OpenAI shut down Sora. The official reason is deepfakes; the real reason is the bill.

OpenAI announced Sora's discontinuation on March 24. As of April 26, the Sora web and app experiences are no longer available. The Sora API will follow on September 24 [OpenAI help].

── What shipped (or rather, didn't) ──

OpenAI's stated reason: deepfake risk during a global election year. Despite C2PA metadata and digital watermarking, bad actors found ways to bypass safeguards [OpenAI announcement].

The unstated reasons, sourced from internal reporting:

• Sora was burning roughly $1 million per day in compute costs. Video generation is materially more expensive than text or image inference.
• Usage declined from a peak of ~1M monthly active users to <500K by the time the shutdown was announced.

── Why it matters ──

Sora was the most capable text-to-video model commercially available. Its shutdown removes the most visible AI-video product from the market and concedes the segment to Runway, Pika, Google's Veo, and a long tail of smaller players.

For builders, two practical effects:

• API-dependent products are exposed. Anyone who built on the Sora API has six months to migrate. The replacements are credible but not drop-in compatible.
• Compute economics are shaping product decisions. OpenAI shutting down a product over compute burn is a public admission that not every modality is unit-economical at consumer scale. Expect similar shutdowns from other vendors as the optimism premium burns off.

── Editor's take ──

The deepfake framing is convenient, but the math is the real story. A $1M/day burn rate on declining usage is a textbook product-market-fit failure. OpenAI saying the quiet part out loud — that some AI products don't work as businesses — sets a useful precedent. Not every modality scales. The next decade will sort which ones do.