For years, the working assumption in tech policy circles has been blunt: no NVIDIA chips, no frontier AI. The export con

For years, the working assumption in tech policy circles has been blunt: no NVIDIA chips, no frontier AI. The export controls on high-end GPUs like the H100 and the newer Blackwell line were designed precisely around that bottleneck. Cut off access to the best silicon, the theory went, and you slow a country’s ability to train the largest, most capable models. Meituan just called that bluff.

The Chinese tech giant, best known for dominating food delivery and local services, has released LongCat 2.0. It is a 1.6-trillion-parameter model built on a Mixture-of-Experts architecture. The kicker? The team trained it entirely on domestic Chinese chips. No NVIDIA H100s. No Blackwell GPUs. And rather than hiding it behind an API, Meituan has dropped the weights, training code, and full data pipeline on GitHub for anyone to inspect.

What LongCat 2.0 Actually Brings

Let’s look at the hardware-agnostic facts first. At 1.6 trillion parameters, LongCat 2.0 sits among the largest open-weight models ever released. Parameters are not the only measure of quality, but at this scale, they signal serious engineering ambition. Because it uses a Mixture-of-Experts design, only a subset of those parameters activates for any given task. That keeps inference costs from ballooning out of control while still allowing the model to store an enormous range of knowledge and reasoning patterns.

The context window hits one million tokens. That matches the high-end offerings from OpenAI and Anthropic, and it changes what the model can practically do. With a window that large, you can feed it entire legal contracts, months of chat logs, or vast code repositories in a single prompt. For developers building applications that require memory across long documents, this is not a marginal improvement. It is a functional necessity.

Then there is the openness. Meituan did not release a sanitized demo or a restricted API. The model weights are available, the training code is public, and so is the pipeline. That matters for researchers who want to reproduce results, for enterprises that need to audit behavior, and for engineers who need to fine-tune the model on proprietary data without sending anything to a third-party server.

The Hardware Story Everyone Missed

The headline here is not just the parameter count. It is the silicon underneath. LongCat 2.0 was trained on domestic accelerators, specifically chips like the Huawei Ascend series. That is a radically different proposition from slapping an existing framework onto an NVIDIA cluster and hitting run.

Training a trillion-parameter model requires solving distributed computing problems at the extreme edge. Memory bandwidth, inter-chip communication, and floating-point performance all have to be managed with ferocious precision. NVIDIA’s advantage has never been just the raw GPUs. It is the CUDA ecosystem, the optimized kernels, and the collective knowledge of how to squeeze performance out of that stack. Building a comparable model on Ascend hardware means Meituan’s engineers had to do the grueling work of adapting training frameworks, rewriting low-level operations, and debugging distributed training runs on a fundamentally different architecture.

That they succeeded suggests something larger than one model. It shows that Chinese hardware-software co-design is maturing to the point where the absence of Western chips is no longer a hard stop. It is a constraint, and an expensive one, but not an impossible barrier.

Where It Actually Performs

LongCat 2.0 scores strongly in Chinese reasoning, mathematics, and long-context retrieval. Those are specific, high-value benchmarks. Mathematical reasoning tests a model’s ability to handle logic and symbolic manipulation. Long-context retrieval tests whether the model can find a needle of information in a million-token haystack without losing track. Passing both is the difference between a model that sounds smart and one that can actually do work.

Meituan inajipanga kama mbadala wa moja kwa moja kwa Llama, DeepSeek, na Qwen. Kwa kazi za lugha ya Kichina, ushindani huo ni mkali sana. Mifumo iliyofundishwa zaidi kwa kutumia data za Kiingereza kwenye mtandao mara nyingi hukwama kwenye marejeleo ya Kichina cha kale, lugha za kanuni za ndani, istilahi za kifedha zinazotumiwa katika masoko ya bara kuu, na kifupi cha mazungumzo isiyo rasmi kinachotawala mitandao ya kijamii ya Kichina. Mfumo uliojengwa kwa ujuzi wa kina wa lugha ya Kichina, uliolishwa na kampuni ambayo biashara yake kuu inategemea tabia za walaji wa Kichina, una faida ya kimuundo katika mazingira hayo.

Hilo lina umuhimu mkubwa zaidi ya roboti za mazungumzo (chatbots). Makampuni ya teknolojia ya kisheria (Legaltech) yanayochambua sheria za mikataba za PRC, watafiti wanaochambua maandishi ya kihistoria, benki zinazochakata nyaraka za mikopo za Mandarin, na majukwaa ya huduma kwa wateja yanayoshughulikia lahaja za kikanda, yote yanahitaji mifumo inayoelewa nuances badala ya kutafsiri kupitia mtazamo unaozingatia Kiingereza pekee.

Silaha ya Siri ya Meituan: Ukubwa na Data

Meituan si maabara ya utafiti yenye programu ya uwasilishaji (delivery app) kama kazi ya ziada. Ni nguvu kubwa ya kiutendaji inayoratibu mamilioni ya waendesha pikipiki, migahawa, na wafanyabiashara kote China kila siku. Ukubwa huo unazalisha mfululizo wa data za lugha za ulimwengu halisi. Malalamiko ya wateja yanayoelekezwa kwa chatbots. Maelezo ya migahawa yanayochanganya lugha za mitaani za kikanda na maelezo rasmi ya usajili wa biashara. Maelekezo ya njia yanayochanganya mifumo ya anwani na alama zisizo rasmi za mahali. Tiketi za huduma kwa wafanyabiashara zinazohusu kodi, kanuni za usafi, na sheria za ndani.

Data hii ni ya mchanganyiko, yenye muktadha, na ya ndani sana kwa namna ambayo data za kawaida za utafutaji mtandaoni (web crawl) haziwezi kuiga kamwe. Kuichanganya hiyo kwenye LongCat 2.0 kunaupa mfumo msingi wa kivitendo ambao mafunzo ya kitaaluma pekee hukosa. Ni jambo moja kufundisha kwa kutumia maandishi yaliyosafishwa ya Wikipedia. Ni jambo lingine kufundisha kwa kutumia lugha ya vurugu na ya kibiashara ya biashara halisi.

Kwa Nini Hali Inabadilika

Ikiwa makampuni yanaweza kufundisha mifumo ya daraja la mbele (frontier-class models) kwa kutumia silikoni ya ndani, mantiki nzima ya kimkakati nyuma ya marufuku za kuuza chips inaanza kuliwa. Marufuku hizo zilijengwa juu ya dhana kwamba kudhibiti njia za usambazaji za NVIDIA kungeudhibiti uwezo wa AI. Dhana hiyo ilichukulia kuwa hakuna mfumo mbadala unaoweza kufanya kazi uliokuwepo.

LongCat 2.0 si tangazo kwamba chips za Kichina zimefanana na NVIDIA katika kila kipimo. Ni ushahidi kwamba haihitaji kufanana kikamilifu ili kukamilisha kazi. Kumbukumbu ya kutosha, upana wa mawasiliano (bandwidth) wa kutosha, na uboreshaji wa programu wenye akili ya kutosha unaweza kuziba pengo kwa kiasi cha kutosha ili kutoa matokeo yenye ushindani. Hiyo ni kiwango cha chini zaidi kuliko usawa kamili, na ni kiwango ambacho kinaonekana kuwa kimevukwa.

Kwa mnyororo wa usambazaji wa AI wa kimataifa, athari ni wazi. Dhana kwamba mafunzo yote muhimu lazima yapitie kwenye vifaa vya NVIDIA sasa imekufa. Hilo linahamisha nguvu kuelekea uhuru wa kitaifa katika maendeleo ya AI. Nchi na makampuni yanayotazama pembeni hayawaoni tena msambazaji mmoja wa kiungo muhimu (chokepoint supplier) kama njia pekee ya kufikia uwezo wa mbele. Wanaona mgawanyiko, na pengine mandhari ya vifaa ya upande nyingi (multipolar hardware landscape), inayojitokeza haraka kuliko ilivyotabiriwa na wengi.

Hitimisho Halisi

LongCat 2.0 ni zaidi ya toleo la kiufundi. Ni jaribio la nadharia ya kisiasa, na nadharia hiyo imefeli hivi punde. Meituan imedhihirisha kuwa kampuni ya teknolojia ya walaji yenye data sahihi, timu sahihi ya uhandisi, na vifaa sahihi inaweza kufundisha mfumo huru wenye vigezo (parameters) trilioni 1.6 bila kugusa hata GPU moja ya Magharibi iliyopigwa marufuku.

Kwa watengenezaji, hii inamaanisha chaguo jipya la open-weight lenye kina halisi katika lugha ya Kichina na kazi za muktadha mrefu (long-context tasks). Kwa watunga sera, inamaanisha kuwa vikwazo vinavyozingatia kukatia vifaa uwezo lazima vizingatie uwezo wa kubadilika, si tu upatikanaji. Kwa sehemu nyingine ya tasnia, inamaanisha kuwa ramani ya nani anaweza kujenga nini, na kwa vifaa gani, inachorwa upya kwa wakati halisi.

Marufuku za usafirishaji huenda yalikuwa yamewapa muda. Pia yanaonekana kuwa yamenunua mbadala.