OpenAI vs Anthropic vs Google: Inside the Trillion-Dollar AI Spending War
Updated July 16, 2026
By Peak of Trending
Four hyperscalers spent close to $700 billion building AI infrastructure in 2026, more than triple the roughly $200 billion the same group spent in 2024, according to Fortune's tracking of quarterly capital filings. Google alone committed $175 billion to $185 billion of that pile, with Sundar Pichai confirming that just over half goes straight into the chips and data centers running Gemini. OpenAI, in the same window, is refusing to list on public markets for anything under a $1 trillion valuation. Anthropic just raised $65 billion at $965 billion, a number that now sits above OpenAI's own private mark. None of the three companies is running the same race, and the spending totals hide that fact instead of revealing it.
Most coverage of this rivalry collapses into a single scoreboard: who has more users, whose model tops which benchmark this week, whose valuation is bigger. That framing misses the actual fault line, which is that OpenAI, Anthropic, and Google are betting on three different theories of how AI money eventually gets made back — and only one of those theories has produced a profitable quarter so far. A reader trying to decide which company to trust with a platform bet, a career, or an investment gets none of that from a headline that just says "AI arms race."
This piece lays out what each company actually spent, earned, and lost through the middle of 2026, where their strategies diverge, and where the industry's own regulators are now flagging risk that the press releases skip over.
- The Capex Number Nobody Can Quite Explain
- OpenAI's Trillion-Dollar Line in the Sand
- Anthropic's Quiet Overtake
- Google Doesn't Have to Raise a Dollar
- The Compute Leash Nobody Priced In
- Who Actually Wins Right Now
- Who This Race Is For
- The Verdict
The Capex Number Nobody Can Quite Explain
Combined AI capital spending across the four biggest hyperscalers is on pace to hit roughly $725 billion in 2026, up from about $410 billion in 2025 and just over $200 billion in 2024, per tracking published by ValueAdd VC. Amazon leads at around $200 billion, Google sits second near $180 billion, and Meta and Microsoft trail at $115–135 billion and $110–120 billion. Layer OpenAI's $500 billion Stargate build-out with Oracle and SoftBank on top, and the four-year infrastructure commitment from this single sector now exceeds the annual GDP of most member states of the United Nations.
None of that capital comes with a guarantee attached. Analysts cited by the 365Outsource coverage of Q1 2026 earnings flagged that much of the spending is justified by projected demand that has not yet shown up in hyperscaler revenue lines. Alphabet has already said 2027 capex will rise again. The bet only pays off if usage keeps compounding at the same rate it has since 2024.
The number nobody uses: Anthropic now spends 2.3 times its payroll on compute, against roughly 0.4 times payroll at the top 1% of ordinary software companies, based on an analysis by investor Tomasz Tunguz. That ratio is the actual cost structure of frontier AI, and it is nothing like running a normal software business.
OpenAI's Trillion-Dollar Line in the Sand
OpenAI filed a confidential draft S-1 with the SEC on June 8, 2026, and the number attached to that filing is $1 trillion. That figure sits well above the company's last private mark of $852 billion, set in March 2026 when OpenAI closed a $122 billion round backed by SoftBank, Amazon, and Nvidia. Sam Altman's advisers reportedly gave him two options, according to reporting picked up by TheStreet from The New York Times: accept a lower price and list in late 2026, or hold the line and wait until 2027. Altman called anything under $1 trillion a non-starter.
The financial picture underneath that stance is not simple. OpenAI's 2025 revenue reached roughly $13 billion to $20 billion depending on the source, with 2026 run rate estimates clustering near $24–25 billion. Against that, the company posted an $8.5 billion net loss in the first quarter of 2026 alone, and Forbes reported projected cumulative losses of up to $115 billion by 2029, with profitability not expected until the 2030s. SoftBank, meanwhile, has a $40 billion loan coming due in March 2027 that depends partly on OpenAI's shares becoming liquid on schedule.
A trillion-dollar number, pinned to a company still years from turning a profit.
Anthropic's Quiet Overtake
Anthropic closed a $65 billion Series H in May 2026 at a $965 billion post-money valuation, according to the company's own funding announcement. That values Anthropic above OpenAI's last private mark, and it came alongside a run-rate revenue figure of $47 billion, up from $30 billion earlier the same year and roughly $10 billion twelve months before that. Enterprise customers, drawn largely by Claude Code, now account for the bulk of that growth.
Here is where the research shifted my read on the whole comparison: I expected Anthropic's overtake to be a valuation story, a round of investor enthusiasm outrunning fundamentals the way OpenAI's own numbers sometimes do. Instead, CNBC reported that Anthropic's projected $10.9 billion in second-quarter 2026 revenue came with roughly $559 million in operating income attached — its first profitable quarter, a milestone SmartAsset's IPO coverage notes OpenAI has yet to reach even once. Anthropic itself has told investors it expects sustained annual profitability by 2029, about a year ahead of its rival's own internal timeline.
A five-year-old research lab just posted the only profitable quarter in this entire comparison.
Google Doesn't Have to Raise a Dollar
Google is the only company in this comparison that never has to file an S-1, because it already trades, and its AI spending simply shows up as a line inside Alphabet's existing cash flow. That structural advantage lets Google absorb a price war it started itself: at I/O 2026, Sundar Pichai cut Ultra subscription pricing by 20% and introduced a cheaper entry tier, a move TradingKey covered as a direct shot at Anthropic and OpenAI's own subscription tiers, which run near $100 and $200 a month respectively.
On raw capability, Gemini is currently ahead. Gemini 3.1 Pro topped 13 of 16 major benchmarks tracked by industry analysts, including a 77.1% score on ARC-AGI-2 reasoning and 80.6% on SWE-bench coding, both the highest deployed figures at time of testing. Distribution tells the same story from a different angle: Gemini's global web-visit share climbed from roughly 6% a year earlier to nearly 28% by mid-2026, and Google is folding the same model family into a rebuilt Siri across more than a billion iPhones under its Apple partnership, expected to ship with iOS 27.
Google is not trying to win the AI argument. It is trying to make the argument irrelevant by owning the pipes underneath it.
The Compute Leash Nobody Priced In
"The wave of AI deals is more like a virtuous circle that helps line up suppliers, builders and customers to meet the exploding demand for computing power," Janus Henderson told Bloomberg, describing the same arrangement the Bank for International Settlements just flagged as a systemic risk.
The Bank for International Settlements named an AI capex bust and the collapse of circular financing among its top three threats to global financial stability in its June 2026 annual report, citing roughly $1.8 trillion in interlocking equity, debt, and supply commitments between chipmakers, hyperscalers, and AI labs, per coverage of the BIS report. Nvidia, OpenAI, Oracle, and Microsoft sit at the center of the web the BIS is worried about, and Anthropic is not exempt from the same structural fragility.
Anthropic's own compute contract with SpaceX covers roughly 325,000 Nvidia GPUs at $1.25 billion a month, and it carries a 90-day termination clause either party can trigger after an initial three-month lock-in, according to research firm Sacra's coverage of the deal. Elon Musk has publicly described the arrangement as a lease with an off-ramp built in, in case SpaceX needs its own capacity back.
The failure named: a study cited in Tom's Hardware found that 95% of enterprise generative AI deployments show no measurable effect on profit or loss, a figure advocates for the spending boom rarely bring up unprompted.
Who Actually Wins Right Now
The scoreboard splits depending on which layer you measure. ChatGPT's consumer market share fell below 50% for the first time since launch in mid-2026, per Sensor Tower's State of AI report, even as OpenAI still leads on raw monthly users worldwide. Anthropic, by contrast, holds roughly a third of the enterprise AI market against 25% for OpenAI and about 20% for Google, according to survey data from Menlo Ventures cited by Yahoo Finance. Google leads on model benchmarks and on total platform reach once Android, Search AI Overviews, and Workspace are folded in.
| Company | 2026 Valuation / Position | Revenue Run Rate | Where It Leads |
|---|---|---|---|
| OpenAI | $852B private mark; targeting $1T IPO | ~$24–25B | Consumer scale, brand recognition |
| Anthropic | $965B post-Series H | ~$47B | Enterprise share, first profitable quarter |
| Already public; ~$180B 2026 AI capex | Folded into Alphabet | Model benchmarks, distribution |
Figures reflect the latest publicly available data at time of writing. Always verify current numbers with each company's own disclosures.
Who This Race Is For
An enterprise buyer choosing a default model provider for 2027 budgets needs a different answer than a retail investor deciding whether to apply for OpenAI's promised retail IPO allocation. The first group should weigh Anthropic's coding-market lead and its unusual quarterly profit against the real possibility of a compute contract cancellation with 90 days' notice. The second group is effectively betting on whether Sam Altman's trillion-dollar floor holds through 2027, or whether the same advisers who raised the SoftBank-loan concern end up right about a delayed, lower-priced listing.
A developer picking an API for a new product faces a third calculation entirely: Gemini's benchmark lead and Google's price cuts against Anthropic's coding-specific tooling and OpenAI's still-larger consumer distribution funnel.
The Verdict
Google is the safest bet of the three because it never needed this race to work in order to survive it. Anthropic is the most interesting bet because it is the only one of the three proving the business model can turn a profit quarter, even if a single Musk phone call could still cut its compute in half. OpenAI is the highest-variance bet, anchored to a valuation number its own advisers reportedly doubt, propped up by a SoftBank loan schedule that does not care whether the IPO timing is convenient.
None of these three companies has to lose for the other two to be right, and that is exactly what makes the BIS warning worth sitting with instead of dismissing.
Frequently Asked Questions
What valuation is OpenAI targeting for its IPO?
OpenAI is targeting a $1 trillion valuation for its planned public listing. That figure sits above its $852 billion private valuation from March 2026, and Sam Altman has reportedly rejected any offering priced below the trillion-dollar mark.
Why did Anthropic overtake OpenAI in revenue?
Anthropic's enterprise business, driven heavily by its Claude Code coding tool, pushed its run-rate revenue to $47 billion by May 2026, ahead of OpenAI's roughly $24–25 billion. Business subscriptions and coding-tool adoption drove most of that growth.
Is Google spending more on AI than OpenAI and Anthropic?
Google's own 2026 AI capex runs $175–185 billion, more than either OpenAI or Anthropic has raised in total private funding to date. Google funds this from Alphabet's existing cash flow rather than dedicated fundraising rounds.
What is the AI circular financing risk?
Circular financing refers to chipmakers, hyperscalers, and AI labs investing in each other while also signing multi-year supply and purchase agreements between the same parties. The Bank for International Settlements flagged roughly $1.8 trillion in these interlocking commitments as a systemic financial risk in its 2026 annual report.
Which company is currently profitable: OpenAI, Anthropic, or Google?
Anthropic posted its first profitable quarter in Q2 2026, with roughly $559 million in operating income on projected revenue of $10.9 billion. OpenAI has not yet reported a profitable quarter and does not expect sustained profitability until the 2030s. Google's AI spending sits inside Alphabet's broader, already-profitable operations.
Which AI model currently leads on benchmarks: Gemini, Claude, or GPT?
Gemini 3.1 Pro led 13 of 16 major benchmarks tracked by industry analysts as of mid-2026, including top scores on ARC-AGI-2 reasoning and SWE-bench coding. Claude and GPT models remain close competitors on several of the same tests.
How much is the AI industry spending on infrastructure in 2026?
Combined AI capital expenditure across the four largest hyperscalers is on pace to reach roughly $725 billion in 2026, up from about $410 billion in 2025. That figure does not include OpenAI's separate $500 billion Stargate infrastructure commitment.
Is Anthropic going public before OpenAI?
Anthropic filed confidentially for an IPO on June 1, 2026, about a week before OpenAI's own confidential filing. Neither company has confirmed a final listing date, and OpenAI's advisers have reportedly discussed pushing its offering into 2027.