Growth That Keeps Getting Faster

There’s a pattern emerging among the hottest AI startups right now. It’s not just that their revenue is climbing. It’s that the climb itself is speeding up. The time between major revenue milestones is shrinking — sometimes dramatically.

A cluster of companies, from young model-makers to older software firms that added AI features, have publicly reported what looks like a classic flywheel effect. Once they cross a certain threshold, the next one comes much faster.

One important caveat: not everyone defines ARR the same way. Some use annualized recurring revenue (contracts signed but not yet billed). Others report an annualized run rate based on the most recent month’s numbers. Gusto, for its part, uses actual trailing 12-month revenue. So direct comparisons are tricky. Still, the direction is unmistakable: up, and accelerating.

Here are the companies that have shared their numbers publicly, listed in reverse order of when they announced their latest milestone.

Mercor: From $1B to $2B in Four Months

Mercor is barely three years old. On Monday, co-founder and CEO Brendan Foody announced the company had crossed $2 billion in gross annualized revenue as of June. That milestone came just four months after hitting $1 billion. In September, the firm — which hires domain experts to train and refine AI models — had already reached a $500 million run rate.

The pace is staggering. Doubling from $1 billion to $2 billion in a third of a year is the kind of growth most startups only dream of. Mercor is proving that the demand for human-in-the-loop AI training is still red-hot.

Anthropic: The Velocity That Stunned the Industry

Few companies have generated as much buzz for revenue speed as Anthropic. In late May, the model maker announced it had crossed $47 billion in revenue run rate. That came less than two months after it reported surpassing $30 billion. To put that in perspective: the company said it reached a $9 billion run rate in late 2025, up from $4 billion in July of the same year.

Anthropic’s growth curve is so steep it has mesmerized the entire AI sector. The company is effectively doubling its run rate in a matter of weeks. That’s not normal — even by Silicon Valley standards.

Sierra: Doubling Down on Enterprise AI Agents

Sierra builds customer service AI agents for large enterprises. Co-founder and CEO Bret Taylor announced in late May that the company reached its first $100 million in ARR in seven quarters. Then it took just two more quarters to add another $100 million.

That acceleration — seven quarters to hit $100M, then two to hit $200M — shows that enterprise buyers are not just dipping their toes into AI customer service. They’re diving in headfirst.

Glean: Cutting the Double Time in Half

Enterprise AI search startup Glean has been around for seven years, but its growth is only getting faster. In May, the company announced it had crossed $300 million in ARR. It took nine months to grow from $100 million to $200 million. It took just six months to go from $200 million to $300 million.

That’s a 33% reduction in the time needed to add $100 million. For a company that’s been in the market for years, not months, that kind of acceleration signals that the product-market fit is deepening, not plateauing.

Gusto: The Old Dog Learning New AI Tricks

Gusto is the veteran of this list. The 14-year-old HR tech startup, last valued at $9.3 billion in early 2022, announced in May that its revenue accelerated in each of the last five quarters. It also surpassed $1 billion in trailing 12-month revenue.

Gusto’s surge proves that you don’t have to be born an AI company to benefit from the boom. By integrating AI into its payroll and benefits platform, the company has reignited its top-line growth. It’s a reminder that incumbents can still sprint if they adopt the right technology.

Clio: Legal Software Gets an AI Boost

Clio has been providing legal practice management software for 18 years. After embedding AI into its offering in 2023, the company’s revenue took off. It surpassed $200 million in ARR in mid-2024, doubled that figure by late last year, and recently announced that ARR had reached $500 million.

That’s a 2.5x increase in roughly 18 months. For an 18-year-old company, that kind of acceleration is rare. Clio shows that even mature markets like legal tech can be disrupted — or at least supercharged — by AI.

What This Tells Us About the AI Market

The common thread across these companies is that AI is not a one-time boost. It’s creating a compounding effect. Early adopters validate the product, which improves the model, which attracts more customers, which generates more data, which makes the product better. That flywheel is spinning faster than ever.

Of course, these are the winners — the ones willing to share their numbers. Many more AI startups are growing quickly but staying quiet. And fast growth doesn’t guarantee profitability or long-term survival. But for now, the revenue acceleration across AI-native and AI-enhanced companies is hard to ignore.

If you’re tracking which startups to watch, these names are setting the pace.

Apple’s latest iOS beta brings a generative fill tool to the Photos app — but is it any good?

When Apple announced the iOS 27 Extend feature at WWDC, I was skeptical. Generative image expansion has been available on Google Pixel phones and through third-party apps for a while, but Apple tends to wait until it can polish a feature before shipping it. After spending several days with the developer beta on an iPhone 17, I can say the tool is genuinely useful — even if it sometimes produces results that look like they wandered out of a dream sequence.

The Photos app overhaul includes three marquee additions: an improved Clean Up tool, Spatial Reframing for spatial video, and the Extend feature. Extend is the headliner — it lets you expand any photo beyond its original boundaries, with Apple Intelligence filling in the missing pixels. I threw everything I could at it: selfies, food shots, portraits taken with a Nikon mirrorless, and casual indoor snaps. Here’s what I learned.

How the Extend tool actually works

Using the feature is straightforward. Open a photo in the Photos app, tap the hamburger menu at the bottom, select Tools, then choose Extend. You can pinch to zoom out, reposition the frame, and drag the edges to any aspect ratio you want. The tool takes about 10 to 15 seconds to generate the expanded image — fast enough that you won’t get impatient.

One smart design choice: Apple baked Extend directly into the Crop tool. If you’re already cropping a photo, just pinch outward beyond the original frame, and the Extend option appears at the bottom of the screen. It feels natural, like the feature was always meant to be there.

Where the AI stumbles — and stumbles hard

Not every result is a winner. In fact, some are downright weird.

I tested Extend on a photo of a mango pastry on a café table. The original frame showed just the pastry and a plastic spoon. After expansion, the AI added a second pastry that never existed, a blurry person in the background, and an extra plate edge on the left. The spoon extension looked convincing; everything else screamed “AI hallucination.”

Another test: I shrank a café interior shot to the center of the frame and let Extend fill all four sides. The tables and chairs at the top turned soft and dreamlike. Gray and black bags lost their definition. Plates on the left shifted in color. It’s the kind of result that looks fine at a glance but falls apart under scrutiny.

I sent one particularly goofy expanded photo — a flower shot extended in all directions — to four friends. Only one thought it was real. The other three spotted the AI artifacts immediately: weird shrubs, a floating flower, leaves that seemed to grow from nowhere.

Some expansion directions are blocked

Extend occasionally refuses to expand in certain directions. I couldn’t figure out a consistent pattern. In some cases, a human subject sat near the edge where I wanted to expand, so that might be a deliberate safety restriction. But other denials had no obvious explanation. The feature also requires an active internet connection — it won’t work offline or over a slow Wi-Fi network, which limits its usefulness on a plane or in a remote area.

But when it works, it’s surprisingly good

For all its quirks, the iOS 27 Extend feature delivers genuinely impressive results often enough to make it worth using.

I tested it on a portrait of a friend taken with a Nikon mirrorless camera at a birthday party in May. The original composition was tight; I expanded the frame to give her more breathing room. The AI filled in shrubs and grass at the bottom left — the texture isn’t perfect, but she couldn’t tell the difference. The tree silhouette and foreground plants came out clean. Only the leaves at the top gave away the trick.

Another strong result: a photo taken at a café with a blurred stranger in the background and reflections in a mirror. Extend preserved the shallow depth of field perfectly, keeping the background person appropriately soft while maintaining the mirror reflections. It’s the kind of subtle edit that would take minutes in Photoshop.

My favorite Extend edit so far

The best result came from a landscape shot of a water bed with mountains in the distance. Extend filled in the water texture to the left, added a canopy stand on the right, and extended the mountain range convincingly. It did add a car that wasn’t there and made my arm look slightly odd, but for a social media post, it’s more than good enough.

An indoor birthday shot also turned out well. The AI added a gift box at the bottom, a lamp on the left, a door on the right, and a partially visible chair behind me — none of which existed in the original. It even bent the ceiling slightly to mimic a wide-angle lens effect, which was either a happy accident or a surprisingly sophisticated touch.

Is the Extend feature worth using?

After a week of testing, I’d say the iOS 27 Extend feature is a solid addition to Apple’s AI toolkit. It’s not flawless — you’ll still get artifacts, especially with complex textures or multiple subjects. But the best results are good enough that you’d need to pixel-hunt to spot the fakery.

Apple marks edited photos with a tag in the metadata (swipe up in the Photos app to see it), which is a nice transparency touch. The feature is best for casual edits: adding a bit of breathing room to a tight portrait, turning a vertical shot into a horizontal one for Instagram, or simulating an ultrawide perspective when you didn’t have the right lens.

Would I trust it for professional work? Not yet. But for everyday use, it’s already more useful than I expected. And for a first-gen beta feature, that’s saying something.

From algorithm to late-stage clinic: a milestone for AI-driven pharma

Insilico Medicine has pushed its lead asset into Phase III human trials. The drug, rentosertib, targets idiopathic pulmonary fibrosis (IPF) — a devastating lung disease with a median survival of just two to four years after diagnosis. This is no small step. It marks one of the few times an AI-discovered molecule has moved beyond early safety checks into large-scale efficacy testing.

The company’s proprietary platform, Pharma.AI, did the heavy lifting. It identified the biological target, designed the molecule, and predicted how it would behave in the body. Now, regulators and patients will see if those predictions hold up at scale.

What is IPF and why rentosertib matters

IPF scars lung tissue relentlessly. Patients lose the ability to breathe properly. Existing drugs slow progression but don’t stop it — and they come with side effects. Rentosertib works differently. It inhibits an enzyme called TNIK (TRAF2- and NCK-interacting kinase), which sits at the crossroads of several pathways driving fibrosis, inflammation, and aging-related tissue damage.

A Phase IIa trial, conducted across 22 sites in China, tested 71 patients. One group got a placebo. Another received 30 mg or 60 mg daily doses of rentosertib for 12 weeks. The results were striking: patients on the 60 mg dose gained an average of 98.4 mL in forced vital capacity (a key measure of lung function), while the placebo group lost 20.3 mL. The U.S. Food and Drug Administration (FDA) had already granted orphan drug designation in February 2023.

How PandaOmics found the target no one else saw

The discovery pipeline started with PandaOmics, one of three engines inside Pharma.AI. This system chews through genomics, clinical data, scientific literature, and patent filings. It builds biological network models and applies causal inference — a statistical method that goes beyond correlation to identify likely drivers of disease.

PandaOmics flagged TNIK as the prime target for IPF intervention. That was unconventional. Most IPF drugs target receptor tyrosine kinases. TNIK sits deeper in the signaling web, regulating pathways like Wnt, TGF-β, Hippo/YAP-TAZ, JNK, and NF-κB. The algorithm also scored TNIK high on a hallmarks-of-aging framework, linking it to chronic inflammation, fibrosis, and cellular senescence.

Feng Ren, PhD, Co-CEO and Chief Scientific Officer of Insilico Medicine, put it plainly: “Rentosertib was not discovered by starting from a conventional target and simply screening more compounds. It came from a biology-first, ageing-informed AI workflow.”

Chemistry42: generative design, not library screening

Once the target was locked, the Chemistry42 engine took over. This is not your typical high-throughput screening, where robots test millions of compounds from a library. Chemistry42 builds molecules from scratch using Generative Tensorial Reinforcement Learning. It designs structures that fit the target protein’s pocket while balancing drug-like properties — solubility, metabolism, toxicity.

The system generated just 79 physical molecules for testing. The team selected the 55th iteration for preclinical development. From project start to preclinical candidate nomination: 18 months. That’s fast by industry standards. The underlying methodology, called GENTRL, was published in Nature Biotechnology back in 2019.

Proteomic clocks and senescence markers in the clinic

Insilico didn’t stop at standard clinical endpoints. The Phase IIa trial included exploratory proteomic analysis using aging-clock frameworks. Tools like ProtAge, OrganAgechrono, and ipfP3GPT tracked predicted biological age changes. The team compared treatment-responsive proteins against UK Biobank trajectories.

Mortality-risk clocks (PAC and OrganAgemortality) provided additional layers of analysis. SenMayo and CellAge signatures measured senescence and senescence-associated secretory phenotype activity. Peer-reviewed work in Aging and Disease confirmed that TNIK inhibition produces senomorphic effects — reducing extracellular matrix remodeling markers.

Alex Zhavoronkov, PhD, Founder and CEO of Insilico Medicine, framed the bigger picture: “This program began with the hypothesis that ageing biology could help identify powerful targets for major diseases. It has now advanced through target discovery, molecular design, preclinical validation, Phase I safety, randomised Phase IIa clinical data, and into Phase III development. For the AI drug discovery field, this is no longer only a speed story — it is a clinical translation story.”

What Phase III will test — and why it matters for AI in biopharma

The upcoming Phase III trial will enroll hundreds of patients across multiple countries. It will measure lung function, progression-free survival, and safety over a longer period. Success would validate not just rentosertib, but the entire AI-driven approach to drug discovery.

Several publications document the journey so far. Nature Biotechnology covered the discovery-to-clinic arc. The Journal of Medicinal Chemistry published the structural biology — including the TNIK kinase domain co-crystal structure. Nature Medicine reported the Phase IIa data.

For the broader industry, this is a test case. Can AI really originate new biology and new chemistry, not just accelerate old methods? The answer will come from patients, regulators, and the data. Phase III is where that answer starts taking shape.

Want to learn more about AI in drug discovery? Check out our coverage of NVIDIA BioNeMo’s role in accelerating research and how generative AI is reshaping clinical trials.