Michael Sikand
Investor, Autopilot Portfolio, Founder of Sikand Media
Oracle is one of the names most levered to OpenAI, so a sentiment swing toward OpenAI's GPT-6 Astra could drive a pre-earnings run-up. OpenAI research VP Aidan Clark says Astra was its first pre-training on over 100,000 GPUs at the Stargate site in Abilene, Texas, which Oracle helped build, and half of Oracle's $600B+ backlog comes from OpenAI. Kalshi odds of OpenAI having the best model by year-end doubled in days. This is a short-term trade. x.com x.com x.com x.comÂ
OpenAI running ahead of its revenue projections derisks Oracle, whose backlog is 54% OpenAI, making it attractive again at 18x forward earnings after its worst week since 2001. Brad Gerstner said Anthropic and OpenAI together were doing about $80 billion in revenue; with Anthropic reportedly near $50 billion, OpenAI is around $30 billion ARR, above the $25 billion it projected for FY2026 last winter. x.comÂ
Diversified cash flows in the billions make the AI-buildout basket more appealing than standalone model companies. Many members look reasonably priced and have longer records of delivering returns; Anthropic and OpenAI look much more like gambles at current valuations, though either could appeal at the right price. A possible equal-weight portfolio to revisit in five years includes Amazon, Alphabet, Microsoft, Meta, Oracle and SpaceX. x.com x.comÂ
Oracle's infrastructure backlog carries a financing risk: a reported ~$300B OpenAI deal accounts for much of its $664B backlog, raising a real question about OpenAI's ability to pay. The reported ~$50B versus ~$70B OpenAI run-rate gap is mostly a difference in how cloud-partner sales are counted, not by itself evidence that physical AI demand has fallen. x.com x.comÂ
Oracle looks mispriced because $11 of current consolidated earnings power puts it at 11x earnings without a breakup. It trades below its software business's value despite the highest gross retention in software; separating pre-AI OCI, AI operations and AI development makes it an obvious long. Future funding will cost more, but current leverage is no issue and all contracts are well below spot, leaving nowhere near neocloud risk. Fundamentals may not matter during panics. x.com x.com x.comÂ
Oracle's AI buildout looks fundable as spending peaks and cash flow turns positive. Spending is forecast to peak in FY27-28, with free cash flow positive in 2H FY28 and above $30bn in FY29, a roughly $54bn swing. New capital raising is capped at $20bn through 2030, half equity and half debt. Revenue is projected to grow from $90bn next year to $225bn+ by FY30. Bullish Oracle. x.comÂ
Oracle's reliance on OpenAI makes its backlog vulnerable. About $250B of backlog comes from OpenAI, which is not making money; that risk is no longer in Oracle's hands, and the stock is very likely to fall further. x.com x.comÂ
Oracle is a buy here if it gets paid. A drop of more than 60% from a 52-week high has happened only once before, and passing on this level looks like a mistake that will be obvious in four years. x.comÂ
Oracle's OpenAI backlog is not just paper: OpenAI just pretrained Astra on more than 100,000 GB200s for about two months at Abilene. That is 144 million GPU-hours, over $1 billion at published Blackwell rates, making it the first billion-dollar training run done on rented compute. Time to accelerate. x.comÂ
Oracle's OpenAI exposure makes buying its shares here a bet too big to take. x.com x.comÂ
Oracle's non-AI revenue and scale could help it outlast debt-starved neoclouds. Elon Musk says debt markets have largely shut out firms lacking substantial non-AI revenue. Oracle raised $20B through an ATM without significant dilution; unlike other major hyperscalers, it is not building Nvidia alternatives, giving Nvidia reason to prioritize it. If Musk is right, consolidation could lift Oracle's pricing power; at 19x earnings, it is without a doubt one of AI infrastructure's best buys. x.comÂ