AI Competition Heats Up: Anthropic Leads on the Enterprise Side, Traffic Reshapes the Internet Landscape

Overview

This report marks the relaunch of Deutsche Bank’s ‘Tokenized Takes’ series, using high‑frequency data to track the release cadences of leading AI models, LMArena rankings, consumer engagement metrics (visits/DAU/duration), and enterprise API token consumption, assessing the impact of the evolving AI ecosystem on the seven major vertical industries covered. The central findings are: AI competition is shifting from a winner‑takes‑all scenario to a multipolar one, with distinct dynamics in the consumer and enterprise segments; AI has emerged as a traffic channel, reaching a platform‑level inflection point in early 2026, but it is a double‑edged sword for downstream destination sites—opening new customer acquisition opportunities beyond Google while introducing an AI assistant layer that could lead to disintermediation.

Core views

The consumer AI market is moving away from dominance by a single player toward a landscape of at least three major contenders. ChatGPT’s share of web visits has declined from 88% in December 2024 to roughly 55% in May 2026, with absolute visits dropping from a peak of 6.2 billion in October 2025 to around 5.5 billion. Gemini has risen to become the second-largest platform (approximately 2.9 billion monthly visits), while Claude is growing fastest, up about fivefold year‑to‑date to nearly 1 billion. On the daily active user (DAU) front, however, ChatGPT still leads its closest rival, Gemini, by roughly four times, indicating strong user stickiness. Claude’s DAU has surged some 1,400% year‑over‑year, far outpacing peers, reflecting the halo effect of enterprise adoption. Competition in the enterprise API market is even fiercer and more volatile than on the consumer side. According to OpenRouter data as of the week ending May 31, 2026, DeepSeek regained the top spot with 18.7% of token share, followed by Anthropic at 14.6%, with Google sliding to third place (11.9%) and OpenAI languishing in seventh position (about 8%), behind MiniMax, Xiaomi, and Tencent. Collectively, Chinese model providers account for roughly 45% of enterprise token consumption, underscoring the accelerating migration driven by cost arbitrage. This starkly contrasts with ChatGPT’s 4:1 lead on the consumer side, highlighting that developer API spending is the most dynamic metric in the ecosystem. In terms of model quality, Anthropic has established a clear advantage on the LMArena leaderboard. Its Fable 5 model tops both the general and code rankings, claiming six of the top ten spots in the former and eight in the latter. Over the past year, the rankings have seen three changes of leadership: OpenAI led until July 2025, followed by Google Gemini for about seven months, before Anthropic took over in February 2026. This rapid turnover underscores the intense competition among cutting‑edge models. Meta’s Muse Spark, the first closed‑source frontier model, debuted directly in the general top ten, representing the largest single‑release jump in this cycle. AI‑recommended traffic across industries experienced a synchronized inflection point in February–March 2026, driven primarily by three platform forces: Google’s acceleration of default referral traffic by 12x after setting Gemini 3 Flash as the default; OpenAI’s shift from in‑app checkout to an ad‑plus‑redirect model, routing all commercial queries to external referrals; and consumers’ trust in AI recommendations crossing a critical threshold (Adobe data show AI channel conversion rates reversing from below non‑AI levels to above). Among the verticals, advertising/content platforms (Pinterest, Reddit, etc.) see AI traffic accounting for 40–65 bps; e‑commerce leader Wayfair reports 70–80 bps; music streaming services Spotify and YouTube Music hit record highs (60–80 bps); and real estate portals like Zillow are steadily climbing (30–40 bps), though they face the most pronounced risk of disintermediation, as AI can answer housing questions without requiring users to visit the portal. API pricing strategies have diverged significantly, reflecting each company’s different approach to monetizing inference. OpenAI doubled GPT‑5.5’s list price to $5/$30 per million tokens, raising the effective request cost by 49–92% after adjusting for efficiency gains; Anthropic kept Opus’s nominal price steady but increased its effective cost by 12–27% through a new tokenizer; frontier models from China (DeepSeek, Qwen, MiniMax) remain 3 to 100 times cheaper. All three major players have introduced premium subscription tiers priced around $100, signaling a convergence toward higher‑end pricing. Despite commanding only 12% of token volume, Anthropic captures about 46% of revenue through its premium pricing and surpasses OpenAI in enterprise ramp-up rate (34.4% vs. 32.3%), demonstrating differentiated monetization capabilities.

Analysis framework

The report employs a multi‑dimensional, high‑frequency data tracking framework to evaluate the AI ecosystem and its impact on internet companies. On the supply side, it tracks the release dates and intervals of each vendor’s flagship models, along with changes in their LMArena Elo scores, to quantify competitive intensity and iteration pace. On the demand side, it disaggregates the market into consumer (web visits, DAU, session counts, usage duration) and enterprise (OpenRouter API token routing shares), revealing two distinctly competitive landscapes. To assess downstream effects, the report introduces “AI recommended traffic as a percentage of total traffic,” covering seven verticals—advertising, e‑commerce, travel, music, gig economy, real estate, and automotive—and identifies a platform‑level inflection point in early 2026 by cross‑referencing behavioral shifts across multiple platforms within the same time window (e.g., Google’s default model switch, OpenAI’s commercialization pivot). This analytical approach translates the abstract concept of “AI disruption” into traceable traffic signals, helping investors distinguish which companies are benefiting from AI discovery and which face disintermediation risks.

Methodology notes

• Industry/Industrial Analysis FrameworkPenetration Rate S‑Curve

  AI recommended traffic as a proxy measure of new channel penetration

  The report uses the share of AI‑recommended traffic (currently below 1% but growing rapidly) to gauge the adoption phase of AI as a discovery channel. When this metric enters the steep early stage of an S‑curve, it indicates that the channel is transitioning from experimentation to scale—a key leading indicator of whether internet companies are benefiting from AI traffic.

• Competition and Strategy FrameworkMoat / competitive advantage

  The speed of rotation in enterprise API market share reflects the height of competitive barriers

  The report observes that enterprise API rankings have changed leaders four times within 12 months, whereas the consumer DAU landscape remains relatively stable. This suggests that the enterprise segment lacks switching‑cost moats, leaving developers highly sensitive to price and performance; in contrast, brand loyalty and network effects provide stronger defensive advantages on the consumer side.

• Industry/Industrial Analysis FrameworkSubstitution Effect Analysis

  The risk of disintermediation posed by the insertion of an AI assistant layer between users and websites

  When AI can directly answer user questions in conversation—particularly in information‑intensive scenarios like property searches—traditional portal visits may be entirely replaced rather than merely redirected. By comparing AI traffic shares across industries with their business characteristics, the report identifies real estate portals as facing the highest substitution risk.

• Industry/Industrial Analysis FrameworkPrice‑Volume Decomposition

  Separating nominal price changes from actual request cost movements in API pricing

  The report notes that OpenAI doubled GPT‑5.5’s list price, but because fewer tokens are needed to complete equivalent tasks, the effective cost increase was only 49–92%; Anthropic kept its nominal price steady but raised its effective cost by 12–27% due to a tokenizer change. This decomposition reveals the true economics of inference, preventing misinterpretation based solely on list prices.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Wayfair (W.US)

  Early beneficiary of AI discovery channels; home goods are well suited to AI‑assisted comparison shopping

  Strengths

  AI traffic share of 70–80 bps, the highest among covered e‑commerce companies, commensurate with a large retailer’s scale

  Weaknesses

  The AI assistant layer may drive up traffic acquisition costs in the future

  Comparison

  Outperforms Amazon (45 bps) and Chewy (30 bps), the latter relying more on habitual app usage

  Risks

  Once AI channels scale, they could evolve from incremental sources into a new cost center

• Expedia (EXPE.US)

  AI traffic spiked then subsided, with incomplete recovery

  Strengths

  An early advertiser for ChatGPT and integrated Anthropic’s Claude

  Weaknesses

  AI traffic is 81% dependent on ChatGPT as its sole source, making Expedia the only OTA whose AI share has failed to grow year‑over‑year

  Comparison

  Weaker than Airbnb (organic growth) and Booking.com (diverse AI partners), with an AI share of just 0.37%

  Risks

  Traffic rebounded after OpenAI abandoned in‑app checkout, but the recovery remains incomplete

• Carvana (CVNA.US)

  AI traffic share doubled but lags behind rival CarMax

  Strengths

  Total traffic up 60% year‑over‑year, with a larger absolute funnel

  Weaknesses

  AI share stands at only 0.25%, about half that of CarMax (0.48%)

  Comparison

  Slower in AI share growth than CarMax, but with stronger overall traffic momentum

  Risks

  If inventory data isn’t sufficiently AI‑readable, Carvana may miss opportunities to leverage AI in its research‑oriented automotive niche

• Pinterest (PINS.US)

  A visual discovery platform naturally suited to AI recommendations, with leading AI traffic shares

  Strengths

  AI traffic share of about 65 bps, the highest among advertising/content platforms

  Weaknesses

  Absolute AI visits total only about 10 million, limiting near‑term contribution to profitability

  Comparison

  Outperforms Reddit (40 bps) and Facebook (50 bps), with visual content more readily referenced by AI

  Risks

  AI traffic remains in the early signal phase, and converting it into revenue will take time

• Zillow / Compass / Redfin

  AI traffic is steadily rising but faces the most severe disintermediation risk

  Strengths

  AI traffic share of 30–40 bps, accelerating in sync with industry inflection points

  Weaknesses

  Property search is information‑intensive, and AI can answer questions directly without requiring portal visits

  Comparison

  Opendoor’s traffic is too volatile and its base too small to serve as a meaningful comparator

  Risks

  Direct AI fulfillment of user needs could fundamentally eliminate portal visits, rather than simply redirecting traffic

Key data

• ChatGPT’s share of web visitsApproximately 55% (May 2026)Down from 88% in December 2024, but DAU still about four times that of the runner‑up
• Chinese models’ share on OpenRouterAbout 45%DeepSeek leads with 18.7%, with Tencent, Xiaomi, MiniMax, and others collectively accounting for the remainder, reflecting accelerated cost arbitrage
• Anthropic’s LMArena rankingsSix of the top ten in the general category, eight in the code categoryFable 5 tops both lists and has held the lead since February 2026
• OpenRouter’s weekly token routing volume36 trillion (week ending May 31, 2026)More than 15 times higher year‑over‑year, signaling explosive growth in enterprise API demand
• GPT‑5.5’s effective cost increase49–92%Nominal price doubled, but efficiency gains offset part of the rise; still significantly higher than Chinese models
• Wayfair’s AI traffic shareAround 70–80 bpsHighest among covered e‑commerce companies, with home goods being particularly well suited to AI‑assisted comparison shopping

Impact & implications

For the internet industry, the rise of AI traffic signifies a structural shift in traffic acquisition logic. On one hand, AI recommendations offer diversified sources of traffic for companies reliant on Google Search; decision‑heavy e‑commerce platforms like Wayfair and Etsy, as well as OTAs such as Airbnb and Booking.com, have already begun to benefit. On the other hand, the introduction of an AI assistant layer may force websites that previously enjoyed free direct traffic to adopt paid acquisition strategies, posing a substantial threat of disintermediation, especially for real estate portals (Zillow, Redfin). At the infrastructure level, the rapid churn in the enterprise API market and the ascent of Chinese models suggest that raw performance alone cannot build lasting barriers; cost efficiency and balance sheet flexibility will be decisive factors in determining who can sustain investment. Anthropic has achieved differentiated monetization through premium pricing, while OpenAI has lost share even as it raised prices, highlighting enterprise customers’ acute sensitivity to value. Hyperscale cloud providers (GCP, AWS), as underlying compute suppliers, will continue to benefit from OpenRouter’s year‑over‑year token growth of more than 15x.

Risks

• The insertion of an AI assistant layer between users and websites may lead to the disintermediation of traditional direct traffic, posing a structural threat especially to information‑intensive portals (e.g., real estate)
• As AI traffic channels scale, websites may be forced to bear new traffic acquisition costs, eroding initial incremental benefits
• Competition in the enterprise API market is extremely fierce and rapidly rotating, making it difficult for model providers to establish durable market share barriers
• OpenAI’s pricing strategy could further drive enterprise customers toward lower‑cost Chinese models, undermining its monetization foundation

What to watch

• Expedia’s progress in recovering AI traffic after withdrawing from ChatGPT and the effectiveness of its integration with Claude
• Improvements in Carvana’s inventory data readability to determine whether it can catch up with CarMax in automotive research scenarios
• Whether the gap in AI traffic shares between YouTube Music and Spotify widens due to internal routing differences within Gemini
• The release of OpenAI’s GPT‑5.6 and how its pricing reset affects enterprise market share
• Whether Google launches Gemini 3.5 Pro as scheduled in June and whether it can reverse the decline in its enterprise share
• Whether AI traffic shares among real estate portals continue to climb and whether direct AI responses begin to suppress portal visits