Steven Sinofsky on Microsoft's antitrust era, Apple's AI talent problem, and parallels to the dot-com boom

Summary

Steven Sinofsky, former head of Windows and Office at Microsoft and now a board partner at a16z, argues that the Microsoft antitrust era—spanning roughly 1990 to 2007—is consistently overstated in its day-to-day operational impact. The legal and regulatory machinery moves too slowly to be a constant disruption, he says, describing months of intense work on the EU-mandated browser choice dialogue punctuated by Brussels going dark for 12-week summer recesses. His sharper critique is that the antitrust pressure distorted Microsoft's internal judgment, causing the company to over-index on Windows as the center of gravity for every product decision—a self-inflicted strategic wound more damaging than any consent decree.

Dot-Com Parallels to the AI Moment

Sinofsky pushes back on the idea that dot-com era entrepreneurs were simply wrong. Every major prediction from 1993 onward—grocery delivery, digital music, same-day logistics—ultimately proved correct. The error was timing, not thesis. Webvan was dismissed by the founder of FedEx as operationally impossible, yet Amazon now runs the model with roughly 1.5 million employees. The pattern repeated across every category: mp3.com preceded Spotify, pets.com preceded Chewy (now a public company worth over $10 billion). The dot-com dead were crawlers that enabled today's walkers.

The critical structural difference between then and now is the denominator. In 1996, the addressable base was roughly 100 million computers globally, with only around 30% of U.S. households owning one and no mobile internet. OpenAI reaching $10 billion in revenue is less surprising when the install base includes every iPhone, every laptop, every app store account, and frictionless payments via Stripe. ChatGPT Pro at $200 per month—roughly $2,400 per year—lands in the same spend bracket as a $3,000 Dell PC of the 1990s, but with three clicks and zero cost of sale. Dell's revenue ramp from $5 billion to $25 billion compounded over years; Cursor and OpenAI are moving faster against a larger base.

Sinofsky's caution is that many current AI companies are still crawlers. Citing Andrej Karpathy's YC talk, he notes the framing of "the decade of the agent" is analytically honest but commercially useless—founders build assuming the revolution is already underway and they are already behind, which is the correct psychological posture even if the technology is early.

Apple's AI Talent Problem

Apple faces a structural compensation mismatch in the AI talent market. Tim Cook's total comp in 2024 was approximately $74.6 million as CEO of a $3 trillion company, yet top-tier AI researchers can command signing packages at that scale or beyond at competitors like Meta on day one. Sinofsky draws a direct parallel to Microsoft's talent wars with Borland International over programming language engineers in the early 1990s—a battle that ended in lawsuits over departures and was fought at six-figure comp levels that look trivial today.

He identifies two compounding risks. First, non-founder CEOs are more constrained in authorizing outlier compensation because they must manage ripple effects across existing pay bands and retention of adjacent talent. Second, adverse selection operates in both directions—departures that look catastrophic from the outside may reflect internal performance problems, while genuine losses of key people can be quietly devastating. Apple Intelligence is a genuinely new discipline for a company whose historical advantage is hardware-software integration and UI polish, not frontier model research, which makes the talent gap structurally harder to close.

Regulatory Competition and the Siri Button

Sinofsky's most pointed observation concerns the competitive dynamics among regulators themselves. During the Microsoft antitrust proceedings, the U.S. DOJ and FTC effectively set global terms, with European regulators largely deferring. That hierarchy has inverted. The EU, operating without meaningful electoral accountability, discovered that aggressive regulation of U.S. tech companies is a reputational product—a gold medal competition with Washington. The result is a coordination dynamic that feels, in Sinofsky's description, close to collusion without technically being so, with the EU, UK, and increasingly Asian regulators triangulating their actions against each other's moves as much as against the companies themselves.

On the specific question of AI assistant invocation—whether regulators will force Apple to allow third-party models like ChatGPT to be triggered by the Siri button—Sinofsky sees an internal contradiction in the EU's position. The Digital Markets Act mandates openness and choice while simultaneously waving at security and privacy protections without reconciling the two. He describes the operative language as literally stating that "security should not be compromised" while requiring sideloading and third-party stores, calling the tension unresolvable by regulation. His view is that the market has already resolved it: iOS and Android represent two coherent, competing philosophies of user trust, and forcing convergence reduces rather than expands meaningful consumer choice. The browser default wars at Microsoft, he argues, produced ballot screens that real users ignored entirely while tech enthusiasts obsessed over file-type-level granularity that nobody outside a Discord server actually wanted.