The Bear Hug: Why Google and Microsoft's "Alliance" Against OpenAI and Anthropic Says More Than It Seems

THESIS: What The Information frames as a pact between Google and Microsoft to "beat back" Anthropic and OpenAI is, in fact, something more interesting—and messier—than an alliance: confirmation that the AI war is shifting away from the terrain where the labs win (raw model quality) toward the terrain owned by the platform giants: distribution, enterprise integration, and above all the standards that will decide how autonomous agents work. It's a battle over the plumbing, not over intelligence. And it deserves a careful reading, because the very word "alliance" hides a huge contradiction.

The facts first. In early June 2026, according to CNBC, Microsoft opened its Build conference with a family of in-house models—the MAI line: MAI-Thinking, MAI Code, MAI Image, MAI Voice and others—explicitly designed to reduce its dependence on OpenAI and lower costs for developers. Microsoft says MAI Code scores above 90% on HumanEval-style tests and improves cost efficiency versus GPT-5.5; the company has signaled these models will power future versions of GitHub Copilot, replacing some of the OpenAI model calls running under the hood today. Almost in parallel, according to Crypto Briefing, Google, Microsoft and Salesforce—joined by Cisco, Databricks, GitHub, NVIDIA, ServiceNow, Snowflake and Hugging Face—unveiled in mid-June a new open standard, ARD (Agentic Resource Discovery), under an Apache 2.0 license and coordinated by the Linux Foundation. Who's missing from the founding list? OpenAI and Anthropic.

OUR READ: ARD is the key piece, and almost no one has given it the attention it deserves. The standard lets an AI agent describe what it's trying to accomplish and automatically discover the available tools and resources across the corporate software stack, with no custom-built connections. It is deliberately complementary—not a replacement—to Anthropic's Model Context Protocol (MCP) and Google's own Agent2Agent protocol. Here's the play: if the enterprise future is agentic, whoever controls the discovery layer controls where the value flows. And a world where agents find and use Google Workspace, Microsoft 365 and Salesforce CRM "out of the box" reinforces precisely the subscriptions millions of companies already pay for. It's no accident that the standard is born from the owners of distribution and not the owners of the best models. It's the classic incumbent response: when you can't win on product, you change the board and compete on standards and channel.

But here comes the contradiction that forces us to qualify the headline. While the talk is of "beating back" Anthropic, Alphabet announced in April, according to Bloomberg, an investment of up to $40 billion in Anthropic—$10 billion immediately at a $350 billion valuation, plus tranches tied to milestones—and the dedication of up to 5 gigawatts of computing capacity. In other words: Google is pouring record sums into the very rival it is supposedly trying to contain. This isn't incoherence; it's the nature of the moment. The platform giants are simultaneously buying options on the leading labs and building their own alternatives so as not to be held captive by them. Microsoft did the same: it restructured its deal with OpenAI—which, according to The New Stack, frees OpenAI to use AWS, Google and others and reshuffles the revenue split—while launching its MAI models. The word that best describes this isn't "alliance" but hedging: each incumbent makes sure it doesn't depend on a single lab and that it owns the layer that truly gives it power—distribution.

It's also worth attributing claims rather than asserting them. That Google and Microsoft "are joining forces to beat back" the labs is The Information's editorial framing, not a verified fact of bilateral coordination. The public evidence points to something different: a broad consortium around an open standard, and parallel—not necessarily concerted—moves by two companies that share an objective interest. The nuance matters, because "secret pact against the upstarts" and "convergence of interests among the channel owners" lead to very different conclusions about regulation and competition.

And who's winning? The data paint a paradox we find the most honest of 2026: Google leads on structural metrics—compute, TPUs, balance sheet—but is losing the race for narrative and momentum, where OpenAI and Anthropic dominate. Anthropic, in fact, leads enterprise adoption and has made coding its stronghold, while OpenAI keeps its primacy among developers. That's the crack Microsoft and Google want to exploit: when model quality converges, the enterprise buyer decides on integration, price and trust in the vendor. And there the owners of the office suites start with a channel advantage no lab can match in the short term.

IMPLICATIONS, short term (and here we must be realistic, without euphoria): the most tangible risk isn't that one side or the other wins, but fragmentation and potential lock-in. A standards war—ARD versus MCP, A2A and whatever comes next—could trap companies in incompatible ecosystems, raising the cost of portability and concentrating power even further in three or four platforms. For developers and small businesses, the promise of "agents that connect themselves" could become "agents that only connect well inside your vendor's walled garden." And the concentration of compute—the gigawatts Google is reserving for Anthropic—reminds us that the physical bottleneck still rules, and that energy and chips are already a brutal barrier to entry. These are real problems, today, about competition and dependence.

IMPLICATIONS, long term (and here we are deliberately optimistic, but with our feet on the ground): the good news hidden in this fight is that the competition is being waged on making things cheaper and standardizing them. Microsoft boasts of cost efficiency ten times better; open standards like ARD, if they truly remain open and under neutral governance such as the Linux Foundation, reduce friction and commoditize infrastructure. When the plumbing becomes cheap and universal, value moves upward—toward what gets built on top. That's the direction we care about. If this battle accelerates AI agents being cheap, interoperable and ubiquitous, the benefit ultimately spills over into medical research, diagnostics, materials science and the productivity of millions of workers. The giants' war over the distribution layer is, seen from afar, the noisy and self-interested process by which an expensive, artisanal technology becomes common infrastructure.

OUR CONCLUSION: let's not read this as David versus Goliath, nor as a masterstroke by two titans against two upstarts. It's something more sober and more structural: the platform owners have understood that the model is being commoditized and have moved to defend the one thing that doesn't commoditize easily—channel and standards—while hedging their bet by funding the very labs they claim to fear. For the attentive observer, the signal isn't the rivalry but where it's being fought: whoever controls how agents discover and connect to one another will hold an enormous lever over the next decade. Watch ARD, watch whether the standards truly stay open, and demand interoperability. Because the future of abundance that excites us only arrives if the plumbing is shared; if it becomes the property of three walled gardens, we'll merely have swapped an oligopoly of models for an oligopoly of pipes.