AMD–OpenAI warrants, 6 GW of GPUs, and MI450 chips: what the viral Reddit post claims
A Reddit post making the rounds argues that AMD has granted OpenAI warrants for roughly 10% of AMD’s equity in exchange for a multi‑year commitment to buy future AMD accelerators. The chips in question – the MI450 series – are said to arrive in 2026, and the purchase commitment is described as around 6 gigawatts (GW) of capacity.
The poster also cites remarks attributed to Nvidia’s CEO, Jensen Huang, suggesting surprise at AMD’s move and noting that OpenAI “doesn’t have the money yet” for its Nvidia orders. The post frames this as part of a broader “circular funding” pattern where chipmakers invest in AI labs that then spend the funds on the chipmakers’ hardware.
Source: Reddit discussion. Figures below reflect the Reddit post; key deal terms have not been independently verified and some details may be not disclosed.
Key details from the AMD–OpenAI deal as described
| Equity instrument | Warrants for ~160 million AMD shares (~10% of AMD), per the post |
| Consideration | OpenAI to buy ~6 GW of AMD accelerators over several years |
| Product | AMD MI450 series, expected 2026 |
| Market context | Nvidia holds ~90% of AI chips, per the post’s characterisation |
| Funding status | OpenAI “doesn’t have the money yet” for Nvidia orders (attributed to Nvidia’s CEO in the post) |
| AMD share reaction | Up ~35% in a week, per the post |
| Warrant price/terms | Not disclosed |
What is a warrant – and why issue it to a customer?
A warrant is a right to buy shares at a set price before a certain date. If AMD’s share price rises above the strike price, exercising the warrant can be valuable. The strike price, vesting, and performance conditions are crucial – none are provided in the post.
Why issue warrants to a customer? It’s a way to secure a strategic anchor buyer, align incentives, and de‑risk a large manufacturing ramp. For AMD, landing OpenAI could accelerate its push against Nvidia’s dominance. The trade‑off is dilution if the warrants are exercised.
“Giving away 10%” before 2026: risk, reward, and timing
The post emphasises that MI450 chips aren’t shipping until 2026. That timing risk is real: performance, yields, and software readiness still need to materialise. If the parts land well, AMD gains a marquee customer and credibility. If they slip or underperform, the warrants could look expensive.
From OpenAI’s side, a second supplier could bring leverage on pricing and availability, and hedge against tight Nvidia supply. But multi‑year purchase commitments are only comforting if funding is secured and the product meets requirements.
Nvidia’s CEO’s reaction – and the “circular funding” concern
“Surprised” AMD gave away 10% before building the product.
“It’s clever I guess.”
OpenAI “doesn’t have the money yet.”
Those are the soundbites attributed to Nvidia’s Jensen Huang in the Reddit post. The poster also references media calling out a pattern: vendors invest in AI labs, who then spend the proceeds on the vendors’ chips. That may be strategic customer financing – or, as critics warn, a sign of bubble dynamics when funding outpaces proven revenue.
What 6 GW signals about the scale of AI compute
6 gigawatts is a measure of power capacity. While the exact mapping from “GW committed” to number of accelerators is not disclosed, it signals enormous data centre build‑outs. For context, UK data centre growth is already straining power availability and planning processes. A single‑digit gigawatt campus is a very big development.
This matters for costs and timelines. Hardware commitments do not equal immediate availability – power, cooling, networking, and software readiness all have to converge. For purchasers, the 2026 date is a reminder to plan around staggered delivery and integration.
Implications for UK developers, teams, and buyers
Competition may lower costs – but not overnight
If AMD successfully lands OpenAI and scales MI450, increased competition could put downward pressure on training and inference pricing. That’s good for UK startups and enterprises frustrated by high GPU rental rates. However, the benefits are likely to phase in around product launch and ecosystem maturity, not before.
Procurement: avoid concentration risk
- Consider multi‑vendor strategies and portability across Nvidia CUDA and AMD ROCm where feasible. Lock‑in can bite if timelines slip.
- Structure contracts with milestones and options rather than all‑in bets on unshipped parts.
- Budget for integration: drivers, libraries, model frameworks, and tooling often lag hardware by months.
Energy, location, and compliance
- Power and planning: UK data centre expansions face local constraints. If you need on‑prem or UK‑hosted compute, engage early with providers.
- Data protection: ensure UK GDPR alignment when using third‑party AI infrastructure, especially for sensitive data and cross‑border processing.
The bear case raised by the Reddit post
- Product risk: MI450 performance and delivery in 2026 are uncertain today.
- Funding risk: the claim that OpenAI “doesn’t have the money yet” implies future raises, debt, or revenue must fill the gap.
- Dilution risk: if warrants are exercised on generous terms, AMD shareholders are diluted; if not, OpenAI’s incentives may weaken.
- Macro risk: if AI spending decelerates or cheaper alternatives emerge, large pre‑commits could be renegotiated or deferred.
The bull case if it all works
- Strategic wedge: AMD secures a flagship customer, catalysing broader adoption and a healthier two‑horse race in AI silicon.
- Ecosystem momentum: major customer commitments attract software investment, improving AMD’s platform and developer tooling.
- Pricing relief: more supply and competition benefit customers globally, including UK SMEs priced out of today’s market.
What to watch next
- Final warrant terms: strike price, vesting, performance conditions (currently not disclosed).
- MI450 disclosures: architecture, benchmarks, availability windows, and ROCm software readiness.
- Funding updates: how OpenAI (and other labs) finance multi‑year hardware commitments.
- Regulatory scrutiny: accounting treatment and disclosures for vendor financing and circular investments.
- Pricing signals: GPU rental rates and cloud accelerator SKUs as 2026 approaches.
Practical takeaways for UK teams today
- Don’t anchor roadmaps to unshipped hardware. Build optionality into your compute plans.
- Prioritise efficiency: smaller models, retrieval‑augmented generation (RAG), and judicious fine‑tuning can cut costs now.
- Focus on near‑term productivity wins while the hardware shake‑out continues. For example, streamline reporting or ops with AI workflows. A practical guide: Connect ChatGPT to Google Sheets.
Bottom line: if the Reddit summary holds, AMD is trading dilution for a shot at relevance in AI accelerators, and OpenAI is trading future equity upside for supplier diversification. It could age like a masterstroke – or a warning case about doing mega‑deals before the product and the money are in hand. The UK angle is straightforward: competition is good, but plan for a 2026 horizon and keep your stack portable.
