_Hi Dev.to developers & friends! Here you have our article from Medium about the AI Chess Match.
_
Something that usually gets missed in our days is that the true AI race stopped being purely about model quality a while ago. Whoever builds the best model still matters, but what matters just as much right now is who controls the infrastructure underneath those models, who they’ve convinced to depend on their ecosystem and whether they can survive long enough for any of it to compound.
The moves that we already saw in 2025 and that we are still seeing now starting second Q of 2026 are strategic in a way that most AI coverage doesn’t quite capture. So to understand this better, let’s go through each major player , not just what they’re releasing, but what they’re actually doing.
*OpenAI: Betting on ubiquity before competition catches up
*
OpenAI’s core strategy has always been to move fast enough that switching costs become unbearable before anyone else gets good enough to really matter. ChatGPT crossed half a billion weekly active users with GPT-5.4 being the current flagship, and rather than staying cozy with Microsoft forever, OpenAI has been quietly diversifying its infrastructure bets.
In February 2026, Amazon announced a crazy 50 billion USD investment in OpenAI , a staggering number that also made AWS the true exclusive third party cloud distributor for OpenAI Frontier, the platform for enterprise AI agent deployment. That’s a deliberate hedge because while Microsoft still runs Azure and Amazon now runs distribution, OpenAI sits at the center of both.
But the more interesting play is at the product layer, where OpenAI struck deals with Walmart, Target and Etsy to let users shop directly inside ChatGPT. It also signed a deal with Anduril to help detect battlefield drones and is quietly becoming a platform , a place where things happen, not just a model you call via API. That’s a very different business than selling tokens.
The risk? OpenAI is spending at a pace that would make most finance teams physically ill. And the Musk lawsuit, which just went to trial this recently in Oakland, creates real uncertainty about its IPO plans. But the strategy is clear and what they basically want is to get embedded everywhere before the window closes.
Google DeepMind: The most underrated position in the entire game
People keep underestimating Google and it keeps not mattering. Gemini 3.1 is already out. Gemini 3.0 embedded advanced reasoning directly into the core model rather than as a toggle, a design decision that sounds minor until you realize it changes how every developer builds on top of it.
But Google real advantage is not any single model, but the stack itself. Search, Chrome, Android, Workspace, YouTube, Cloud …Google has more surfaces where AI can be embedded than any other company on earth. When Gemini gets better, it gets better in every one of those places simultaneously. No other company has that kind of distribution flywheel.
Google also launched the Agent2Agent protocol, an open standard for letting AI agents communicate across platforms. Over 50 partners joined at launch, from Salesforce to SAP to Accenture. This is Google doing what Google does best, basically define the protocol, make it open and become indispensable infrastructure. And nobody can deny that it has worked before.
Anthropic: The long game that might actually work
Anthropic is not trying to win on speed, and their bet here is that reliability and safety aren’t just ethical positions but they are a business model. Enterprises building production AI systems don’t want models that hallucinate aggressively and break unpredictably. They want something that behaves consistently enough to be auditable.
Claude Opus 4.6 currently leads SWE-bench Verified at 80.8% , and this is the top score for software engineering benchmarks.
Anthropic also created the Model Context Protocol, and it’s now the de facto standard for how AI agents connect to external tools. As of March 2026, just less than two months ago, MCP crossed 97 million installs, in a move that is not just a research project anymore but pure and simple infrastructure.
Anthropic also joined the Agentic AI Foundation under the Linux Foundation in December 2025, alongside OpenAI and others in another notable move for a company often cast as the cautious alternative.
When labs that compete this intensely contribute to the same neutral body, it usually means they’ve agreed the protocol layer should be shared even if the model layer stays competitive.
The risk we see is pace because Anthropic moves deliberately and in a market where the speed of improvement has been genuinely shocking, deliberate is a real vulnerability.
Microsoft: The relationship that quietly started to crack
For years, Microsoft’s AI strategy was simple…Invest in OpenAI, embed Copilot everywhere and let Azure be the infrastructure. That’s still mostly true, but in April 2026 Microsoft released three proprietary models under its MAI brand ( MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 ) the first major output from a superintelligence team formed in November 2025 under Mustafa Suleyman, the DeepMind co-founder who joined Microsoft in 2024.
This is the first clear signal that Microsoft is not content being a distributor forever. It spent 37.5 billion USD on AI capital expenditure in a single quarter and it is building its own foundation models. They still need OpenAI for now, but it is very clearly not building toward needing OpenAI forever.
Developers using Azure should be watching this closely because what seems clear is that the platform is becoming less neutral.
*Meta: Money, open source and a whole lot of catching up
*
Meta is spending between 115 and 135 billion USD on AI infrastructure in 2026 alone. For context, it’s roughly twice what they spent in the entire last year, and the reason behind that huge spending is that despite being by far the world’s most used social media company, Meta has watched OpenAI and Anthropic run ahead on models while Google has pulled away on enterprise.
The open source Llama family remains Meta’s most important move. It made Meta the default starting point for millions of developers who needed to self host, fine-tune or avoid API costs.
More than anything, it bought goodwill in the developer community that money alone couldn’t have purchased.
More recently, Meta launched Muse Spark (internally code named Avocado) under its new Meta Superintelligence Labs, led by Alexandr Wang, who joined after Meta’s 14.3 billion USD investment in Scale AI. It’s basically a closed model ( a departure from the open source posture ), and a sign that Meta knows open weights alone won’t be enough to compete at the frontier.
*The wild card nobody expected: China’s open source strategy
*
This is the part most Western developers haven’t fully processed yet. Chinese open weight models now account for more global AI model downloads than American ones. Alibaba’s Qwen family has over 300 million downloads and more than 100.000 derivative models built on top of it.
DeepSeek V3.2 delivers roughly 90% of GPT-5.4 quality at about 1/50th the cost.
The strategy is deliberate and clearly signs that without access to the most advanced chips due to American export controls, Chinese labs have leaned into open source as a feedback loop: release models, let developers worldwide improve and build on them, absorb those contributions, and iterate. It’s basically the Linux and Android playbook, but applied to frontier AI.
For developers, the practical implication is uncomfortable to ignore and means that for most tasks (code generation, summarization, analysis ), the performance difference between a 0.10/million USD token open weight model and a 5.00/million USD token closed frontier model has largely collapsed. The assumption that open source was always two years behind is now empirically wrong.
Having said all the above, what we clearly see is that the model layer is getting commoditized faster than anyone expected, and that what isn’t commoditized yet is infrastructure, protocols, distribution and trust. That’s where the real bets are being placed right now , and something that explains why you’re seeing moves like OpenAI embedding into e-commerce, Google defining agent communication standards, Anthropic building the protocol that connects AI to tools and Microsoft quietly starting to build models of its own.
For developers, the practical takeaway is that the API you choose today is also a vote for which ecosystem you’ll be inside in two years. These companies know that. They’re engineering for lock in while the window for lock in is still open. A good and simple strategy.
So the model benchmarks matter, but the strategy underneath them matters even more.
