The AI Platform Economy: How OpenAI, Anthropic and Google Are Building Different Businesses

Executive Intelligence Series · Company Analysis | September 26, 2026

Three companies now sit at the center of artificial intelligence, and they are not building the same business. OpenAI is spending at a scale unmatched in corporate history to make ChatGPT the default interface to computing. Anthropic has built the fastest, most capital-efficient scale-up in enterprise software history by concentrating on Claude and Claude Code. And Google is folding frontier AI into an already-dominant advertising, cloud, and productivity empire it never had to build from scratch. All three are chasing the same AI platform economy — from three fundamentally different starting positions.

The AI Platform Economy: How OpenAI, Anthropic and Google Are Building Different Businesses

Executive Overview

The question is no longer which lab has the best model. Frontier systems from OpenAI, Anthropic, and Google now trade leadership every few months, and the gap between the best available models has narrowed to the point that model quality alone rarely decides a customer's choice of vendor. What has emerged instead is a platform logic — a foundation model is one layer in a stack that includes infrastructure, distribution, tooling, and an ecosystem of dependent companies, and the real question is where each company sits in that stack and how much value it can capture.

By mid-2026 the three companies' reported scale had become almost impossible to compare on a like-for-like basis: OpenAI at roughly $40 billion in annualized revenue against an $852 billion valuation and more than $1 trillion in compute commitments; Anthropic at roughly $47–65 billion in annualized revenue against a $965 billion valuation reached in barely 14 months; and Google Cloud alone crossing a $100 billion run rate inside a parent company generating over $400 billion a year, funded almost entirely from existing cash flow rather than external capital.

This is not an attempt to declare a winner. The value in comparing OpenAI, Anthropic, and Google side by side is in showing how three fundamentally different corporate structures are attempting to capture value from the same emerging market — and what that divergence means for enterprise buyers, developers, and the rest of the technology industry watching the AI platform economy take shape.

The AI Platform Economy

For the first two years of the generative AI boom, the industry's center of gravity was the model itself — which lab had the most capable system. That framing has largely collapsed. A foundation model is now one layer in a stack that includes infrastructure (chips, data centers, power), distribution (a way to reach hundreds of millions of users or millions of developers), tooling (APIs, coding agents, enterprise integration), and an ecosystem of dependent companies building on top of the platform.

Foundation-model companies bear enormous fixed costs — training runs and compute contracts measured in tens or hundreds of billions of dollars — while much of the commercial upside may ultimately accrue to the applications built on top of the models, or to the infrastructure providers underneath them. OpenAI, Anthropic, and Google are making different bets about where they can defend a durable position: OpenAI on being the default consumer and developer interface to AI; Anthropic on being the trusted enterprise and coding layer; Google on owning the full stack, from custom silicon to Search, so it can capture value at every layer simultaneously.

The framework running through this analysis: Models → Infrastructure → Distribution → Customers → Monetization → Ecosystem → Platform Expansion. Each company is strong at different points on that chain — and weak at others.

The CODEW Lens: "The AI company" is no longer a useful frame. There are at least three different companies pursuing three different theories of where value accumulates in this stack — and the next 18 to 24 months should start to show which bets were right.

Table 1 · Three Different AI Business Models

Dimension	OpenAI	Anthropic	Google
Core platform	ChatGPT (900M+ weekly users) / API	Claude / Claude Code / API	Gemini app (950M+ MAU) inside Search, Workspace, Android, Cloud
Distribution	Direct consumer + enterprise sales + developer API	Enterprise + developer API; minimal consumer push	Search, Cloud, Workspace, Android, YouTube — pre-existing at global scale
Infrastructure	External partnerships at extreme scale; ~$1.1–1.4T committed through mid-2030s	Deliberate multi-vendor hedge (AWS Trainium, Google TPU/Broadcom, Nvidia)	Vertically integrated custom TPU silicon + GPUs, self-funded
Monetization	Subscriptions + enterprise + API; ~$25–40B ARR (company-reported)	Enterprise + API + coding agents; ~$47–65B ARR (company-reported)	Advertising + Cloud + Workspace seats + devices; Cloud alone >$100B run rate
Strategic model	AI-native platform, funded by external capital and vendor financing	Model/enterprise platform, funded by strategic investment from cloud/chip partners	AI integrated across an already-dominant, self-funding technology empire

Note: OpenAI and Anthropic revenue and valuation figures are company-reported and unaudited unless otherwise cited to SEC filings; treat as directional and subject to revision.

OpenAI — Building an AI-Native Platform

OpenAI's business is built outward from ChatGPT, which functions as both product and distribution channel. By mid-2026, the company reported more than 900 million weekly active users and over 1 million business customers — a scale of consumer reach no rival, inside or outside AI, has matched this quickly. ChatGPT is no longer positioned as a chatbot, but as an operating layer people return to for search, work, coding, and increasingly, agentic tasks.

Monetization runs on three tracks — consumer subscriptions, an API/developer business serving roughly 4 million developers, and a growing direct enterprise sales motion. Combined, these pushed OpenAI's annualized revenue run rate from roughly $12 billion at the start of 2026 to more than $40 billion by August 2026 — though independent verification is thin. Microsoft's own FY2026 filings disclosed $24.1 billion in revenue attributable to OpenAI (inclusive of revenue-sharing) for the twelve months through June 2026 — the most reliable third-party figure available, since OpenAI does not publish audited financials.

The more consequential number may be what OpenAI is spending, not earning. The company has committed to roughly $1.1–1.4 trillion in compute obligations through the mid-2030s across Oracle ($300B/5yrs for Stargate-linked cloud capacity), Nvidia (up to $100B tied to at least 10GW of Nvidia-based data centers), AMD (warrants covering ~10% of AMD in exchange for 6GW of GPU capacity), Broadcom, Microsoft, AWS, and CoreWeave. The Stargate joint venture — unveiled at the White House in January 2025 — has grown to nearly 7 gigawatts of confirmed capacity and more than $400 billion in near-term commitments. Analysts have flagged the circularity in this financing: several of OpenAI's largest infrastructure partners are simultaneously its investors, paid back in part through the same purchase commitments they finance.

Platform expansion is where OpenAI's ambitions are broadest and least proven — agentic products, a developer ecosystem around GPT models, Sora for video, and reported ambitions in commerce, browsers, and hardware. The strategic logic: make ChatGPT the default interface through which consumers and businesses access AI-native software, competing for the role Google Search played for the web.

The CODEW Lens: OpenAI has the clearest consumer distribution of the three and the least defensible balance sheet. The bet is that scale and speed outrun the financing risk before it comes due.

Anthropic — Enterprise AI and the Model Layer

Anthropic has taken the narrowest and, by several measures, the most capital-efficient path of the three. Rather than building a mass-consumer surface, it has concentrated on enterprise customers and developers through Claude and Claude Code, its agentic coding product, which reportedly reached roughly $8 billion in annualized revenue by mid-2026 after launching in May 2025.

The financial trajectory has been extraordinary. Anthropic's annualized revenue climbed from roughly $1 billion at the end of 2024 to about $47 billion by May 2026, and reportedly to around $65 billion by July 2026 — among the fastest scaling curves ever recorded for an enterprise software business. That growth funded a rapid string of private valuations: $61.5B (Series E, Mar 2025) → $183B (Series F, Sep 2025) → $380B (Series G, Feb 2026) → $965B (Series H, May 2026) — briefly the most valuable private AI company in the world, ahead of OpenAI's $852B March 2026 mark. Anthropic has reportedly filed confidentially for an IPO, with valuation chatter reaching toward $2 trillion — figures that remain unconfirmed and should be treated as developing.

Anthropic's enterprise positioning rests on safety-and-reliability messaging aimed at risk-conscious buyers, deep API tooling, and reported dominance in enterprise LLM usage — Menlo Ventures data cited in mid-2026 put Anthropic at roughly 40% share of enterprise LLM usage, ahead of OpenAI's 27% and Google's 21%, concentrated in coding and agentic workloads. The company reports more than 300,000 business customers, 70% of the Fortune 100, and over 1,000 accounts spending $1 million-plus annually.

Where Anthropic differs most from OpenAI is infrastructure strategy: a deliberately multi-vendor hedge across AWS Trainium (backed by up to $25B+ from Amazon), Google TPUs and Broadcom silicon (a partnership that grew from an initial $40B Google investment tied to 5GW of TPU capacity into a reported $200 billion, five-year compute commitment beginning 2027), and Nvidia GPUs (Microsoft and Nvidia together committing roughly $15B). Anthropic still expects to spend close to $19 billion on training and inference in 2026 — an amount roughly matching its full-year revenue.

The CODEW Lens: Anthropic isn't trying to own the interface between consumers and AI — it's trying to own the interface between AI and enterprise software development, and to monetize that position at a premium reported average revenue per account many multiples above OpenAI's blended, consumer-weighted average.

Google — AI Inside an Existing Technology Empire

Google is not building an AI company from scratch — it is embedding frontier AI inside a business that already generates several hundred billion dollars a year from Search, YouTube, Cloud, Workspace, and Android. Alphabet's full-year 2025 revenue exceeded $400 billion, with Google Services still the overwhelming majority, and Google Cloud the fastest-growing segment.

That existing distribution is Google's single largest structural advantage. Gemini doesn't need to build a user base from zero — it's embedded in Search (AI Overviews, AI Mode), Android, Chrome, Workspace, and YouTube. By Q2 2026, the standalone Gemini app alone had grown to roughly 950 million monthly active users, with Google's models processing about 22 billion API tokens per minute — figures that likely undercount total Gemini exposure once Search and Workspace integrations are included.

Google Cloud is where the AI investment converts most visibly into new revenue. Cloud revenue accelerated from roughly $17.7 billion in Q4 2025 (+48% YoY) to $20.0 billion in Q1 2026 (+63%) and $24.8 billion in Q2 2026 (+82%), with operating margin expanding from the high teens to above 35%. Cloud backlog grew from roughly $240 billion at the end of 2025 to $514 billion by mid-2026. Google began recognizing revenue from direct TPU hardware sales to customers in 2026 — including the Anthropic compute deal — turning custom chip design from an internal cost center into an external revenue line.

That is the core of Google's edge: vertical integration. Google designs its own TPU chips (seventh-generation "Ironwood," with an eighth generation announced), operates its own data centers, and trains and serves Gemini on its own infrastructure. Alphabet's 2026 capex guidance rose over the year from roughly $175–185 billion to as high as $195–205 billion — roughly double 2025's $91 billion — funded largely out of existing free cash flow rather than external capital raises.

The CODEW Lens: Google doesn't need to win a new category to win the AI transition. It needs AI to deepen the categories it already dominates — and the early Cloud margin data suggests that's exactly what's happening.

Table 2 · AI Infrastructure Commitments, 2026

Company	2026 Compute/Capex	Funding Source	Chip Strategy
OpenAI	~$50B (2026 spend, testified); ~$1.1–1.4T committed through mid-2030s	Private capital + vendor financing (Oracle, Nvidia, Microsoft, AMD, Broadcom)	Nvidia GPUs, AMD, custom Broadcom chips (H2 2026)
Anthropic	~$19B (2026 training/inference); ~$200B Google Cloud, 5 yrs from 2027	Strategic investment (Amazon, Google, Microsoft, Nvidia) + revenue	Multi-vendor hedge: AWS Trainium, Google TPU/Broadcom, Nvidia
Google / Alphabet	~$195–205B (full-year capex guidance, raised twice in 2026)	Operating cash flow + equity/debt raises	Proprietary TPU (Ironwood, gen 7) + Nvidia GPUs

The Infrastructure Question

Compute has become the defining cost — and the defining strategic variable — for all three, but each is solving for it differently. OpenAI has taken on the largest and most externally financed infrastructure burden, spread across seven major vendors in a structure critics call circular, since several of those vendors are simultaneously OpenAI investors being paid back through the purchase commitments they're financing. Anthropic has spread training and inference across three distinct chip architectures explicitly to avoid single-vendor lock-in, while still committing to roughly $200 billion in Google Cloud spend over five years and reportedly exploring direct data-center leases to bypass cloud providers for some capacity. Google is the only one of the three that owns its core AI silicon outright — a cost and margin structure neither OpenAI nor Anthropic can access, since both remain dependent on Nvidia GPUs at high margins, or on TPU/Trainium capacity rented from Google and Amazon respectively.

The capital intensity across the industry is difficult to overstate. Alphabet raised its 2026 capex guidance twice within the year, ending near $200 billion, funded through equity, debt, and operating cash flow. OpenAI and Anthropic are funding comparable ambitions primarily through private fundraising and vendor financing — a structure leaving both more exposed to a growth slowdown or a dip in investor appetite than Google, whose AI spend is large but not existential relative to a profitable, diversified parent.

The CODEW Lens: Owning the silicon is the closest thing to a durable moat in this market. Google has it. OpenAI and Anthropic are renting or co-financing their way around not having it.

Distribution as a Competitive Variable

Distribution may be the sharpest point of divergence among the three. OpenAI's distribution is built almost entirely on ChatGPT itself — an app people choose to open — supplemented by a large developer API base and a growing direct enterprise motion. Anthropic has deliberately built almost no consumer distribution; its reach runs through developers, Claude Code embedded in engineering workflows, and direct enterprise relationships — a narrower but higher-value funnel reflected in reported average revenue per account many multiples above OpenAI's blended average.

Google's distribution is categorically different because none of it had to be built for AI specifically. Search remains the most-used entry point to the internet; Workspace is embedded in daily business workflow; Android ships on most of the world's smartphones; YouTube commands a dominant share of global video consumption. Gemini rides on top of all of it — a form of distribution neither rival can replicate quickly, though it also means Gemini's contribution is blended into Search, Cloud, and Workspace revenue rather than standing alone as a discrete line item the way ChatGPT or Claude revenue can be discussed.

The CODEW Lens: OpenAI had to build its audience. Google already had it. That single fact may matter more over the next five years than any model benchmark.

The Enterprise AI Market

Enterprise adoption is where the three companies' different bets are being tested most directly. Anthropic has built the strongest position by several measures — roughly 40% share of enterprise LLM usage against OpenAI's 27% and Google's 21%, concentrated in coding and agentic workflows, with 70% of the Fortune 100 among its reported customers. OpenAI's enterprise push is broader but less concentrated, spanning ChatGPT Enterprise, API integrations, and a fast-growing base of over 1 million business customers, weighted more toward volume than Anthropic's premium per-account economics. Google's enterprise position runs primarily through Google Cloud, where Gemini, Vertex AI, and TPU access are sold alongside — and often bundled with — existing GCP infrastructure and Workspace contracts, leveraging procurement relationships that already exist rather than requiring enterprises to adopt an entirely new vendor.

All three are converging on agents — AI systems that take multi-step action rather than simply answering prompts — as the next enterprise battleground. Claude Code is the clearest current proof point, reportedly holding a majority share of the AI coding-agent market on its own. OpenAI and Google are both racing to build comparable agentic capability, though as of mid-2026 neither has published revenue figures isolating agent-specific products the way Anthropic has for Claude Code.

The CODEW Lens: Enterprise share is currently Anthropic's strongest card. Whether it holds as OpenAI and Google scale their own enterprise motions is the single most watchable metric in this market over the next year.

Table 3 · Enterprise LLM Market Share, Mid-2026 (Menlo Ventures)

Company	Share	Primary Strength
Anthropic	~40%	Coding and agentic workflows (Claude Code)
OpenAI	~27%	Broad deployment volume, ChatGPT Enterprise
Google	~21%	Bundled Cloud/Workspace procurement

The Economics of AI Platforms

The three companies' underlying unit economics differ enough to complicate direct comparison. Google Cloud's operating margin expanded from roughly 18% to above 35% through 2026 even as Gemini usage scaled — evidence Google is converting AI-driven Cloud growth into profit at the segment level, aided by lower-cost custom silicon and the fact that AI spend is layered onto an already-profitable core business. OpenAI and Anthropic don't disclose comparable profitability; both are widely understood to run at a loss at the company level once capital-intensive compute commitments are included, even as reported usage and revenue scale rapidly. Goldman Sachs analysis cited in 2025 estimated that once OpenAI's capital commitments were included alongside operating expenses, internal revenue covered under half of total costs, with the remainder financed through vendor arrangements and external capital.

Inference costs — running a trained model in production, as opposed to the upfront cost of training it — have become the more consequential economic variable as usage scales into billions of daily queries. That's part of why all three are investing heavily in custom or lower-cost silicon (Google's TPUs, OpenAI's Broadcom-designed inference chips expected in late 2026, Anthropic's multi-vendor hedge) rather than relying solely on Nvidia GPUs, whose gross margins above 70% are increasingly viewed industry-wide as a cost worth engineering around.

The CODEW Lens: Google is the only one of the three currently proving, in audited numbers, that AI can be run at expanding margin. That's a meaningfully different claim than growing revenue fast.

The Ecosystem Battle

Each company sits at the center of a different ecosystem of dependent businesses. OpenAI's spans a large developer base building on GPT models, a growing set of enterprise integration partners, and — more unusually — its own infrastructure partners, several of which are simultaneously investors, customers, and suppliers in overlapping capacities. Anthropic's ecosystem is narrower and more concentrated around software development: coding tool vendors, DevOps platforms, and enterprise software companies building agentic features on Claude's API. Google's is the broadest by definition, spanning every developer, enterprise, and consumer already inside Google Cloud, Android, and Workspace, alongside emerging AI-specific developer tools on Gemini and Vertex AI.

Cloud providers are increasingly part of each other's competitive ecosystems in unusual ways: Google is simultaneously a model-layer competitor to OpenAI and Anthropic and a critical infrastructure supplier to Anthropic through its TPU and Cloud partnership — a dynamic without a clean historical precedent in enterprise technology.

The CODEW Lens: Google selling compute to a direct model-layer rival is the clearest evidence yet that infrastructure and applications have become separate businesses inside the same AI stack, even when one company touches both.

The Platform Expansion Opportunity

Each company's expansion path reflects its starting position. OpenAI's most plausible routes run through agents, commerce, and consumer software more broadly — leveraging ChatGPT's existing reach to become a default interface for tasks well beyond chat, with reported ambitions in browsers, hardware, and shopping. Anthropic's expansion is narrower and more likely to run through deeper enterprise software integration, developer tooling, and the agentic coding space it already leads, rather than a consumer push it has so far avoided. Google's expansion opportunity is less about entering new categories and more about deepening AI's presence inside categories it already dominates — Search, Cloud, Workspace, Android — while using Cloud and TPU sales to capture value from AI infrastructure demand generated by companies, including rivals, that aren't otherwise Google customers.

The CODEW Lens: OpenAI is trying to create new categories. Anthropic is trying to own one category deeply. Google is trying to make AI invisible inside categories it already owns. Each is a coherent strategy — but they aren't the same strategy.

Key Risks

OpenAI carries the heaviest financing risk of the three: more than $1 trillion in compute commitments financed substantially through vendor arrangements and private capital, unaudited revenue disclosure, and a circular financing structure with several of its largest infrastructure partners that analysts have flagged as a vulnerability if growth slows or external capital tightens. Governance remains a live question following the company's 2025 restructuring into a for-profit public benefit corporation still partially controlled by its nonprofit foundation.

Anthropic faces concentration risk on the opposite end: extraordinary reliance on continued enterprise and coding-market growth, compute spend (~$19B in 2026) running close to full-year revenue, and a valuation trajectory — $61.5B to $965B in roughly 14 months — with few historical comparisons and little room for a growth slowdown without a sharp valuation correction.

Google faces the industry's largest capex exposure in absolute dollar terms — guidance that rose to as much as $205 billion for 2026 alone — a bet that AI-driven Cloud and Search demand keeps converting into the operating-margin gains seen through 2026. Google also carries regulatory risk distinct from its rivals given its dominant Search and advertising position, and faces the risk that AI-driven changes to search behavior (fewer clicks to external sites, more AI-generated direct answers) could cannibalize the advertising revenue that still funds most of its AI investment.

Industry-wide, all three share one risk: the enormous capital committed across the sector — collectively approaching $2 trillion in long-term cloud and compute backlog across the major providers — assumes a pace of enterprise AI adoption and monetization that hasn't yet been proven at that scale over a full economic cycle.

The CODEW Lens: Each company's biggest risk is the mirror image of its biggest bet. OpenAI's leverage funds its reach; Anthropic's concentration funds its premium; Google's capex funds its moat. None of the three risks is hypothetical — all three are already showing up in the numbers.

Table 4 · Risk Profile Comparison

Company	Primary Risk	Mitigation in Place
OpenAI	Financing/circularity risk on >$1T compute commitments	Vendor diversification (Oracle, CoreWeave, in-house Broadcom chips); potential IPO by 2027
Anthropic	Valuation-growth mismatch; compute spend near full-year revenue	Multi-cloud chip hedge; strategic investor backing (Amazon, Google, Microsoft, Nvidia)
Google	Capex scale; regulatory exposure; Search cannibalization	Self-funded from diversified, profitable core business; owned TPU silicon

What to Watch

Enterprise adoption trends will be the clearest near-term signal — whether Anthropic's reported lead in enterprise LLM share holds as OpenAI and Google scale their own enterprise motions. API usage and pricing will show whether the industry is settling into stable unit economics or continuing a race-to-the-bottom on inference pricing. Agent adoption outside coding — where Anthropic currently leads — will indicate whether any of the three can extend agentic capability into broader enterprise workflows at scale.

Compute capacity coming online in 2026 and 2027 — especially OpenAI's Stargate sites and the Anthropic-Google TPU buildout — will test whether committed spend translates into usable capacity on the promised timeline. Financial disclosure will matter more than usual: Anthropic's reported confidential IPO filing, and persistent reporting that OpenAI may also pursue a public listing as early as 2027, would each bring the first audited, independently verifiable financial picture of either company — a meaningful shift from the unaudited, company-reported figures that currently define most public understanding of their businesses.

The CODEW Lens: The single most important date on this list may not be a product launch — it's whichever IPO filing lands first. Audited numbers will settle more of this debate than any new model release.

The CODEW Takeaway

As AI moves from models toward platforms, the defining competition may not simply be who builds the most capable model, but who can build the most durable combination of intelligence, infrastructure, distribution, customers, and ecosystem around it.

OpenAI is betting that consumer-scale distribution and unprecedented compute commitments will let it define the next default interface to computing, even at extraordinary financial risk. Anthropic is betting that a narrower, enterprise-first focus and a deliberately diversified infrastructure hedge can build a more capital-efficient and defensible position, even without OpenAI's consumer reach. Google is betting that owning the full stack — from custom silicon to the world's most-used search engine — lets it absorb the AI transition without having to win any single new category outright, because it already dominates most of the categories AI is reshaping.

None of these bets is settled. What is clear is that "the AI company" is no longer a useful frame for understanding this market.

The CODEW verdict: There are at least three different companies, three different business models, and three different answers to where value will accumulate in the AI platform economy — and the next 18 to 24 months, as compute comes online, IPOs approach, and enterprise budgets get tested, should start to show which bets were right.

The CODEW Lens: Don't ask which company "wins" AI. Ask which layer of the stack you're exposed to — model, infrastructure, or distribution — because that answer, more than any single company's fortunes, is what will actually move.

The AI Platform Economy Glossary

ARR (Annualized Revenue Run Rate) — Most recent monthly revenue × 12; not audited calendar-year revenue. The primary metric OpenAI and Anthropic use to report growth.

TPU (Tensor Processing Unit) — Google's custom AI chip, now in its seventh generation ("Ironwood"), designed to compete with Nvidia GPUs on cost and efficiency for training and inference.

Backlog / RPO — Signed contracted revenue not yet recognized. Google Cloud's backlog reached $514 billion by mid-2026, a key forward-demand signal for AI infrastructure.

Vendor financing / circular financing — A structure where an infrastructure supplier (chipmaker, cloud provider) is simultaneously an investor in the customer buying its products, raising questions about whether reported revenue reflects independent demand.

Multi-cloud / multi-vendor hedge — Anthropic's strategy of splitting training and inference across AWS, Google, and Nvidia hardware to avoid dependency on a single chip architecture or cloud provider.

Agentic AI — AI systems that take multi-step action on a user's behalf, rather than simply responding to a single prompt; the current frontier of enterprise AI competition, led commercially by Claude Code.

Series H / post-money valuation — The valuation of a private company immediately after a funding round closes, including the new capital raised. Anthropic's Series H (May 2026) set its valuation at $965 billion.

FAQ

Q: Which company has the most revenue — OpenAI or Anthropic?

By company-reported figures, Anthropic's annualized revenue (~$47–65B by mid-2026) has overtaken OpenAI's (~$25–40B) — but neither figure is audited. The most reliable third-party number is Microsoft's SEC-disclosed $24.1 billion in FY2026 revenue attributable to OpenAI.

Q: Why is Anthropic valued higher than OpenAI if OpenAI has more users?

Investors appear to be pricing Anthropic's faster revenue growth, higher reported enterprise market share, and premium per-account economics more heavily than OpenAI's larger but lower-monetizing consumer user base. Both valuations are private-market marks, not public listings, and can move quickly.

Q: Does Google make its own AI chips?

Yes. Google designs its own TPUs in-house, now on a seventh generation ("Ironwood"), and began selling TPU systems directly to customers — including Anthropic — in 2026, alongside continuing to offer Nvidia GPUs on Google Cloud.

Q: Is OpenAI's infrastructure spending sustainable?

That's an open and contested question. OpenAI has committed to more than $1 trillion in compute obligations through the mid-2030s, financed largely through vendor arrangements and private capital rather than current revenue. Analysts have flagged the circular financing structure with several major partners as a risk if growth slows.

The CODEW Stat

$1.1T+ OpenAI compute · $965B Anthropic valuation · $205B Google capex Across just three companies, disclosed and committed AI infrastructure spending now approaches $1.5 trillion for 2026 alone once Google's capex is added to OpenAI's and Anthropic's compute commitments — a scale of capital deployment with few precedents in corporate history, financed through three entirely different balance sheets: external debt and vendor financing, strategic partner investment, and existing operating cash flow.

The CODEW