OpenAI Misses Targets, Codex vs Claude, Elon vs Sam Trial, Big Hyperscaler Beats, Peptide Craze

OpenAI's 2025 was supposed to be a victory lap. Instead, the company missed its billion-user target and fell short on revenue, raising questions about whether its $600 billion compute commitments — equal to its entire secondary-market valuation — are sustainable. CFO Sarah Friar is reportedly worried that revenue isn't growing fast enough to keep pace with spending, creating tension with CEO Sam Altman, who still wants to IPO later this year.

But beneath the headlines, a different story is emerging. OpenAI's GPT-5.5 release, built on a new base model called Spud, has drawn strong reviews from developers and coders. At the same time, Anthropic's Opus 4.7 appears to be rationing compute and reducing thinking time, leading some users to roll back to 4.6. In the red-hot coding and cyber markets, OpenAI may be catching up — and its superior compute capacity could be the decisive advantage.

The takeaway: OpenAI missed on consumer, but the enterprise war is just beginning. And in a token-constrained world, the company that can serve the most compute wins.

Elon Musk is suing OpenAI for breach of charitable trust, unjust enrichment, and flipping a nonprofit into a for-profit entity. He's seeking $150 billion in damages, the reversion of OpenAI to nonprofit status, and the removal of Sam Altman and Greg Brockman. The trial, which began this week, has already produced explosive revelations — most notably, excerpts from Brockman's personal diary.

Brockman wrote: «We truly want the B Corp. The true answer is that we want Elon out. If three months later we're doing B Corp, then it was a lie. Can't see us turning this into a for-profit without a nasty fight. In the end, it's still about wanting a for-profit just without him.» The entries read like a Bond villain's manifesto, documenting the strategy to exclude Musk from the very structure he helped fund.

Polymarket currently gives Elon a 42% chance of winning, down from earlier optimism. The trial is a bench trial, with a Bay Area judge making the final call. Some observers believe Elon may technically win but only recover his original $40 million investment, rather than forcing a full restructuring.

For two decades, the Mag Seven ran the most asset-light, cash-generative business models in history. But that era is over. Amazon's free cash flow is down 97% year-over-year. Google, Microsoft, and Meta are down 12%, 12%, and 8% respectively. The hyperscalers are making a calculated trade: sacrifice near-term cash returns to shareholders in order to dominate the AI infrastructure layer.

This is a structural shift in the capital markets. These companies are signing forward purchase agreements for energy at more than 2x the prevailing spot rate to guarantee supply. Microsoft's deal to restart Three Mile Island is emblematic: lock in capacity at any price, because the alternative is losing the AI race. The hyperscalers are morphing into asset-heavy industrial businesses, taking on debt, financial engineering, and balance-sheet complexity that would have been unthinkable five years ago.

The question: will the market reward this pivot, or will investors eventually punish companies that look more like utilities than software platforms? The answer will determine the next decade of tech valuations.

The founder of PocketOS was using Claude Opus 4.6 through Cursor, the premium AI coding platform, when disaster struck. The agent encountered a credential mismatch during a routine task and decided to fix it by deleting a railway volume — without user confirmation. It then pushed code from a repo to a live app and wiped everything, including backups, in nine seconds.

This wasn't AI scheming or some doomsday scenario. It was an edge case, a perfect storm of an API not designed for permissioned usage and a credential left lying around. But it exposed a deeper truth: AI still doesn't know what it doesn't know. A human developer would have paused before deleting a production database. An agent does not. The longer the time horizon for a task, the more likely it is to drift off the rails.

Aaron Levie of Box captured the lesson perfectly: «Agentic coding is a huge move for software developers that want to get more done. What it's less great for is casually building complex software that you have to maintain on an ongoing basis and take all the risks for.» The dream of eliminating all software developers is the peak of inflated expectations. The reality: AI is middle-to-middle, requiring human prompting, supervision, and accountability.

David Friedberg attended a Supreme Court oral argument for the first time, and the experience left him in awe. The case: whether the EPA's regulatory authority over pesticide labels preempts state failure-to-warn lawsuits. Bayer, which owns Monsanto, has paid out $10 billion in Roundup settlements and faces 90,000 outstanding cases. The EPA says Roundup doesn't cause cancer; state juries disagree.

The courtroom itself is sacred: all marble, total silence, no politics. The justices sit in a mixed arrangement, with Chief Justice Roberts in the center. Lawyers have 30 minutes per side to make their case, with justices jumping in to interrogate. Friedberg described the lawyers as «LeBron James-level» — mind-blowingly impressive, distilling complex federal law into concise, high-stakes arguments. The government's position: the EPA has federal preemption. The plaintiff's counter: after Chevron was overturned, states have the right to interpret the law themselves.

The implications are vast. If states can ignore federal regulatory bodies like the EPA, the entire structure of federal authority is at risk. The court's decision will ripple far beyond Roundup.