The AI Landscape: February 2026
OpenAI turns to ads, Google embeds Gemini everywhere, and DeepSeek watches the calendar.
The AI industry loves its product launches and benchmark wars, but the real story of any given moment is usually quieter: what are the major labs actually building, and what does it reveal about where they think the value lies? As January ends and earnings season collides with the ongoing buildout, the five companies that matter most are each making very different bets.
OpenAI: Ads, Healthcare, and the Access Question
OpenAI’s January was defined less by model releases than by distribution moves. The company launched ChatGPT Go globally on January 16th, an $8/month tier positioned between free and the $20 Plus subscription. The same day, it announced plans to begin testing advertisements in the U.S. for free and Go users.
The advertising move drew immediate skepticism, not least because CEO Sam Altman had previously called the idea of AI-plus-ads “uniquely unsettling.” OpenAI’s framing leans heavily on access: ads will help fund intelligence for people who cannot or will not pay. The company promises ads will not influence ChatGPT’s responses, will be clearly labeled, and will not appear for users under 18 or in conversations about health, politics, or mental health. Whether those commitments hold under revenue pressure remains to be seen.
The other thrust of OpenAI’s January was healthcare. ChatGPT Health launched January 7th as a dedicated section for medical conversations, with additional privacy protections and the ability to connect medical records. A day later came OpenAI for Healthcare, the enterprise counterpart: HIPAA-compliant API access now rolling out to institutions including Cedars-Sinai, Stanford Medicine, and Memorial Sloan Kettering. The company claims 230 million people ask ChatGPT health questions weekly.
Taken together, the pattern is clear: OpenAI is no longer just building models. It is building a consumer platform with tiered access, advertising infrastructure, and vertical plays in high-stakes domains. The question is whether it can do all of this while maintaining the trust that makes the product valuable in the first place.
Google: Gemini Everywhere, All at Once
Google spent January embedding Gemini into every surface it controls. The push has been relentless.
On January 14th, the company launched Personal Intelligence, a feature that lets Gemini reason across Gmail, Photos, Drive, and other Google services to surface “proactive insights.” It is Google’s direct answer to Apple Intelligence, though with the crucial difference that Google actually has the models to power it. The feature is off by default and limited to AI Pro and Ultra subscribers.
On January 27th, Google made Gemini 3 the default model for AI Overviews in Search globally, with the ability to ask follow-up questions directly from the overview panel. The same week, Chrome gained a persistent Gemini sidebar, the Nano Banana image generator, and “Auto browse,” an agentic feature that lets users command the browser to perform multi-step tasks across the web.
Gmail, meanwhile, now uses Gemini 3 to power AI Overviews for email, synthesizing long threads into summaries and answering natural-language questions about one’s inbox.
The strategy is unmistakable: Google is betting that distribution wins. It has more surfaces than anyone, and it is wiring Gemini into all of them. The risk is that “AI everywhere” becomes “AI nowhere in particular,” a feature layer rather than a destination. But if the integrations work, Google’s advantage in sheer reach could prove decisive.
Anthropic: Constitution, Code, and Quiet Confidence
Anthropic’s January was characteristically low-key, but the moves it did make were substantive.
On January 21st, the company published a new “constitution” for Claude, replacing the 2023 version that drew principles from sources like the UN Declaration of Human Rights. The new document emphasizes teaching Claude why to behave in certain ways rather than simply what to do. It also includes a striking passage acknowledging uncertainty about whether Claude might have “some kind of consciousness or moral status,” and states that Anthropic cares about Claude’s “psychological security” and “well-being.”
On the product side, Anthropic added health data integration to Claude on iOS and Android (January 12th), expanded Claude Code access to all Team plan seats, and continued iterating on Claude Code with quality-of-life updates.
Perhaps more revealing than any announcement was President Daniela Amodei’s comments to CNBC at the start of the month. While competitors announce trillion-dollar infrastructure commitments, Anthropic’s mantra is “do more with less.” Amodei argued that scaling is not the only lever that matters, and that the winner of the next phase may be the lab that keeps improving while spending at a rate the economy can sustain.
With a reported $10 billion fundraise in the works at a $350 billion valuation, Anthropic is not exactly operating on a shoestring. But its relative restraint, and its emphasis on efficiency over brute-force scale, marks a genuine strategic divergence from the rest of the field.
Meta: Spending Big, Stumbling Publicly
Meta’s January was dominated by two stories: the scale of its AI spending, and the public unraveling of its Llama ambitions.
The company’s Q4 earnings, reported this week, beat expectations with $59.9 billion in revenue. But the headline was the capex guidance: Meta now expects to spend $115 billion to $135 billion on AI infrastructure in 2026, roughly double what it spent in 2025. CEO Mark Zuckerberg said the company remains “compute-constrained” and expects to stay that way for much of the year.
The spending is driven in part by Meta’s Superintelligence Labs, the team formed in 2025 to pursue frontier capabilities. CTO Andrew Bosworth told Davos the team delivered its first internal models in January, though details remain scarce.
But the Llama story has grown more complicated. Reports emerged that Llama 4 Behemoth, the largest model in the Llama 4 family, has faced significant delays. More damaging: Yann LeCun, Meta’s departing chief AI scientist, confirmed in early January that Llama 4’s benchmark results had been “fudged” before its April 2025 release, describing CEO Zuckerberg as “really upset” and saying he “lost confidence in everyone who was involved.”
LeCun’s departure, announced alongside plans to start a new AI research venture called Advanced Machine Intelligence Labs, came with pointed criticism. He described Meta’s recent AI hires as “completely LLM-pilled,” embracing a technology LeCun has called “a dead end when it comes to superintelligence.”
Adding to the uncertainty: CNBC reports Meta is considering making its next major AI model proprietary, abandoning the open-weights strategy that has defined Llama’s appeal to developers.
Meta remains a formidable player, its AI-powered ad targeting drives real revenue, and its spending capacity is essentially unlimited. But the gap between ambition and execution has grown visible in ways it was not a year ago.
DeepSeek: Efficiency as Strategy, Not Just Constraint
The most interesting AI story of the past year has arguably been DeepSeek, the Hangzhou-based lab that shocked the industry in January 2025 when its R1 reasoning model matched OpenAI’s o1 at a fraction of the cost. As 2026 begins, the company is signaling that its next move is imminent.