KEY POINTS
- Oracle shed roughly 4%, Broadcom 4%, AMD 3% and SoftBank near 10% on Tuesday after the WSJ reported OpenAI missed user-growth and revenue targets, ending ChatGPT's run of unbroken expectations.
- Anthropic crossed a $30 billion annualized revenue run rate in April, passing OpenAI for the first time and reframing the consumer AI race that had powered every chip multiple since late 2023.
- Today's Big Tech capex guidance from Microsoft, Alphabet, Meta and Amazon will determine whether the chip de-rate that started this week becomes a sector trend or a one-day shakeout.
The single biggest pillar under the AI trade cracked this week. Oracle dropped about 4%, Broadcom 4%, Advanced Micro Devices 3% and SoftBank Group nearly 10% in Tokyo on Tuesday after the Wall Street Journal reported that OpenAI has missed multiple internal targets for ChatGPT user growth and monthly revenue, including its goal of reaching one billion weekly active users by the end of 2025. Nvidia, fresh off a record close last week and a market capitalization north of $5 trillion, also slipped as traders rushed to reprice every stock whose growth narrative leans on OpenAI's compute appetite.
This is the first time the OpenAI flywheel has visibly stalled in a way the public market can price. Finance chief Sarah Friar has reportedly warned colleagues that revenue growth must accelerate to fund future compute commitments, including the company's $300 billion, five-year deal with Oracle. That single contract is the entire bull case for Oracle's cloud re-rating. When the customer behind it is missing its own numbers, the multiple goes with it.
The Anthropic Crossover
The bigger story under the headline is competitive. Anthropic's annualized revenue run rate crossed $30 billion in April, the first time it has surpassed OpenAI. Anthropic now leads in enterprise coding workloads and is winning an outsized share of new API spend, while Google's Gemini has eaten into ChatGPT's consumer share. OpenAI is no longer the obvious default, and that matters because the AI capex thesis has been priced as if every dollar of model spend flows back to a single hyperscaler customer base anchored by ChatGPT subscriptions.
It also matters for chip allocation. Anthropic's compute is heavily routed through Amazon's custom Trainium silicon and Google TPUs, which cuts into the Nvidia-only narrative that has been pricing AMD's MI400 ramp and Broadcom's custom-ASIC business as derivatives of OpenAI demand. If the world's number-two and number-three model builders are on different chip stacks, the second-derivative names need to be repriced individually rather than as one beta.
The Memory and Power Squeeze
The OpenAI report landed on a market that was already nervous about AI economics for a different reason. IDC projects DRAM will cost $9.71 per gigabyte in 2026, up from $3.76 in 2025, a 158% jump that flows directly into hyperscaler bills of materials. Micron has guided that demand will exceed supply for server DRAM, HBM stacks and data-center SSDs throughout 2026. Energy is the other constraint. The recent spike in Brent crude past $118, after the UAE's exit from OPEC and renewed tensions with Iran, has reignited the question of whether AI campuses are pricing power correctly into their unit economics.
Stack those costs against an OpenAI that is missing revenue, and the math behind the $700 billion of combined 2026 AI capex commitments from Microsoft, Alphabet, Meta and Amazon starts to look less linear. Free cash flow at three of the four hyperscalers has been falling faster than the consensus model assumes, with Microsoft's last-reported quarterly capex at $29.88 billion, up 89% year over year, against single-digit operating margin expansion in cloud.
What Today's Tape Tells You
The pre-market on Wednesday showed a partial bounce in semis but lower volume than Tuesday's selloff, which reads as positioning rather than conviction. Traders are squaring up ahead of after-the-bell prints from Microsoft, Alphabet and Meta, with Amazon following Thursday. The market is no longer rewarding AI capex as a self-justifying line item. It is asking whether spend turns into Azure bookings, Google Cloud backlog and Meta ad revenue at rates that justify the run rate.
Watch the questions on tonight's earnings calls more than the headline numbers. If Microsoft, Alphabet, or Meta walks back any portion of their capex envelope or hedges on training-cluster timing, the chip de-rate from this week becomes a trend, and the natural next target is Oracle's $300 billion backlog at OpenAI, which is now visibly weaker than it was two weeks ago. If, instead, all three hold or raise capex and report accelerating cloud revenue, the OpenAI miss gets reframed as one customer's problem and the chip names retake recent highs by Friday. Either way, the stock to watch on Thursday open is Oracle. It owns the cleanest read on whether the OpenAI revenue miss is being treated as a credit event or a near-term valuation reset.
