While Cerebras builds custom AI chips, French startup Kog squeezes performance out of the GPUs enterprises already own. By rewriting low-level software for existing datacenter hardware, Kog eliminates the latency bottlenecks that slow AI workflows.

Challenging the Necessity of Custom AI Silicon

The AI industry has pivoted to purpose-built chips for inference. Kog's CEO, Gaël Delalleau, says the belief that standard GPUs can’t handle decoding is wrong. Modern GPUs—Nvidia H200, AMD MI300X—offer memory bandwidth that current software leaves idle.

Kog’s Kog Inference Engine (KIE) targets "extremely fast single-request decoding," a must-have for real-time apps. In a recent demo the company hit 3,000 tokens per second (TPS) with Laneformer 2B, an open-source 2-billion-parameter model tuned for their stack. The demo uses a small model, but Kog plans to extend those gains to much larger LLMs.

A "Hacker" Approach to GPU Engineering

Unlike hardware-agnostic providers such as ZML, Kog digs deep. The team reverse-engineers GPUs at the assembly and binary levels, treating the silicon as a set of physical laws to master.

That low-level focus extracts every ounce of efficiency, but it costs time. With a lean crew of 11 engineers, Kog spends weeks or months dissecting each new GPU architecture before adding support. The method delivers top-tier performance but limits how quickly Kog can cover new hardware.

Targeting High-Value AI Use Cases

Kog focuses on sectors where latency equals lost revenue:

  • Software engineering: Tools like Claude Code stall for minutes. Kog aims to turn "waiting hours" into near-instant results.
  • Generative app/game design: Prompt-to-app pipelines need rapid iteration to keep users engaged and revenue flowing.

The company’s next milestone is a 10× speedup on its first major large-scale model. Backed by Scaleway, Bpifrance and the French Tech 2030 program, Kog positions itself as a key player in Europe’s AI sovereignty drive.

Key Takeaways

  • Optimization over hardware: Kog pushes the memory bandwidth of Nvidia H200 and AMD MI300X GPUs to the limit with extreme low-level software engineering.
  • Breaking the latency barrier: The startup targets a 30× boost in LLM inference speed for real-time professional workflows such as automated coding and generative design.
  • Deep-level engineering: By adopting a "hacker" mindset, Kog reverse-engineers GPU assembly and binary code to achieve performance that standard stacks cannot reach.

Kog, a French startup, says its Kog Inference Engine aims to run large-language-model (LLM) decoding up to 30 times faster on the same Nvidia H200 or AMD MI300X GPUs that data-center operators already own.

Why the push for speed matters now

LLM inference now throttles products like code-completion assistants, on-the-fly content generators, and interactive game-design tools. In those settings, the gap between a user’s prompt and the model’s reply directly impacts productivity and revenue. High-level APIs such as CUDA leave a large slice of GPU memory bandwidth idle, especially during token-by-token decoding, which dominates real-time use cases.

Kog’s low-level answer

Kog skips conventional software layers and talks straight to the silicon. Its engineers reverse-engineer the GPU at the assembly level, treating the hardware as a set of constraints to obey rather than a black box to abstract. The result is a custom execution path that keeps data flowing through the GPU’s memory pipes at near-full capacity.

In a recent demo the company ran Laneformer 2B—a 2-billion-parameter open-source model tuned for its stack—and reported a high token throughput. The model is modest compared with larger commercial systems, but the speed gain shows what a purpose-built software stack can extract from the same hardware.

The engineering trade-off

The upside comes with a steep cost curve. Kog’s 11-engineer team spends weeks or months dissecting each new GPU architecture before it can be supported. That depth yields high performance on the targeted cards, but it also means Kog cannot instantly add support for every new GPU that hits the market. Customers benefit only if their fleets include the Nvidia H200 or AMD MI300X GPUs Kog has already optimized.

Who stands to gain

  • Alat kejuruteraan perisian – Produk yang menjana kod, seperti Claude Code, sering terhenti selama beberapa minit sementara model memproses permintaan. Memendekkan masa menunggu tersebut kepada beberapa saat sahaja akan menjadikan pembangunan interaktif jauh lebih lancar.
  • Saluran reka bentuk generatif – Studio yang menukarkan arahan teks kepada prototaip aplikasi atau aset permainan memerlukan iterasi pantas untuk memastikan pencipta terus terlibat. Penyahkodan yang lebih pantas memendekkan kitaran reka bentuk dan meningkatkan kadar penukaran.

Kedua-dua sektor ini menterjemahkan latensi secara langsung kepada kehilangan jam yang boleh dibilkan atau kehilangan pelanggan, jadi peningkatan kelajuan 30× boleh menjadi kelebihan daya saing yang menentukan.

Gambaran lebih luas

Strategi Kog bertentangan dengan trend industri dalam membina silikon AI tersuai. Syarikat seperti Cerebras melaburkan berbilion dalam cip yang menjanjikan throughput yang lebih tinggi bagi setiap watt. Kog berhujah bahawa GPU hari ini sudah mempunyai lebar jalur memori yang mencukupi untuk penyahkodan; bahagian yang hilang ialah perisian yang benar-benar boleh menggunakannya. Jika dakwaan ini terbukti pada skala besar, pembangun boleh menangguhkan naik taraf perkakasan yang mahal dan bergantung kepada naik taraf perisian untuk mencapai inferens hampir masa nyata.

Cabaran berpotensi

  • Beban penyelenggaraan – Setiap generasi GPU baharu akan memerlukan kejuruteraan berbalik yang baharu. Apabila pasaran semakin pelbagai, pasukan kecil tersebut mungkin akan terbeban.
  • Kebolehskalaan kepada model yang lebih besar – Demo tersebut menggunakan model 2 bilion parameter. Menskalakan teknik yang sama kepada sistem yang jauh lebih besar mungkin akan menghadapi had kapasiti memori atau memerlukan teknik kejuruteraan tambahan yang belum didedahkan.
  • Laluan alternatif – Penyedia awan sudah menawarkan instans yang dioptimumkan untuk inferens yang menggabungkan cip khusus dan perisian. Bagi sesetengah beban kerja, peningkatan marginal daripada timbunan Kog mungkin tidak mengatasi kemudahan perkhidmatan terurus.

Perkara untuk diperhatikan

  • Pelancaran model besar pertama – Kog menyatakan bahawa pencapaian penting seterusnya ialah peningkatan kelajuan 10× pada "model skala besar utama." Jurang antara demo 2 bilion parameter dan model gred perusahaan akan menjadi ujian paling jelas bagi pendekatan tersebut.
  • Perluasan sokongan perkakasan – Penambahan sokongan untuk GPU baharu atau pemecut lain akan menandakan sama ada model tahap rendah tersebut dapat mengikut rentak kepantasan perkakasan.

Kesimpulan

Kog menunjukkan bahawa memaksimumkan prestasi daripada GPU sedia ada adalah mungkin apabila anda menulis semula timbunan perisian dari peringkat asas. Janji penyahkodan LLM yang 30× lebih pantas boleh membentuk semula cara pembangun meletakkan harga dan merangka seni bina perkhidmatan AI—dengan syarat syarikat dapat mengekalkan usaha kejuruteraan intensif yang diperlukan untuk setiap komponen silikon baharu.