Is DeepGlint profitable in 2026?

No. DeepGlint posted a net loss of roughly CNY -182.59 million on TTM revenue of about CNY 168.58 million as of July 15, 2026, per Financial Times data, continuing a pattern of losses exceeding revenue.

Is DeepGlint laying off staff?

Headcount fell from 367 employees in an April 16, 2026 filing snapshot to 325 employees by July 15, 2026, a roughly 11% contraction within a few months, consistent with ongoing net losses.

Could AI replace DeepGlint's products?

DeepGlint's core face and body recognition engines face commoditization from frontier multimodal models like Google's Gemma 4 (Apr-Jun 2026) and Gemini 3.6 Flash (July 21, 2026), which perform similar visual recognition tasks cheaply.

Who is DeepGlint's biggest competitor?

Hikvision, a much larger surveillance-hardware incumbent, dwarfs DeepGlint in scale. DeepGlint also competes with fellow Chinese 'AI little dragons' SenseTime, CloudWalk, Megvii and Yitu in face-recognition and smart-city AI.

DeepGlint (Beijing Deep Glint Technology Co Ltd, ticker 688207) is publicly traded on Shanghai's STAR Market since its March 17, 2022 IPO. Pre-IPO backers included Sequoia Capital China, ZhenFund, and Ceyuan Ventures.

Is DeepGlint pivoting to foundation models?

Yes. Since late 2024 DeepGlint-AI has released open-source multimodal models (MLCD, UniME families) and its commercial roadmap now lists 'inspiration multimodal large models,' though no revenue contribution from this pivot had been verified as of July 2026.

AI companies and labs · China

DeepGlint

deepglint.com ↗

Chinese computer-vision firm selling face and video recognition to police and banks.

Little protection left

Assessed 26 July 2026 · Evidence as of 26 July 2026

Not much stands between this business and its pressures.

On the recordSince 26 July 2026 · Checkable from August 22-24, 2026 · Every call on the record Checkpoint: Q2 2026 earnings report covering multimodal pivot revenue and headcount

The call

DeepGlint is a shrinking, loss-making vendor whose core recognition products are being absorbed by bigger rivals and general-purpose AI models alike.

DeepGlint makes software that recognizes faces and bodies in video and photos, mostly for government and finance customers in China. It is getting squeezed from two sides: larger companies like Hikvision and SenseTime already own those same customers and have way more resources, and new AI models from Google and Chinese labs can now do much of what DeepGlint's products do without needing special software. The company is burning cash (losing $180 to $200 million a year on $23 to $24 million in sales) and shrinking its team, so the one thing that could change the picture is if its new multimodal AI products start bringing in real money by late August 2026, instead of just being research projects that cost more than they earn.

Why

AI pressure: Frontier multimodal models like Gemini and DeepSeek. General multimodal models now handle face and video recognition tasks that used to require DeepGlint's specialized engines. Could AI do the core job? Yes, likely.

Bigger players: Hikvision. Hikvision sells the same surveillance and recognition systems at far larger scale and already dominates the government and finance client base. Could a giant copy it? Yes, easily.

What protects it. Existing ties to public-security and state-bank buyers who need cleared vendors, though no contract shows this converting into growth.

What would change this call

DeepGlint's August 22-24, 2026 earnings report on revenue, net loss, and headcount. Shows whether the multimodal pivot generates real revenue or remains R&D-only spending.
Whether DeepGlint headcount continues below 325 in subsequent filings. Would confirm continued contraction rather than stabilization or turnaround.
Any public announcement of new multi-year public-security or state-bank contracts won by DeepGlint. Would indicate defensible customer lock-in rather than commoditized, interchangeable offerings.

Key numbers

TTM revenue CNY 168.58 million (2026-07-15)
Net loss CNY -182.59 million (TTM) (2026-07-15)
Share buyback of RMB 50-100 million announced (2026-07-21)

Named risks

Persistent net losses exceeding revenue, CNY -182.59M loss on CNY168.58M TTM revenue as of July 2026. Severity: high
Core face/body recognition products face commoditization from cheap frontier multimodal models. Severity: high
Headcount contraction from 367 to 325 employees within 2026 signals ongoing restructuring. Severity: medium
Historical customer concentration, Agricultural Bank of China was ~33% of 2019 revenue. Severity: medium
High stock volatility with 52-week range CNY9.70-25.10 and large single-day moves. Severity: low

Rivals, as read in July 2026

Rival	Direction	Evidence
Hikvision	losing to	Described as much larger incumbent dwarfing DeepGlint in scale as of 2026 evidence brief
SenseTime	losing to	2021 third-party analysis said DeepGlint is 'no longer in the first camp' vs the 4 AI Little Dragons
CloudWalk	losing to	Named peer in same smart-city/finance AI segment, DeepGlint judged sub-scale as of 2021-06
Megvii	losing to	Fellow 'AI little dragon' occupying same face-recognition customer base
Google (Gemini/Gemma multimodal models)	losing to	Gemma 4 (Apr-Jun 2026) and Gemini 3.6 Flash (Jul 21 2026) commoditize core recognition tasks

Questions people ask

Is DeepGlint profitable in 2026?: No. DeepGlint posted a net loss of roughly CNY -182.59 million on TTM revenue of about CNY 168.58 million as of July 15, 2026, per Financial Times data, continuing a pattern of losses exceeding revenue.
Is DeepGlint laying off staff?: Headcount fell from 367 employees in an April 16, 2026 filing snapshot to 325 employees by July 15, 2026, a roughly 11% contraction within a few months, consistent with ongoing net losses.
Could AI replace DeepGlint's products?: DeepGlint's core face and body recognition engines face commoditization from frontier multimodal models like Google's Gemma 4 (Apr-Jun 2026) and Gemini 3.6 Flash (July 21, 2026), which perform similar visual recognition tasks cheaply.
Who is DeepGlint's biggest competitor?: Hikvision, a much larger surveillance-hardware incumbent, dwarfs DeepGlint in scale. DeepGlint also competes with fellow Chinese 'AI little dragons' SenseTime, CloudWalk, Megvii and Yitu in face-recognition and smart-city AI.
Who owns DeepGlint?: DeepGlint (Beijing Deep Glint Technology Co Ltd, ticker 688207) is publicly traded on Shanghai's STAR Market since its March 17, 2022 IPO. Pre-IPO backers included Sequoia Capital China, ZhenFund, and Ceyuan Ventures.
Is DeepGlint pivoting to foundation models?: Yes. Since late 2024 DeepGlint-AI has released open-source multimodal models (MLCD, UniME families) and its commercial roadmap now lists 'inspiration multimodal large models,' though no revenue contribution from this pivot had been verified as of July 2026.

Timeline

2026-07-21DeepGlint announces RMB 50-100 million share buyback program
2026-07-15FT snapshot: TTM revenue CNY168.58M, net loss CNY-182.59M, headcount 325
2026-07-14Stock shows 52-week range CNY9.70-25.10 amid high volatility
2026-04-16FT snapshot: TTM revenue CNY150.80M, net loss CNY-201.32M, headcount 367
2026-02-25Product listing shows pivot to inspiration multimodal large models, metaverse apps
2025-08DeepGlint ICCV 2025 highlight paper 'RICE' accepted
2025-01DeepGlint-AI releases UniME-Phi3.5-V-4.2B and MLCD-bigG-14-448px models
2024-12DeepGlint-AI releases MLCD-Seg-7B open-source model
2024-11DeepGlint-AI releases MLCD-Embodied-7B open-source model
2022-03-17DeepGlint IPO on Shanghai STAR Market, ticker 688207

Revision history

26 July 2026

First assessed. Little protection left. The text above is fixed as of this date and is not edited afterwards.

In the same band

Inflection AI Mobvoi Xiaoi Robot