The Wall Street Journal reported on April 28 that OpenAI missed an internal goal of one billion weekly ChatGPT users by the end of 2025 and missed multiple monthly revenue targets in early 2026, losing ground to Google's Gemini in consumer markets and to Anthropic in coding and enterprise. Sam Altman and CFO Sarah Friar issued a joint statement calling the report "prime clickbait" and saying OpenAI is "firing on all cylinders." The market disagreed by tens of billions of dollars. Oracle dropped 7.7%, CoreWeave fell 7.4%, SoftBank sank nearly 10% in Tokyo, and Nvidia, Broadcom, AMD, and Arm all declined between 2% and 6%.
For enterprise buyers, the headline volatility is noise. The signal underneath it is not. The signal is that the company whose growth assumptions underpin roughly $600 billion in committed compute spending is no longer the runaway leader in the market segment that matters most to most enterprises—and the vendor diversification conversation that some CIOs have been deferring for two years just got urgent.
What the Journal Actually Reported
Three things, all factual, none of them dependent on interpretation:
One billion weekly active users by end of 2025 was an internal target. OpenAI did not hit it. The company reached approximately 900 million weekly active users by February 2026—a 125% year-over-year increase at a scale where most products have plateaued. By any normal product standard, this is a generational consumer franchise. OpenAI is not operating by normal standards. It is operating by the standards required to justify $600 billion in compute commitments through 2030.
OpenAI missed multiple monthly revenue targets in early 2026. Annualized revenue sits at approximately $25 billion. CFO Sarah Friar warned colleagues internally that if revenue growth does not accelerate, OpenAI could face difficulty funding its future compute agreements. That warning is not hypothetical. Hundreds of billions in cloud commitments to Oracle, CoreWeave, and others assume revenue scales from $25 billion today to roughly $280 billion by 2030.
Friar reportedly told colleagues the company is not organizationally ready for the Q4 2026 IPO Altman has been targeting, preferring a 2027 listing. The joint "totally aligned" statement from Altman and Friar came after the Journal made the disagreement public. That is the order of operations communications teams resort to when the disagreement is real.
OpenAI's response—calling the report clickbait, pointing to "incredibly positive" internal mood, and emphasizing strength in consumer revenue and enterprise momentum—does not contradict the specific factual claims. It reframes them.
Why The Market Reaction Was So Sharp
Tens of billions of market cap moved on the report because OpenAI's commitments are contractual and OpenAI's revenue projections are aspirational. Three names absorbed the heaviest hits, and each one tells you something specific:
- Oracle (-7.7%) signed a $300 billion five-year partnership to supply OpenAI with computing power. If OpenAI can't pay, Oracle's data center buildout thesis cracks.
- CoreWeave (-7.4%) has an $11.9 billion infrastructure contract with OpenAI. CoreWeave's entire business model assumes continued hyperscale GPU demand from a small number of frontier labs.
- SoftBank (-10% in Tokyo) has committed $60 billion to OpenAI. Masayoshi Son has staked a chunk of his second AI bet on OpenAI being the dominant platform.
The chip stocks—Nvidia, Broadcom, AMD, Arm—fell because their forward growth math also assumes OpenAI keeps buying. When a single buyer represents that much pull-through demand, any sign that the buyer's revenue doesn't justify the commitments hits every link in the chain at once.
This is a concentration risk story, not an AI demand story. Enterprise AI demand is real and accelerating. The question is whether one company's revenue can justify one company's spending.
The Anthropic and Gemini Picture
Two competitive facts matter more for enterprise buyers than the OpenAI revenue miss itself:
Anthropic's annualized revenue crossed $30 billion in April 2026, passing OpenAI for the first time, with roughly 80% of that revenue coming from enterprise customers spending more than $1 million annually. The number of those $1M+ accounts doubled from approximately 500 to over 1,000 in less than two months. Anthropic is doing this while spending roughly a quarter of what OpenAI spends on training. The Claude family has been the consistent enterprise pick for code generation, agentic workflows, and high-stakes reasoning since mid-2025, and the customer concentration suggests that lead is widening, not narrowing.
Google Gemini gained consumer market share throughout 2025 and continues to outpace ChatGPT on several benchmarks, particularly in long-context, multimodal, and price-per-token comparisons. Inside Google Cloud accounts, Gemini is increasingly showing up as the default option because it is bundled, billed, governed, and audited inside infrastructure enterprises already own.
Add DeepSeek, Mistral, Meta's Llama, and the broader open-weight ecosystem—each competing on price and customization in ways that make it harder for any single vendor to capture enough share to justify $600 billion in infrastructure—and the picture is not "OpenAI is failing." The picture is "the market is splintering, and the splinters are competitive."
What OpenAI Is Doing Right—And Why It May Not Be Enough
OpenAI has genuine momentum to point to, and it is fair to acknowledge it before judging the math:
- Enterprise revenue now exceeds 40% of total revenue, with nine million paying business users—a fourfold increase since September 2025.
- The advertising business, launched in February, crossed $100 million in annualized revenue within six weeks and is projected to scale to $2.5 billion this year and $100 billion by 2030.
- GPT-5.5 and rapid model releases demonstrate a company shipping aggressively, not retreating.
- Approximately 50 million paying ChatGPT subscribers is a generational consumer franchise that most technology companies would never touch.
The problem is not that the business is bad. The problem is the asymmetry. To justify the $852 billion valuation set in March's $122 billion funding round, OpenAI needs to grow revenue more than ten times in four years from a $25 billion run rate to roughly $280 billion. The company's thesis has always been that scale wins: build the biggest models, deploy the most compute, acquire the most users, and revenue follows. The Journal's report is the first significant public evidence that scale alone may not deliver the revenue path required to fund the commitments.
The Enterprise Decoder
If you're a developer, ML engineer, or hands-on platform builder, here is what the report actually changes for your day-to-day:
OpenAI is not going away, and the API is not at near-term risk. Rumors of imminent service degradation are wrong. OpenAI has 50M consumer subscribers, $25B in annualized revenue, and Microsoft's distribution. None of that disappears because of one Journal article.
But model lock-in just got more expensive. If your stack assumes "OpenAI will be the cheapest, fastest, and best for the next three years," that assumption now carries measurable risk. Anthropic is winning on coding, agents, and long-horizon enterprise tasks. Gemini is winning on long-context, multimodal, and price-per-token. The right answer in mid-2026 is multi-model by default, with model-agnostic prompt and tool layers in front of provider SDKs.
Watch the compute commitments, not the press releases. The most important number for OpenAI's trajectory is not its valuation or its weekly active users. It is the rate at which it draws down the $600 billion in committed compute. If that drawdown decelerates because revenue lags, expect price changes, rate-limit changes, and tier restructuring. None of that breaks production today, but it changes long-range capacity planning.
Open-weight and self-hosted options are now real options for the right workloads. DeepSeek V4, Llama, and Mistral are no longer second-tier. For internal coding assistants, RAG pipelines on private data, and agentic workflows where latency and cost dominate, open-weight on owned GPUs is increasingly the right answer—especially if your security posture wants the model in your VPC.
If you're a CIO, CTO, head of AI, or platform owner, the strategic implications cut differently:
Vendor concentration is the risk, not vendor selection. If OpenAI is more than 70% of your AI inference spend, you are exposed to a single company's compute math working out. Diversify before you have to. The cleanest test: can your platform team swap Claude for GPT-5.5 in a production agent in under a week without changing business logic? If not, that is the first thing to fix in Q2.
Renegotiate enterprise contracts now, not later. OpenAI, Anthropic, and Google all want enterprise commitment in 2026 because each one is trying to lock down the segment that pays. Use the competitive pressure. Anthropic specifically is hungry for accounts in regulated industries. Google is bundling Gemini into Workspace, GCP, and Vertex with pricing that is designed to displace OpenAI inside customers it already serves.
The "single AI vendor" architecture is now a board-level risk. When OpenAI's largest backer (SoftBank) drops 10% in a single session and the largest infrastructure partners (Oracle, CoreWeave) drop nearly 8%, that is the market pricing in concentration risk. Your AI architecture should not assume any single frontier lab is invariant over the next 24 months. Build for substitutability. Audit for it. Make it a procurement requirement.
Cost models built on assumed OpenAI price decreases need a sensitivity analysis. Many enterprises baked aggressive token-cost-decline assumptions into 2026 and 2027 AI budgets. Those assumptions implicitly required OpenAI to keep undercutting itself on price. If OpenAI's revenue pressure forces it to hold prices firm or restructure tiers, your TCO model misfires. Sensitivity test your AI budget against flat token costs, not declining ones.
Governance matters more, not less. As the AI vendor landscape splinters, the governance overhead—prompt logging, model routing, data residency, audit trails, identity binding for agents—becomes the platform layer that survives any vendor shift. If you have not centralized it, this is the quarter to start. The operational pain of running multiple frontier vendors without a control plane is the real cost of a multi-model strategy. A real MCP gateway, a real model router, a real prompt registry, and real DLP at the prompt boundary stop being nice-to-haves.
Three Concrete Moves for the Next 30 Days
-
Pull a vendor concentration report from your AI cost data. Sum spend by model family across teams. If OpenAI is more than 60% of total inference spend, schedule a Q2 review with platform and finance leadership specifically on diversification and contract renegotiation leverage.
-
Run a head-to-head bake-off on your top three production prompts. Use GPT-5.5, Claude Opus 4.5 or 4.7, and Gemini 2.5 Pro on the same evaluation set with the same scoring rubric. Three prompts, three models, real numbers—not vendor benchmarks. Most enterprises discover that one of their top workloads is materially better and cheaper on a non-OpenAI model. That single finding usually justifies the multi-model investment.
-
Add concentration risk to your AI vendor scorecard. If your scorecard measures latency, accuracy, and price but not vendor financial health and substitutability, you are missing the dimension this week made most expensive. Add a column. Score it. Track it.
The Bottom Line
OpenAI did not collapse. The headlines in the next week will overstate what changed and the rebuttals will understate it. What actually changed is the public evidence that the OpenAI thesis—scale wins, revenue follows, commitments justified—has a measurable gap between contractual spending and current revenue trajectory. The competitive picture shifted from a near-monopoly with two challengers to a three-horse race where each horse is winning a different segment.
For enterprise buyers, the right response is not panic and not denial. It is a clean-eyed audit of how exposed your AI architecture is to any one vendor's roadmap, pricing, or solvency, and a deliberate rebalancing toward substitutability before the market makes that decision for you. The companies that are going to win the next phase of enterprise AI are the ones whose platform engineers can route a workload to whichever model wins on quality and cost this quarter, without rewriting the application.
That capability is not a model choice. It is an architecture choice. And April 28, 2026 was the day the market made that choice less optional.
Sources:
- If OpenAI Is Missing User and Revenue Targets, Are They Falling Behind the AI Race? — The New York Times, April 28, 2026
- OpenAI says it is "firing on all cylinders" after missing revenue and user targets, but market wipes billions off AI stocks — The Next Web, April 28, 2026
- Market slumps as OpenAI reportedly misses internal targets for active users and revenue — Tom's Hardware, April 28, 2026
- OpenAI reportedly missed revenue targets. Shares of Oracle and these chip stocks are falling — CNBC, April 28, 2026
- OpenAI misses key revenue, user targets in high-stakes sprint toward IPO — Wall Street Journal, April 28, 2026
Want to calculate your own AI ROI? Try our AI ROI Calculator — takes 60 seconds and shows projected savings, payback period, and 3-year ROI.
