AI Companies Ranked by Durability: 263 Assessed
Of 263 AI companies assessed in July 2026, 170 are losing its protection and only 2 rank hard to displace. See who holds real moats and who's under pressure.
The headline numbers
WillItSurvive's dataset covers 10,525 companies assessed in July 2026, and the AI companies and labs vertical is one of its most crowded and most stressed slices. Of the 263 AI companies and labs assessed in July 2026, 170 were rated losing its protection and only 2 earned the hard to displace standing. Another 34 were rated little protection left, 33 protection under test, and 24 defensible for now.
Across the 263 AI companies assessed in July 2026, 88 companies were rated as facing high competitive pressure. That is a heavier load than most verticals in the assessment, and it maps onto one dividing line: whether a company's value lives above the model layer or below it.
The line that decides standing
The pattern in the July 2026 data is consistent. Companies selling a thin product layer on top of general-purpose models are being absorbed by the labs and hyperscalers that own those models. Companies that own hard infrastructure, physical hardware, regulated data, or hard-won operational leads are holding.
The only two built to last ratings in the entire 263-company AI vertical assessed in July 2026 went to Applied Intuition and Dataiku. Applied Intuition sells simulation and testing software that carmakers and defense agencies depend on, with no visible customer loss and among the lowest competitive pressure in the vertical, alongside Waymo and Enveda Biosciences.
Just below that top tier, the defensible for now group is dominated by companies with physical or infrastructure advantages: Nvidia, Broadcom, AMD, Marvell, Waymo, Shield AI, Palantir, and PathAI. Waymo already runs paid robotaxi service in more than 10 US cities with Alphabet's backing, a physical and regulatory lead that is hard to copy. Broadcom makes the custom AI chips hyperscalers actually rely on. PathAI sells image analysis to labs and drug researchers, a high-trust niche with no reported customer defections as of the July 2026 assessment.
Who has already been absorbed
Some names in the vertical exist mostly as cautionary data points. Phind shut down in January 2026 after ChatGPT, Claude, and Gemini absorbed its developer-search niche natively. Humanloop's team was absorbed by Anthropic. Rewind AI was acquired and shut down by Meta. Covariant became a shell after Amazon hired its founders in 2024. Soul Machines entered receivership, with avatar rivals Synthesia and HeyGen picking up its enterprise customers.
These cases share a mechanism: big labs acquihired the founding teams and folded the tooling into their own products, leaving no independent company behind. The value lived above the model layer, and the model owners took it back.
The frontier labs themselves are not exempt
A striking detail in the July 2026 data: OpenAI, Mistral AI, xAI, Safe Superintelligence Inc., and Thinking Machines Lab were all rated losing its protection, while Anthropic was rated protection under test, meaning no frontier lab earned better than a contested standing in the 263-company vertical assessed in July 2026. High-profile names like Character.AI, Perplexity AI, Midjourney, and Grammarly also sit in the shaky ground column. Prominence and standing are different measurements.
Pressure sources, named
The assessment maps who is pressing on whom:
- Chinese standalone vendors vs Chinese hyperscalers. 4Paradigm, Xiaoi Robot, CloudWalk, and AISpeech face Alibaba, Baidu, Tencent, and Ant Group bundling cheaper foundation models (Qwen, Ernie, Spark, DeepSeek) directly into enterprise platforms. 4Paradigm's commodity LLM business fell 70% year over year in Q3 2025, per the July 2026 assessment. Mobvoi's revenue fell 17.2% overall and about 41% in the second half as voice AI demand shifted to iFlytek, Huawei, and Doubao.
- UBTech vs domestic robotics rivals. AgiBot and Unitree outship UBTech on humanoid and industrial robot volume, and new July 2026 Chinese consumer AI rules hit UBTech's U1 and UWorld product line directly.
- Nvidia and AMD vs hyperscaler chips. Google, Amazon, Microsoft, and Meta are building in-house silicon (TPU, Trainium, Maia, MTIA) to shift inference off third-party GPUs. So far this shift hasn't shown up as real share loss for Nvidia in the July 2026 assessment.
- Palantir vs Salesforce. Salesforce's Missionforce won a $5.6 billion Army contract in January 2026, breaking into what had been Palantir's government stronghold. Palantir still rates defensible for now.
- Edtech vendors vs bundled apps. Riiid, Yuanfudao, and NetEase Youdao face free or bundled tutoring and translation inside Khan Academy, Duolingo, ByteDance, and general LLM apps.
Watch-items (dated, verify yourself)
WillItSurvive publishes no outcome declarations; forward-looking material appears only as dated items readers can check:
- By December 2026: whether 4Paradigm's commodity LLM segment, down 70% year over year in Q3 2025, shows another year-over-year decline in its next reported quarter.
- By March 2027: whether Palantir's international commercial growth rate still sits below 8% in its next earnings report, following the Salesforce Army contract win.
- By June 2028: whether MediaTek publicly confirms shipping a second-source TPU design for Google, cutting into Broadcom's custom AI chip share.
Broader signals worth tracking: more wrapper startups getting quietly acquihired by OpenAI, Anthropic, Google, or Meta instead of raising new rounds; Chinese hyperscalers launching more bundled enterprise AI platforms; and new hyperscaler in-house chip generations with expanded internal deployment.
How much to trust these ratings
Before trusting any of this, it's worth asking how the approach holds up against outcomes we already know. WillItSurvive checked its standings against 116 companies that later failed, using only information that was available 18 months before each one went under, and against a matched set of companies that survived. The two groups separated cleanly: the companies that were later confirmed dead looked troubled well ahead of time, while genuine survivors mostly did not. Ten companies that looked like healthy survivors in that comparison set, including Chegg, Fisker, Nikola, and Getir, turned out to already be failed or hollowed-out businesses that had been mislabeled as peers. The point of this check isn't that the assessment can predict the future; it's that, working only from facts knowable ahead of time, it reliably tells doomed companies apart from real survivors, and it never gave a clean bill of health to a company that was dead within 18 months.
Questions this data answers
How many AI companies are actually in trouble? Of the 263 AI companies and labs assessed in July 2026, 170 (63%) were rated losing its protection and 34 more were rated little protection left. Only 2 companies, Applied Intuition and Dataiku, earned the hard to displace standing.
Which AI startups have already shut down or been absorbed? Phind shut down in January 2026 after ChatGPT and Gemini absorbed its search niche, Humanloop's team was absorbed by Anthropic, Rewind AI was acquired and shut down by Meta, and Covariant became a shell after Amazon hired its founders in 2024. Soul Machines is in receivership, with Synthesia and HeyGen picking up its customers.
Is Nvidia's position under threat? Nvidia rated defensible for now in July 2026; hyperscaler in-house chips (TPU, Trainium, Maia, MTIA) have not yet cost it real share. The watch-item is expanded internal deployment of those chips, which would slowly narrow Nvidia's inference workload share.
Are Chinese AI vendors under more pressure than Western ones? Standalone Chinese chatbot, voice, and vision vendors like Xiaoi Robot, CloudWalk, Mobvoi, and Megvii were rated little protection left or losing its protection in the July 2026 assessment, among the more exposed positions in the vertical. The main squeeze comes from Alibaba, Baidu, Tencent, and iFlytek bundling cheaper models into the same enterprise market.
What kind of AI company holds up best in this data? Companies whose advantage sits below or around the model layer: chips (Nvidia, Broadcom, AMD), physical operations (Waymo's robotaxi service in more than 10 US cities), simulation data (Applied Intuition), or regulated medical data (PathAI). Companies whose product is a layer on top of someone else's model dominate the shaky ground and wide open columns.
Assessments reflect evidence available July 2026. Companies change; treat watch-items as things to verify, not conclusions.
Figures are as of each company's assessment date unless dated otherwise.