The 2026-2028 AI Chip IPO Pipeline Is Now 5 Names Deep - SambaNova, Rebellions, Tenstorrent, Cerebras, and Baidu Kunlunxin

CNBC reported on July 8, 2026 that SambaNova CEO Kunle Olukotun said the company is 'strongly considering' an IPO, likely in the U.S. The news joins the existing 2026-2028 AI chip IPO pipeline: Rebellions (targeting a South Korea IPO in 2026), Tenstorrent (rumored 2027-2028), Cerebras (re-filed S-1 in 2025, targeting 2026), and Baidu Kunlunxin (targeting Hong Kong IPO at $50B). The 2026-2028 AI chip IPO pipeline is now 5 names deep, and the public-market pricing will determine whether the GPU-vs-custom-chip competition is structurally bifurcated.

The reason the timing matters more than a normal AI-product cycle is that the AI chip sector has been a private-market phenomenon for the last five years. Nvidia has been the dominant public-market AI chip name, with AMD as the distant second and Broadcom as the custom-chip enabler. The 2026-2028 AI chip IPO pipeline is the first time the public market will be able to underwrite multiple AI chip alternatives in the same calendar window, and the relative pricing will determine whether the GPU-vs-custom-chip competition is structurally bifurcated or whether the cohort compresses to a single dominant name.

For Nvidia, AMD, Broadcom, TSMC, ASML, Marvell, and the broader AI capex stack, the read-through is direct. The 2026-2028 AI chip IPO window is the cleanest single read on whether the AI chip demand is concentrated (Nvidia wins) or distributed (custom chips and AI chip startups take share). A clean round of IPOs at premium valuations validates the distributed thesis; a messy round of IPOs at discounted valuations validates the concentrated thesis. The trade split is long the AI capex stack (Nvidia, TSMC, ASML) and watch for the IPO window as the structural test.

For most of 2023-2025, the playbook for AI chip public-market exposure was 'long Nvidia, wait for the next round of private-market entrants.' The private-market entrants - SambaNova, Rebellions, Tenstorrent, Cerebras, and Baidu Kunlunxin - have been waiting for the public-market window. The 2026-2028 IPO pipeline is the cleanest single signal that the public-market window is opening, and the trade is no longer 'Nvidia is the only AI chip public-market bet'; it is 'Nvidia is the incumbent, and 5 AI chip startups are the new entrants.'

The first piece of the new framing is 'Nvidia is the incumbent.' Nvidia is the dominant AI chip public-market name with a ~$3T market cap. The 2026-2028 AI chip IPO window is not a threat to Nvidia's incumbent position in the near term - the IPO candidates are smaller and earlier-stage. But the public-market pricing of the IPO candidates is the cleanest single read on whether the AI chip demand is concentrated or distributed, and that read is the binding test for Nvidia's multiple over the next 24-36 months.

The second piece is '5 AI chip startups are the new entrants.' SambaNova is targeting a U.S. IPO with a dataflow architecture that competes with Nvidia on inference workloads. Rebellions is targeting a South Korea IPO with an NPU architecture that competes with Nvidia on edge AI. Tenstorrent is rumored for a 2027-2028 IPO with an RISC-V architecture that competes with Nvidia on cost-per-watt. Cerebras re-filed its S-1 in 2025 with a wafer-scale architecture that competes with Nvidia on training throughput. Baidu Kunlunxin is targeting a Hong Kong IPO at $50B with a China-AI-chip architecture that competes with Nvidia on sovereign AI. The 5-name cohort is the new entrants; the IPO window is the structural test.

The combined target valuations of the 5 AI chip IPO candidates is meaningful. Baidu Kunlunxin is targeting $50B. Cerebras had a previous private mark of ~$4-6B before the 2024 IPO withdrawal; the re-filed S-1 in 2025 is likely targeting a similar or higher valuation. SambaNova is rumored at a ~$5-10B private mark. Tenstorrent is rumored at a ~$2B private mark. Rebellions is rumored at a ~$1-2B private mark. The combined target valuation is roughly $60-70B at the conservative end and $80-100B+ at the high end.

The public-market window is the binding test. A clean round of IPOs at premium valuations would validate the distributed-thesis: the AI chip demand is large enough to support multiple winners. A messy round of IPOs at discounted valuations would validate the concentrated-thesis: the AI chip demand is large enough to support only one winner (Nvidia). The next 24-36 months will tell which way the cohort goes.

The 5-name cohort is also a clean signal of the AI chip demand-side. Each of the 5 candidates is targeting a different workload (inference, edge AI, cost-per-watt, training, sovereign AI), and the 5-name structure is the cleanest single read on the demand fragmentation. If the demand is concentrated on training (Nvidia wins), the inference-focused IPO candidates will struggle. If the demand is distributed across training, inference, edge, and sovereign AI, the 5-name cohort will clear.

The macro question underneath the 5-name AI chip IPO pipeline is whether the AI chip demand is concentrated or distributed. The 2023-2025 trade was that Nvidia was the only meaningful AI chip public-market exposure. The 2026-2028 IPO pipeline is the first time the public market will be able to underwrite multiple AI chip alternatives in the same calendar window, and the relative pricing will determine whether the cohort is concentrated (Nvidia wins) or distributed (custom chips and AI chip startups take share).

For Nvidia, the 5-name IPO pipeline is a multiple-compression catalyst if the cohort clears at premium valuations (distributed demand means Nvidia's incumbent position is being eroded) and a re-rating catalyst if the cohort fails (concentrated demand means Nvidia's incumbent position is structurally durable). The next 24-36 months will tell which way the cohort goes. The cleanest single read is the first IPO: Rebellions is targeting a 2026 South Korea IPO, which is the cleanest single test of the public-market window for the AI chip cohort.

For the broader market, the read-through is that the AI chip IPO window is the structural test for the AI capex stack. A clean round of IPOs at premium valuations validates the distributed thesis and re-rates the entire AI capex stack (Nvidia, TSMC, ASML, Broadcom, Marvell). A messy round of IPOs at discounted valuations validates the concentrated thesis and pressures the AI capex stack. The 2026-2028 calendar is the binding window for the AI chip IPO thesis.

The first tell is the Rebellions 2026 South Korea IPO. Rebellions is the first AI chip IPO candidate in the pipeline, and the timing of the S-1 filing is the first clean signal of the IPO calendar. A 2026 H1 filing means a 2026 H2 listing; a 2026 H2 filing means a 2027 listing. Watch the corporate-governance steps as the first signal of IPO preparation.

The second tell is the Cerebras S-1 re-filing progress. Cerebras re-filed its S-1 in 2025 after the 2024 withdrawal; the timing of the next S-1 update is the cleanest read on the Cerebras IPO calendar. A 2026 H1 update means a 2026 H2 listing; a 2026 H2 update means a 2027 listing.

The third tell is the SambaNova U.S. IPO filing. SambaNova CEO Kunle Olukotun said the company is 'strongly considering' an IPO, likely in the U.S.; the timing of the S-1 filing is the cleanest read on the SambaNova calendar. The fourth tell is the Baidu Kunlunxin Hong Kong filing. Kunlunxin is targeting a $50B Hong Kong IPO; the timing of the HKEX filing is the cleanest read on the Kunlunxin calendar. The fifth tell is the Tenstorrent rumor cycle. Tenstorrent is rumored for a 2027-2028 IPO; any formal announcement is a re-rating catalyst for the AI chip cohort.