AI Wave Winners: Companies the Buildout Is Carrying
Of 263 AI companies assessed July 2026, only 2 rated hard to displace. Who the AI buildout is actually carrying, and why chips and hard data beat wrappers.
The AI buildout is enormous, but the money is not lifting everyone in the boat. Of the 263 AI companies and labs assessed in our July 2026 assessment of 10,525 companies, only 2 earned a Hard to displace standing: Applied Intuition and Dataiku. The rest of the category tells a harsher story. Of the 263 AI companies assessed in July 2026, 170 carry an losing its protection standing, 34 are rated Little protection left, and 88 face high revenue pressure right now.
So who is the wave actually carrying? The pattern in the data is consistent: the companies holding position own something below or around the model layer. Chips, physical fleets, simulation data, regulated medical niches. The companies getting swallowed sell a thin layer on top of general models, a layer the labs keep absorbing into their own products.
The low-pressure names
Start with the strongest positions. Waymo stood out as one of the least exposed of the 263 AI companies assessed in July 2026, reflecting a paid robotaxi service already running in more than 10 US cities with Alphabet's backing. That is a physical and regulatory lead, not a feature a chatbot can replicate.
Applied Intuition, one of the two Hard to displace names, sells simulation and testing software that carmakers and defense agencies depend on to build autonomy, with no visible customer loss in the assessment record. Dataiku, the other Hard to displace name, also stands among the least exposed in the July 2026 assessment.
The chip layer is the clearest winner group. Nvidia carries a defensible for now standing in the July 2026 assessment, still running the chips and software stack behind most AI training and inference. Broadcom, rated on the same date, makes the custom AI chips hyperscalers rely on while selling enterprise software through VMware. AMD and Marvell Technology are both rated defensible for now as well.
Other comparatively sheltered names in the July 2026 data include Shield AI, rated defensible for now, in defense autonomy; PathAI in lab image analysis, a specialized high-trust niche with no reported customer defections; Isomorphic Labs, OpenEvidence, Crusoe Energy, Gatik, and Halter. Palantir, also rated defensible for now, sits on government contracts, though not untouched: the assessment notes Salesforce's Missionforce won a $5.6 billion Army contract in January 2026, breaking into what had been Palantir's stronghold.
What the winners have in common
Three assets show up again and again in the low-pressure records: hardware or chips that the buildout physically requires, exclusive data that a model upgrade cannot erase (self-driving miles, simulation libraries, pathology images), and contracts in regulated or defense settings where switching costs are high. Companies with value above the model layer, wrappers, avatar makers, prompt tools, mostly sit at the other end of the distribution.
The contrast is stark in the assessment record. Humanloop and Phind were rated among the most exposed of the 263 companies in the July 2026 assessment; Phind's developer search product was absorbed by built-in search and code help in ChatGPT, Claude, and Gemini and shut down in January 2026, while Humanloop's team was absorbed by Anthropic. Rewind AI, Covariant, Soul Machines, MultiOn, Contextual AI, Synthesis AI, Robin AI, and LivePerson were all rated Little protection left in the same assessment. Big labs, Anthropic, Google DeepMind, Meta, and Amazon, keep acquihiring founding teams and folding the tooling straight into their own products, leaving no independent company behind.
Even the famous labs are not sheltered. OpenAI carries an losing its protection standing in the July 2026 assessment, and xAI, Mistral AI, and Perplexity AI are rated the same way. Anthropic is rated protection under test. High valuation and high pressure coexist in this category.
The winners' own watch-items
Even the carried companies have dated risks worth checking, and we never declare an outcome; these are things to verify.
- Hyperscaler chips: Google, Amazon, Microsoft, and Meta are building TPU, Trainium, Maia, and MTIA to shift inference off third-party GPUs over time. Nvidia's position in the July 2026 assessment shows the shift has not cost real share yet; watch new in-house chip generation announcements with expanded internal deployment.
- Broadcom second-sourcing: a dataset watch-item asks whether MediaTek publicly confirms shipping a second-source TPU design for Google, checkable by mid-2028.
- Palantir: the assessment flags whether international commercial growth still sits below 8% in the next earnings report after the January 2026 Missionforce loss, checkable by March 2027.
- Wrapper consolidation: if more AI tooling startups get quietly acquihired by OpenAI, Anthropic, Google, or Meta rather than raising new rounds, the thin layer above the models keeps collapsing into the labs beneath it.
- Chinese bundling: Alibaba, Baidu, Tencent, and Ant Group keep bundling cheaper foundation models into enterprise platforms; 4Paradigm's commodity LLM business fell 70% year over year in Q3 2025 in the assessment record, and a watch-item asks whether the next reported quarter shows another decline, checkable by December 2026.
Why we frame it this way
WillItSurvive publishes standings and dated watch-items, not outcome declarations. In a July 2026 backtest, we reviewed 116 companies that later failed, using only facts that were knowable roughly 18 months before each one's terminal event, and compared them against matched companies that survived the same period. Ninety percent of the companies that failed were rated negatively in that backtest, and none were rated positively. Ten companies initially expected to survive, including Chegg, Fisker, Nikola, and Getir, were instead rated negatively, and each went on to fail or be substantially gutted. That supports separating positioned companies from exposed ones on present evidence; it does not support forecasting any individual firm.
Questions this data answers
Which AI companies are rated strongest in the July 2026 assessment? Only Applied Intuition and Dataiku earned Hard to displace among the 263 AI companies assessed in July 2026. The next tier, defensible for now, includes Nvidia, Broadcom, AMD, Marvell, Waymo, Palantir, Shield AI, and PathAI, mostly companies with chips, physical fleets, or regulated data.
Is Nvidia still the safest AI chip position? Nvidia carries a defensible for now standing in the July 2026 assessment: real pressure exists from hyperscalers building their own chips, but it has not yet translated into share loss. The dated watch-item is hyperscaler in-house chips (TPU, Trainium, Maia, MTIA) expanding internal deployment.
Why are most AI startups rated poorly despite the boom? Of the 263 AI companies assessed in July 2026, 170 are losing its protection and 88 face high pressure, because foundation model labs keep shipping the same features natively. Humanloop, Phind, Rewind AI, and Covariant were each absorbed or shut down after the labs replicated or acquihired their layer.
Are the big model labs themselves winners? Not on this data. OpenAI is rated losing its protection, xAI, Mistral, and Perplexity carry the same standing, and Anthropic is rated protection under test in the July 2026 assessment, while chip and infrastructure names are rated more favorably.
What separates a durable AI company from a fragile one? In the July 2026 assessment the dividing line is whether value lives below or around the model layer, in chips, robots, simulation data, self-driving miles, or regulated contracts, versus above it. Value above the model layer keeps getting absorbed by the labs that own the models.
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.