous defense systems amid Ukraine conflict and rising defense spending.
Outlier
Workers Should Control AI Agents, Says WorkBeaver CEO — One of the few voices questioning the default assumption that AI agents exist to automate workers away is the founder of WorkBeaver, who argues that platforms should empower workers to control their own agentic tools rather than serving as replacements. This is a signal that the co-agent model (humans and AI systems collaborating at parity) may be gaining traction as an alternative to full automation. If this framing catches on with labor unions or regulators, it could reshape the entire narrative around AI adoption in knowledge work, turning what looks like displacement into something more like augmentation with worker agency. Worth watching as a contrarian bet against the automation-first consensus.
See you in the next one. We’re halfway through the turn, and the winners are already separating from the pack.