Feb 28, 2026 · 24-Hour AI Briefing: OpenAI’s $110B Raise, Meta Reconsiders Olympus, PayPal’s 6-Month Data Leak
In the last 24 hours, three updates put both the ceiling and the floor of the AI industry on display. OpenAI is doubling down on a platform-scale trajectory with massive funding. Meta’s Olympus rethink highlights how hard it is to build training silicon at frontier scale. And PayPal’s prolonged data exposure is a reminder that in AI-plus-finance, security failures don’t just cost money—they erode trust.
1) OpenAI announces a new $110B funding round; 900M+ weekly active users, 50M+ consumer subscribers, 9M+ paid enterprise users; Codex weekly actives more than doubled to 1.6M
Commentary:
Based on the structure you provided, this looks like pre-loading ammunition for sustained compute intensity and product portfolio expansion. Amazon signals cloud and distribution, NVIDIA signals core compute supply, and SoftBank signals long-horizon capital and global leverage. At this scale, OpenAI is behaving less like a single-product company and more like infrastructure.
The 9M+ paid enterprise users reinforces B2B durability: if OpenAI can keep translating productivity into measurable ROI, enterprises will treat it as budgeted infrastructure rather than a nice-to-have tool. Codex growing to 1.6M weekly actives also matters—once you own the programming workflow entry point, the advantage extends beyond model quality into ecosystem, integrations, permissions, auditability, and team collaboration.
The key question is how long ChatGPT’s user growth and paid growth can keep compounding—and whether OpenAI can convert scale into a stable, high-retention paid structure before growth inevitably slows.
2) Meta reconsiders its Olympus second-gen training chip planned for 2026 due to technical complexity and manufacturing risk
Commentary:
Olympus struggling may be less about “capability” and more about how high the ecosystem barrier really is. With a SIMT-style design aimed at CUDA compatibility and trillion-parameter-scale training, you’re not just building a chip—you’re trying to replicate years of systems engineering, compiler maturity, kernels, and cluster orchestration.
What stands out is Meta’s pragmatic pivot. While maintaining its NVIDIA relationship, Meta reportedly signed a massive inference-oriented MI450 deal with AMD and also moved to rent Google TPUs for new model training. The strategy is clearly shifting from a single bet to a diversified compute portfolio—NVIDIA GPUs + Google TPUs + in-house MTIA—focused on supply certainty and $/token economics.
It’s also a signal to everyone else: AI chip self-development is a long, unforgiving road, even for the largest companies.
3) PayPal discloses a roughly 6-month data exposure caused by a code error in its business lending system; sensitive data involved; remediation includes two years of credit monitoring
Commentary:
The incident reportedly stems from a change in the PayPal Working Capital (PPWC) loan application, which unintentionally exposed customer PII in API responses. The most alarming part is duration: July 1, 2025 to Dec 13, 2025—about 165 days—without logging or alerts catching abnormal data flow, creating a real possibility that data was harvested at scale long before discovery.
Even with two years of Equifax monitoring, identity risk can persist for years, and the long-tail cost often lands on users. The remediation is standard, but the broader lesson is sharper: in an AI-enabled financial world, a single unreviewed line of code can become a trust-destruction event.